275 Commits

Author SHA1 Message Date
5ea60e9b96 Prepare v0.24.0 release documentation 2026-11-21 18:09:47 +00:00
ba5ac651e2 Share immutable backend defaults during profile capture 2026-11-21 17:55:06 +00:00
6f45fd9fd0 Avoid inheritance expansion during input presence capture 2026-11-21 17:53:40 +00:00
79e23ecdba Preserve defaulted prompt input inspection metadata 2026-11-21 17:52:56 +00:00
63e2968038 Keep application extension sensitivity policy with consumers 2026-11-21 17:51:52 +00:00
1408001fce Clarify configuration handoff consistency guidance 2026-11-21 17:51:36 +00:00
df8384426d Record configuration capture scalability opportunities 2026-11-21 17:49:05 +00:00
8758c350e9 Record defaulted input metadata integration gap 2026-11-21 17:27:06 +00:00
4a9def2138 Record configuration migration review findings 2026-11-21 16:48:34 +00:00
40f8a68253 Record final configuration migration verification 2026-11-21 14:50:06 +00:00
08d3d0ea3b Document shared configuration and repair defaults 2026-11-21 14:48:19 +00:00
56795bb49a Update offline examples for shared configuration 2026-11-21 14:43:14 +00:00
891feb8c27 Cover focused configuration inspection behavior 2026-11-21 14:42:02 +00:00
424d35c798 Cover application-owned configuration handoff 2026-11-21 14:39:48 +00:00
bab3422e62 Migrate facade tests to the new configuration contracts 2026-11-21 14:36:07 +00:00
cf03bca753 Migrate engine tests to captured configuration 2026-11-21 14:30:15 +00:00
92e1e225b2 Migrate configuration tests to captured resources 2026-11-21 14:29:06 +00:00
74a77b3c32 Construct engines from resolved application configuration 2026-11-21 14:24:30 +00:00
3a948a1e02 Mount configuration fragments under PromptKit 2026-11-21 14:22:41 +00:00
eb66f82b4f Use focused native configuration inspection 2026-11-21 14:21:39 +00:00
4d5841e087 Capture owned resources from resolved settings 2026-11-21 14:20:46 +00:00
34c0cf3f14 Compose reusable configuration rules at the PromptKit path 2026-11-21 14:18:15 +00:00
76c8789e65 Share configuration settings across the public boundary 2026-11-21 14:15:14 +00:00
38a2dcca52 Separate output contract inputs from resolved values 2026-11-21 14:12:34 +00:00
dad302f1ff Resolve repair budgets from validation modes 2026-11-21 14:10:22 +00:00
645bff20d7 Upgrade ConfigKit to version 1.4.0 2026-11-21 14:07:51 +00:00
248d0dbcf5 Adopt shared configuration model in architecture and implementation plan 2026-11-21 13:51:26 +00:00
515649fee0 Plan configuration handoff and repair defaults 2026-11-21 13:42:10 +00:00
07aaee8ef6 Add shared configuration model to future roadmap 2026-11-21 13:29:06 +00:00
49ea85680e Record configuration handoff integration results 2026-11-21 13:13:49 +00:00
16f8197a2d Update configuration roadmap for ConfigKit v1.4.0 2026-11-21 12:45:17 +00:00
2a726d60ff Set inspection boundary for application-specific fields 2026-11-21 04:18:22 +00:00
d3523000bf Allow application-specific configuration fields in roadmap 2026-11-21 04:13:27 +00:00
bb8ec2c792 Clarify resource version policy and roadmap planning scope 2026-11-21 04:09:32 +00:00
f80ec1c42c Adopt focused ConfigKit inspection in configuration roadmap 2026-11-21 03:39:11 +00:00
649bb7dcf2 Narrow resolved configuration handoff to typed value lookup 2026-11-21 01:50:53 +00:00
55276c9429 Plan application-owned ConfigKit resolution and whole-result handoff 2026-11-21 00:12:38 +00:00
aae523c916 Clarify repair defaults and configuration inspection boundaries 2026-11-20 19:20:32 +00:00
dfe9806228 Record fixed PromptKit configuration root decision 2026-11-20 19:13:46 +00:00
63c96b7d61 Record default repair budget decisions 2026-11-20 19:10:24 +00:00
e1af2d6d9c Retire completed schema planning documents 2026-11-20 16:24:16 +00:00
bd6b2848bb Avoid copying unused slice capacity 2026-11-20 16:22:09 +00:00
eae4d18a9c Clarify request value limits for inline schemas 2026-11-20 16:20:27 +00:00
2c436d14d9 Keep prompt selection free of schema payloads 2026-11-20 16:20:27 +00:00
1bfb6eb301 Omit inactive runtime schemas from prepared snapshots 2026-11-20 16:19:29 +00:00
37582e9c30 Record schema copying efficiency finding 2026-11-20 16:13:53 +00:00
4120d784a2 Record schema integration review findings 2026-11-20 16:09:20 +00:00
3bc16bffe6 Record schema snapshot ownership finding 2026-11-20 15:56:42 +00:00
e9a12d43c2 Document prompt-owned schemas and finalize migration 2026-11-20 15:46:39 +00:00
9bed36282a Cover mixed schema sources and captured execution 2026-11-20 15:41:13 +00:00
950d3ccfb2 Verify schema projection and snapshot ownership 2026-11-20 15:38:40 +00:00
6fbbaa9cfa Bind schema selections throughout execution 2026-11-20 15:36:44 +00:00
2d96dfa696 Unify inline and bound schema preparation 2026-11-20 15:31:30 +00:00
be5a508554 List feature considerations for the next release 2026-11-20 15:28:08 +00:00
e8420d229e Capture prompt schema selections and source references 2026-11-20 15:27:46 +00:00
596e985cff Capture schema roots from configuration sources 2026-11-20 15:25:00 +00:00
fbe3d427a2 Add schema selection values and ownership 2026-11-20 15:20:29 +00:00
e775a3c116 Plan prompt-owned schema support 2026-11-20 14:41:43 +00:00
adfccb5370 Settle schema selection and runtime API decisions 2026-11-20 14:19:34 +00:00
0603a4f0dd Record collection schema roots and native composition 2026-11-20 13:10:42 +00:00
eaf9a76fd3 Plan prompt-owned inline and source-relative schemas 2026-11-20 12:52:26 +00:00
3d4c4c940e Update ConfigKit to version 1.3.0 2026-11-20 12:46:09 +00:00
6668673c8d Retire completed message hash roadmaps 2026-11-20 12:32:50 +00:00
019e077ba9 Document prepared message content hashes 2026-11-20 04:22:33 +00:00
718cd05043 Verify prepared message content hash semantics 2026-11-20 04:18:33 +00:00
dd40171c01 Expose content hashes on prepared runs 2026-11-20 04:16:50 +00:00
e415ad43ce Add versioned message content hashing 2026-11-20 04:15:25 +00:00
ae1efc8049 Plan prepared message-content hashing 2026-11-20 04:10:15 +00:00
0f8e0e49b1 Settle message hash compatibility across releases 2026-11-20 03:56:18 +00:00
ffce326dd7 Describe prepared message-content hashing 2026-11-20 03:41:51 +00:00
804a4d32b2 Prepare v0.23.3 release notes 2026-11-19 14:27:21 +00:00
67d446f611 Preserve context and original causes in HTTP failures 2026-11-19 14:07:43 +00:00
bcdd6fe06c Preserve operation timeout identity during HTTP generation 2026-11-19 13:44:42 +00:00
382ccfb14b Refresh dependencies and prepare v0.23.2 release notes 2026-11-17 16:49:53 +00:00
7406616794 Prepare v0.23.1 release notes 2026-11-16 19:55:18 +00:00
4939fdb283 Preserve construction error causes without extra branches 2026-11-16 19:44:09 +00:00
fcbff80ab2 Prepare v0.23.0 release notes and retire completed roadmaps 2026-11-14 19:46:54 +00:00
d3bc439bb8 Clarify prepared notification timing and context 2026-11-14 19:45:18 +00:00
3ec68d509c Record the execution controls review finding 2026-11-14 18:30:15 +00:00
57b62d391e Strengthen continuation notification ownership coverage 2026-11-14 17:49:44 +00:00
8831ae35f7 Document and demonstrate execution controls 2026-11-14 17:47:55 +00:00
3de3e60c0c Verify coordinated execution controls and observers 2026-11-14 17:44:36 +00:00
a8698d45a8 Verify prepared notification ownership and cleanup 2026-11-14 17:43:40 +00:00
2b9badb3a5 Test continuous execution budgets across collaborators 2026-11-14 17:41:15 +00:00
79699acbb9 Notify callers from the prepared execution snapshot 2026-11-14 17:39:17 +00:00
4cc5a55579 Retain execution controls in prepared handles 2026-11-14 17:38:09 +00:00
d42b131de2 Bound ordinary runs with attributable operation timeouts 2026-11-14 17:37:25 +00:00
2066579e09 Add execution control request contracts 2026-11-14 17:36:03 +00:00
0e4f590f5a Define the execution controls implementation plan 2026-11-14 17:15:46 +00:00
f17e0620a5 Clarify execution control lifecycle boundaries 2026-11-14 17:07:02 +00:00
5ad1b7aa6e Record accepted execution control decisions 2026-11-14 16:59:50 +00:00
3ace6ddb57 Plan execution controls for coordinated turns 2026-11-14 14:53:18 +00:00
65546d794e Prepare v0.22.0 release notes and retire completed roadmaps 2026-11-14 14:33:41 +00:00
892cb74e6e Keep observer composition storage proportional to selected callbacks 2026-11-14 14:29:39 +00:00
64cf84c9b5 Avoid intermediate conversation history copies 2026-11-14 14:28:08 +00:00
c23393b2cc Document conversation capture and continuation contracts 2026-11-14 14:26:39 +00:00
5aae5df5fb Record conversation algorithm efficiency opportunities 2026-11-14 14:24:50 +00:00
aaaa2816a4 Record conversation API documentation review finding 2026-11-14 04:06:09 +00:00
5c135c5b28 Finalize conversation and execution group documentation 2026-11-14 01:35:06 +00:00
31cc0069fc Demonstrate coordinated conversation workflows 2026-11-14 01:31:22 +00:00
68daa61867 Expose explicit execution group summaries 2026-11-14 01:29:42 +00:00
159d369b9b Aggregate execution group terminal observations 2026-11-14 01:27:58 +00:00
23fb55f9df Compose execution observers in order 2026-11-14 01:26:49 +00:00
a07e95fd17 Cover coordinator lifecycle boundaries 2026-11-14 01:25:47 +00:00
c4d7cf8cd4 Expose conversation coordination through the engine 2026-11-14 01:24:13 +00:00
5a6f73e0bb Add sequential conversation coordinator 2026-11-14 01:23:20 +00:00
c149f7274c Verify independent conversation continuations 2026-11-14 01:22:22 +00:00
e5c42de06d Include conversation ancestry in execution observations 2026-11-14 01:20:06 +00:00
04569e44df Capture completed conversation exchanges 2026-11-14 01:18:55 +00:00
26530808e0 Prepare requests with captured conversation history 2026-11-14 01:16:38 +00:00
fcd0485340 Add immutable conversation exchange storage 2026-11-14 01:14:08 +00:00
5cc3fa0168 Plan conversation continuation and execution groups 2026-11-14 01:08:18 +00:00
b423de12c3 Clarify continuation ordering and API planning 2026-11-14 00:54:59 +00:00
72955aaaa3 Rewrite conversation roadmap around the target capabilities 2026-11-13 17:32:22 +00:00
85dae9b443 Record accepted execution group decisions 2026-11-13 17:13:33 +00:00
70db0ad614 Add optional execution groups to conversation roadmap 2026-11-13 17:01:51 +00:00
37a48f2e20 Record remaining conversation capture decisions 2026-11-13 16:41:36 +00:00
6c14379026 Record captured exchange inspection decision 2026-11-13 16:35:51 +00:00
45bbf7bb2c Record per-run accounting and continuation relationships 2026-11-13 16:33:21 +00:00
ee810b8fbe Record continuation turn-budget consumption decision 2026-11-13 00:21:34 +00:00
bc95d1f9fd Keep coordinator termination contract minimal 2026-11-12 19:58:24 +00:00
c948fef8fc Record zero-turn budget and callback cancellation decisions 2026-11-11 21:54:06 +00:00
cb180e8728 Record final response callback and exhaustion semantics 2026-11-11 21:50:24 +00:00
de2a7fcd6f Record sequential coordinator decision and consumer sketches 2026-11-11 21:12:13 +00:00
f20020d9f4 Record continuation execution settings decision 2026-11-11 20:57:59 +00:00
9396183bdb Record logical conversation capture decision 2026-11-11 19:41:23 +00:00
dba6d7d68c Frame conversation roadmap around content-neutral capabilities 2026-11-11 19:24:44 +00:00
d515a7e49b Record conversation continuation design decisions 2026-11-11 18:53:07 +00:00
171fb11f4e Link to v0.21.0 release notes 2026-11-10 15:18:46 +00:00
c9cbe7d3b3 Draft v0.21.0 release notes 2026-11-10 15:17:28 +00:00
7df7e420aa Adopt ConfigKit 1.2.0 2026-11-10 15:14:31 +00:00
8ff9aa3efd Prepare v0.20.0 release documentation 2026-11-06 18:13:52 +00:00
78b78ea00b Avoid repeated prompt metadata lookups 2026-11-06 18:00:32 +00:00
63e9f0b731 Clarify resource inspection documentation 2026-11-06 17:59:40 +00:00
20b0f78af2 Record prompt inspection lookup optimization 2026-11-06 17:56:06 +00:00
101a2840d4 Record resource inspection documentation gaps 2026-11-06 16:40:33 +00:00
f083ad5309 Complete resource inspection validation and coverage 2026-11-06 15:34:21 +00:00
19a0604874 Document resource inspection and source migrations 2026-11-06 15:31:16 +00:00
ea31e71280 Test typed sources through configuration composition 2026-11-06 15:28:45 +00:00
27966c1686 Add typed prompt sources and backend inheritance 2026-11-06 15:27:08 +00:00
ea63a7b0ff Verify resource inspection isolation and selection 2026-11-06 15:26:16 +00:00
b5b9b1f209 Verify JSON and YAML resource inspection rendering 2026-11-06 15:24:55 +00:00
129959b24c Migrate inspection consumers to resource values 2026-11-06 15:23:59 +00:00
dfe9d6566f Expose captured resource inspection through the engine 2026-11-06 15:21:00 +00:00
bf3d2e19bd Capture complete prompt resources for inspection 2026-11-06 15:18:57 +00:00
4cbd4e2da0 Capture resolved backend and profile resources 2026-11-06 15:17:36 +00:00
b0a16ccce8 Simplify profile null fallback behavior 2026-11-06 15:16:35 +00:00
8add7c49e6 Define resource inspection result types 2026-11-06 15:14:59 +00:00
8563379216 Plan resource inspection and typed configuration sources 2026-11-06 15:03:30 +00:00
951739eda0 Require generically renderable inspection results 2026-11-06 13:58:28 +00:00
a0f696181f Plan null simplification after ConfigKit resolution 2026-11-06 12:56:26 +00:00
3951dcf1f3 Select typed inspection with meaningful null distinctions 2026-11-06 12:43:56 +00:00
363920deb2 Plan symmetric typed prompt source helpers 2026-11-06 12:15:00 +00:00
06315062b3 Define inspection around resolved resource configuration 2026-11-06 03:43:00 +00:00
c669f14610 Record backend inspection error decision 2026-11-06 03:29:41 +00:00
52287ab2d1 Keep inspection roadmap contracts proportionate 2026-11-06 03:04:00 +00:00
a43e0f17ee Align inspection roadmap with operator reporting needs 2026-11-06 02:49:13 +00:00
ffaaf552c8 Expand backend roadmap with configuration API recommendations 2026-11-06 02:40:56 +00:00
d7007eb072 Plan backend inspection API 2026-11-06 02:34:38 +00:00
4f3212bfb6 Write release notes for v0.19.0 2026-11-06 01:20:09 +00:00
4faf43cc29 Remove completed audit and implementation roadmaps 2026-11-06 01:19:33 +00:00
c9663ec63c Protect retained metadata from custom JSON encoders 2026-11-06 01:15:55 +00:00
d8c98ac575 Record the responsibility simplification review 2026-11-06 01:01:14 +00:00
5ee602cc4b Record final validation and repair audit links 2026-11-05 20:19:06 +00:00
96c619f7d4 Update documentation for simplified responsibilities 2026-11-05 20:15:51 +00:00
b3b5ec3820 Verify ownership across execution and repair 2026-11-05 20:14:37 +00:00
5dd1418953 Simplify smoke report ownership 2026-11-05 20:12:40 +00:00
f5c900cb0a Share immutable prepared execution metadata 2026-11-05 20:11:26 +00:00
c5f3de13f3 Transfer owned provider settings during resolution 2026-11-05 20:09:17 +00:00
88725e79ab Detach captured backend and profile values 2026-11-05 20:07:55 +00:00
6c015a17fa Detach schemas without provider value limits 2026-11-05 20:06:37 +00:00
9463c9463a Migrate validator tests to prepared plans 2026-11-05 20:04:55 +00:00
1fbd175e53 Remove direct output validation 2026-11-05 20:01:28 +00:00
4b4ce43796 Prepare both smoke output validators 2026-11-05 20:00:33 +00:00
d78474a3bf Move renderer tests to captured templates 2026-11-05 19:59:34 +00:00
acaf2ccd8b Render only captured prompt templates 2026-11-05 19:57:55 +00:00
97e5739bfb Update tests for captured repository lookups 2026-11-05 19:57:13 +00:00
90f796aef7 Require captured repository lookups 2026-11-05 19:55:10 +00:00
0dec3dcec0 Build provider requests directly as payloads 2026-11-05 19:53:41 +00:00
448e64dc40 Use resolved targets for client routing 2026-11-05 19:52:48 +00:00
de46935a0e Plan the responsibility simplification sprint 2026-11-05 19:24:09 +00:00
c69703b2b6 Define the responsibility simplification roadmap 2026-11-05 19:14:48 +00:00
f660b6e0fc Record prepared validation ownership decision 2026-11-05 19:07:52 +00:00
5f56c65427 Record responsibility audit decisions and ownership guidance 2026-11-05 18:47:51 +00:00
cfa3a85308 Record the architectural responsibility audit 2026-11-05 18:10:01 +00:00
1ff6d778af Preserve operation ownership in terminal observations 2026-11-05 16:03:26 +00:00
3f93763c9c Record the completed audit remediation sprint 2026-11-05 16:00:39 +00:00
4aeb58f913 Avoid duplicating nested error causes 2026-11-05 16:00:39 +00:00
ac1e150168 Validate schemas without relying on source filenames 2026-11-05 15:59:30 +00:00
ff57dbf8b3 Preserve authentication cancellation at the response limit 2026-11-05 15:58:32 +00:00
beb18cc9d6 Clarify algorithmic cost of nested error projection 2026-11-05 15:49:32 +00:00
7203e5d6a1 Record observer integration audit finding 2026-11-05 15:40:18 +00:00
356563a6e5 Record audit remediation review findings 2026-11-05 14:59:41 +00:00
6673284312 Complete documentation reconciliation and validation 2026-11-05 14:37:12 +00:00
6f42e1e007 Retire the legacy capacity client wrapper 2026-11-05 14:35:10 +00:00
cee9bbc7b2 Remove redundant captured prompt fields 2026-11-05 14:34:05 +00:00
7db5572b62 Remove unused captured state and orphan helpers 2026-11-05 14:33:04 +00:00
a91bc1608b Test artifact hashes as opaque values 2026-11-05 14:31:14 +00:00
e7a3665443 Consolidate bundled catalog test expectations 2026-11-05 14:30:14 +00:00
59e8010623 Clarify credential inspection and protect request privacy 2026-11-05 14:28:45 +00:00
ada5377f32 Share generation attempt accounting 2026-11-05 14:27:28 +00:00
df208e132b Return generation attempt results as one value 2026-11-05 14:25:07 +00:00
8c0fb15727 Separate prompt and target capture 2026-11-05 14:24:19 +00:00
2e2fba4993 Separate final configuration record validation 2026-11-05 14:22:34 +00:00
68668cee56 Simplify diagnostic report ownership 2026-11-05 14:21:33 +00:00
f745b14b71 Move authentication probes into the outbound adapter 2026-11-05 14:20:04 +00:00
dbba1485a7 Select inventory namespaces before copying 2026-11-05 14:11:41 +00:00
7c435032b6 Bound repair feedback construction 2026-11-05 14:07:34 +00:00
e834e7d806 Add bounded diagnostic prefix encoding 2026-11-05 14:06:09 +00:00
a4901cbb0e Share requirement branch traversal 2026-11-05 14:04:58 +00:00
5dcfc2e953 Merge condition evidence without repeated copies 2026-11-05 14:03:45 +00:00
a02561d40f Centralize trusted JSON value cloning 2026-11-05 14:02:20 +00:00
6112ee8fb4 Merge provider defaults into one owned copy 2026-11-05 14:00:57 +00:00
0aceff8c08 Verify self-contained schema preparation and submission 2026-11-05 13:59:40 +00:00
cddb2b389c Check declared schema locations for external resources 2026-11-05 13:57:21 +00:00
e056cb1607 Share self-contained schema compilation 2026-11-05 13:56:03 +00:00
1e122f0473 Require validation preparation internally 2026-11-05 13:54:06 +00:00
2f0f5fa607 Detach prepared execution ownership from runners 2026-11-05 13:52:36 +00:00
7d500ee62f Clarify artifact reader ownership transfer 2026-11-05 13:51:29 +00:00
7be677c80b Bound authentication response recognition 2026-11-05 13:49:51 +00:00
1c075bd510 Identify built-in generation preparation failures 2026-11-05 13:48:16 +00:00
b8dc8a49a9 Retain successful response read failures 2026-11-05 13:46:56 +00:00
cd373ee46b Preserve owning categories in public errors 2026-11-05 13:45:00 +00:00
e881bbdcf4 Track the operation owning internal errors 2026-11-05 13:42:56 +00:00
106c808442 Use ConfigKit mapping values directly 2026-11-05 13:41:01 +00:00
4ec35f5ca6 Clarify local Go toolchain selection 2026-11-05 13:39:31 +00:00
6a1a32ae49 Plan the audit remediation sprint 2026-11-05 13:36:46 +00:00
e85ea2db35 Consolidate assessment recommendations into remediation roadmap 2026-11-05 13:15:09 +00:00
3ef2a9e18e Complete the remediation planning assessment 2026-11-05 12:48:29 +00:00
a85a6788b4 Record the first six remediation assessments 2026-11-05 12:38:10 +00:00
21fea46fb3 Plan the audit remediation assessment 2026-11-05 12:32:34 +00:00
62bf82a56b Add audit remediation roadmap 2026-11-05 04:37:26 +00:00
bd3dfa992c Record legacy capacity wrapper retirement decision 2026-11-05 04:31:14 +00:00
425f53a62e Record independent Go toolchain and consumer minimum policy 2026-11-05 04:28:41 +00:00
fcc94eb9e9 Record direct ConfigKit value handoff decision 2026-11-05 04:21:13 +00:00
95730a3c11 Record authentication response budget decision 2026-11-05 04:06:23 +00:00
c542b81666 Record self-contained schema decision and future bundling idea 2026-11-05 04:03:38 +00:00
e8e112186e Record ArtifactReader ownership decision 2026-11-05 03:48:04 +00:00
aa08d46ef6 Clarify audit remediation implementation cautions 2026-11-04 18:08:53 +00:00
668d09997a Consolidate audit findings and record open decisions 2026-11-04 18:05:43 +00:00
69203dcb12 Record dead code audit findings 2026-11-04 16:30:56 +00:00
3b760310a6 Record package boundary audit findings 2026-11-04 16:17:03 +00:00
b39460a507 Record behavioral equivalence audit findings 2026-11-04 16:06:17 +00:00
5177db0611 Record API and interface contract audit findings 2026-11-04 15:29:58 +00:00
31f1e6a89d Record test coverage and effectiveness audit findings 2026-11-04 15:11:58 +00:00
f7085d76f6 Record security audit findings 2026-11-04 14:50:24 +00:00
cf0bfd4d76 Record error handling audit findings 2026-11-04 14:12:57 +00:00
6b1323d1fe Record resource management audit findings 2026-11-04 14:02:50 +00:00
fb87fabc2e Record complexity and maintainability audit findings 2026-11-04 13:27:01 +00:00
5adc1c1e64 Record algorithmic efficiency audit findings 2026-11-04 13:18:32 +00:00
0d27252999 Record functional correctness audit findings 2026-11-04 13:12:28 +00:00
d9f6aad305 Adopt ReleaseKit v0.9.0 and refresh recovery guidance 2026-11-04 02:10:48 +00:00
fda0338134 Finalize v0.18.0 notes and retire completed plans 2026-11-04 00:33:44 +00:00
2b3d165ce2 Document the bundled OpenRouter configuration 2026-11-03 15:39:04 +00:00
43f55edb98 Test bundled OpenRouter smoke diagnostics 2026-11-03 15:36:26 +00:00
4950e1dee2 Test bundled profile request construction 2026-11-03 15:34:10 +00:00
d240364776 Test bundled configuration composition 2026-11-03 15:32:25 +00:00
6f57f2f30a Enable bundled OpenRouter configuration by default 2026-11-03 15:30:47 +00:00
2c50eca033 Add embedded OpenRouter configuration 2026-11-03 15:27:26 +00:00
ac44a4f325 Plan bundled OpenRouter configuration 2026-11-03 15:24:06 +00:00
9e05c3ac8b Clarify bundled profile composition and compatibility 2026-11-03 15:11:45 +00:00
2c83925c88 Plan inherited auto-router cost-tier profiles 2026-11-03 14:48:19 +00:00
e0cd63923d Set bundled OpenRouter smoke-test defaults 2026-11-03 14:40:58 +00:00
75204d6134 Set root identities and expand bundled profile plan 2026-11-03 14:26:50 +00:00
5ce6e01f27 Plan bundled OpenRouter backend and profiles 2026-11-03 13:11:54 +00:00
5cec25a40c Prepare v0.18.0 release documentation 2026-11-03 12:50:19 +00:00
855aae6daa Document recorded configuration normalization 2026-11-03 03:32:59 +00:00
43abbcb414 Test normalized configuration inspection and provenance 2026-11-03 03:31:37 +00:00
9f33fed823 Use resolved configuration for construction 2026-11-03 03:29:41 +00:00
7a98feb399 Record prompt field normalization 2026-11-03 03:28:31 +00:00
626134e5e1 Record backend and profile normalization 2026-11-03 03:27:17 +00:00
e7ed315671 Update ConfigKit to version 1.1.0 2026-11-03 03:24:54 +00:00
d7d9a7dd65 Target normalization release notes at v0.18.0 2026-11-03 03:23:28 +00:00
18f6a474e0 Plan recorded field normalization 2026-11-03 03:17:55 +00:00
fb7ebe6a6c Plan field-level configuration normalization 2026-11-03 03:10:04 +00:00
251 changed files with 13911 additions and 4896 deletions

View File

@@ -26,6 +26,8 @@ source-relative prompt content, supplies an inline input, and prepares through
a scoped view. The example also demonstrates collection inventories and a scoped view. The example also demonstrates collection inventories and
namespace-filtered discovery, without contacting a provider or requiring namespace-filtered discovery, without contacting a provider or requiring
credentials. credentials.
The example loads application-owned settings once and selects no bundled
underlay, keeping inventories limited to its own offline fixtures.
The [consumer guide](docs/consumers/pkg-promptkit.md) also shows a local backend, The [consumer guide](docs/consumers/pkg-promptkit.md) also shows a local backend,
ordered file/embedded sources, programmatic mappings, and how to replace an ordered file/embedded sources, programmatic mappings, and how to replace an
engine when source files change. Applications adapting the prior keyed engine when source files change. Applications adapting the prior keyed
@@ -43,6 +45,47 @@ boundary and constraints that framework work must preserve.
## Release Guidance ## Release Guidance
The [v0.24.0 release notes](docs/releases/v0.24.0.md) cover application-owned
configuration loading, prompt-owned schemas, default output repair, and the
prepared message-content hash. Upgrading consumers should review the breaking
configuration and request-API changes in the
[migration guide](docs/consumers/configuration-migration.md#move-settings-under-promptkit).
The [v0.23.3 release notes](docs/releases/v0.23.3.md) cover HTTP operation-timeout
attribution, custom caller cancellation causes, and response-read error identity.
The [v0.23.2 release notes](docs/releases/v0.23.2.md) cover configuration,
validation, and release-tool dependency updates without an API migration or
Go minimum increase.
The [v0.23.1 release notes](docs/releases/v0.23.1.md) cover corrected constructor
error wrapping for downstream classification, with loading diagnostics and
underlying causes preserved.
The [v0.23.0 release notes](docs/releases/v0.23.0.md) cover per-operation
timeouts and opt-in prepared-request notifications for ordinary and coordinated
execution.
The [v0.22.0 release notes](docs/releases/v0.22.0.md) cover captured conversation
exchanges, application-directed turn coordination, and explicit execution groups.
The [v0.21.0 release notes](docs/releases/v0.21.0.md) cover ConfigKit v1.2.0
and main-relative companion imports through the existing configuration-source
API.
The [v0.20.0 release notes](docs/releases/v0.20.0.md) cover resolved resource
inspection, generic JSON/YAML reports, typed prompt sources, and backend
inheritance. Upgrading consumers should review the changed inspection result
types and profile null-fallback behavior.
The [v0.19.0 release notes](docs/releases/v0.19.0.md) cover audit remediation
and compatibility changes to artifact ownership, programmatic sources, and schemas.
The [v0.18.0 release notes](docs/releases/v0.18.0.md) cover the default-on
OpenRouter backend and profiles, recorded configuration normalization, and
ConfigKit v1.1. The notes explain inventory changes and the bundle opt-out;
the Go 1.27 minimum remains unchanged.
The [v0.17.0 release notes](docs/releases/v0.17.0.md) cover the vanity module The [v0.17.0 release notes](docs/releases/v0.17.0.md) cover the vanity module
path, ConfigKit v1, identified resource lists, removal of prompt-version path, ConfigKit v1, identified resource lists, removal of prompt-version
metadata, and the Go 1.27 minimum. This is a breaking pre-v1 migration. metadata, and the Go 1.27 minimum. This is a breaking pre-v1 migration.

View File

@@ -0,0 +1,325 @@
package promptkit_test
import (
"bytes"
"context"
"errors"
"fmt"
"reflect"
"strings"
"sync"
"sync/atomic"
"testing"
"testing/fstest"
"go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit"
)
type integrationComponent struct {
ID string `yaml:"id"`
}
type integrationApplication struct {
Version string `yaml:"version"`
Sibling string `yaml:"sibling"`
Components []integrationComponent `yaml:"components"`
Prompts []string `yaml:"prompts"`
PromptKit promptkit.Settings `yaml:"promptkit"`
}
func integrationDefinition(resolves, validates *atomic.Int64) configkit.Definition[integrationApplication] {
definition := configkit.Definition[integrationApplication]{
Defaults: func() integrationApplication { return integrationApplication{Sibling: "default"} },
Version: &configkit.VersionRule{Path: configkit.Keys("version"), Check: func(version configkit.Version, validation *configkit.Validation) error {
if version.Present && version.Text != "app-v1" {
validation.Error(configkit.Keys("version"), "invalid application version")
}
return nil
}},
SequenceIdentities: []configkit.SequenceIdentity{configkit.SequenceIdentityAt(configkit.Keys("components"), "id")},
Sensitive: func(path configkit.Path) bool { return path.String() == "/sibling" },
Resolve: func(_ *integrationApplication, _ *configkit.Resolution) error { resolves.Add(1); return nil },
Validate: func(_ *integrationApplication, validation *configkit.Validation) {
validates.Add(1)
validation.Warning(configkit.Keys("sibling"), "application warning")
},
}
configured := promptkit.Configure(definition, func(value *integrationApplication) *promptkit.Settings { return &value.PromptKit })
previous := configured.Validate
configured.Validate = func(value *integrationApplication, validation *configkit.Validation) {
previous(value, validation)
if len(value.PromptKit.Backends) != 1 || value.PromptKit.Backends[0].Endpoint == nil || strings.Contains(*value.PromptKit.Backends[0].Endpoint, " ") {
validation.Error(configkit.Keys("promptkit"), "backend normalization missing")
}
if value.PromptKit.Application["section_extra"] != "section" || value.PromptKit.Backends[0].Application["backend_extra"] != "backend" || value.PromptKit.Profiles[0].Application["profile_extra"] != "profile" || value.PromptKit.Prompts[0].Application["prompt_extra"] != "prompt" || value.PromptKit.Collections["group"].Application["collection_extra"] != "collection" {
validation.Error(configkit.Keys("promptkit"), "application fields missing")
}
}
return configured
}
const integrationDocument = `version: app-v1
prompts: [application-only]
components: [{id: second-component}]
promptkit:
section_extra: section
backends:
- id: local
endpoint: ' https://example.test/v1 '
extra_params: {provider: {mode: original}}
backend_extra: backend
profiles:
- id: profile
backend: local
model: original-model
profile_extra: profile
prompts:
- id: example
default_profile: profile
inputs: [{name: document, required: false}]
messages: [{role: user, content: hello}]
output:
format: json
validation_mode: json_schema
schema: {type: object, required: [ok], properties: {ok: {type: boolean}}}
prompt_extra: prompt
collections:
group: {collection_extra: collection}
`
func TestApplicationOwnedLoadAndCapture(t *testing.T) {
var resolves, validates atomic.Int64
definition := integrationDefinition(&resolves, &validates)
result, err := configkit.Load(configkit.Bytes("application", []byte(integrationDocument)), configkit.Options[integrationApplication]{Definition: definition})
if err != nil {
t.Fatal(err)
}
if result.Config.Sibling != "default" || result.Config.Prompts[0] != "application-only" || len(result.Diagnostics()) != 1 || result.Config.PromptKit.Prompts[0].Output.RepairAttempts == nil || *result.Config.PromptKit.Prompts[0].Output.RepairAttempts != 1 {
t.Fatalf("loaded application = %+v, diagnostics = %+v", result.Config, result.Diagnostics())
}
client := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: `{"ok":true}`}}
engine, err := promptkit.NewFromConfig(result, promptkit.RuntimeConfig{}, promptkit.WithLLMClient(client))
if err != nil {
t.Fatal(err)
}
if resolves.Load() != 1 || validates.Load() != 1 {
t.Fatalf("callbacks rerun during handoff: %d %d", resolves.Load(), validates.Load())
}
before, err := engine.InspectPrompt(context.Background(), "example")
if err != nil {
t.Fatal(err)
}
app := &result.Config.PromptKit
*app.Backends[0].Endpoint = "https://changed.test/v1"
app.Backends[0].ExtraParams["provider"].(map[string]any)["mode"] = "changed"
*app.Profiles[0].Model = "changed-model"
*app.Prompts[0].Messages[0].Content = "changed"
app.Prompts[0].Inputs[0].Name = "changed"
app.Prompts[0].Output.Schema.(map[string]any)["required"].([]any)[0] = "changed"
app.Prompts = nil
app.Collections = nil
after, err := engine.InspectPrompt(context.Background(), "example")
if err != nil || !reflect.DeepEqual(before, after) {
t.Fatalf("captured prompt changed: %+v, %+v, %v", before, after, err)
}
run, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "example"})
if err != nil || run.Validation.Status != promptkit.ValidationPassed || len(client.requests) != 1 || client.requests[0].Prompt.Messages[0].Content != "hello" || client.requests[0].Target.Model != "original-model" {
t.Fatalf("run = %+v, requests = %+v, error = %v", run, client.requests, err)
}
if strings.Contains(fmt.Sprintf("%+v", client.requests[0].Target.ExtraParams), "backend_extra") || strings.Contains(fmt.Sprintf("%+v", after), "prompt_extra") {
t.Fatal("application fields leaked into resources or provider parameters")
}
view, err := engine.InspectConfiguration()
if err != nil {
t.Fatal(err)
}
var report bytes.Buffer
if err := view.WriteEffective(&report); err != nil || !strings.Contains(report.String(), "section_extra: section") || !strings.Contains(report.String(), "prompt_extra: prompt") {
t.Fatalf("inspection = %q, %v", report.String(), err)
}
if len(view.Diagnostics()) != 1 {
t.Fatalf("diagnostics = %+v", view.Diagnostics())
}
var wg sync.WaitGroup
for range 4 {
wg.Go(func() {
_, err := configkit.Load(configkit.Bytes("repeat", []byte(integrationDocument)), configkit.Options[integrationApplication]{Definition: definition})
if err != nil {
t.Error(err)
}
})
}
wg.Wait()
if resolves.Load() != 5 || validates.Load() != 5 {
t.Fatalf("reused callbacks = %d %d", resolves.Load(), validates.Load())
}
}
func TestHandoffMountedSourcesAndExecution(t *testing.T) {
baseFS := &countingMountFS{FS: fstest.MapFS{
"base/collection.yaml": {Data: []byte("prompts:\n - id: parent\n default_profile: profile\n messages: [{role: user, content_file: original.txt}]\n output: {format: json, validation_mode: json_schema, schema_file: schema.json}\n")},
"base/original.txt": {Data: []byte("original")}, "base/schema.json": {Data: []byte(`{"type":"object"}`)},
}}
operatorFS := &countingMountFS{FS: fstest.MapFS{
"operator/item.yaml": {Data: []byte("id: child\nextends: {id: parent}\nmessages: [{role: user, content_file: replacement.txt}]\n")},
"operator/replacement.txt": {Data: []byte("replacement")},
}}
base := configkit.NewFilesystem("embedded", baseFS)
operator := configkit.NewFilesystem("operator", operatorFS)
collection, err := promptkit.CollectionInput("team", base.File("base/collection.yaml"))
if err != nil {
t.Fatal(err)
}
collection.Filesystem = &configkit.FilesystemContext{FS: &base, Root: "."}
item := configkit.Input{Source: operator.File("operator/item.yaml"), Destination: configkit.Keys("promptkit", "collections", "team", "prompts"), PlacementMode: configkit.PlacementModeSequenceItem, Filesystem: &configkit.FilesystemContext{FS: &operator, Root: "."}}
profile := promptkit.SettingsInput(configkit.Mapping("profile", map[string]any{"profiles": []any{map[string]any{"id": "profile", "endpoint": "https://example.test/v1", "model": "model"}}}))
type root struct {
PromptKit promptkit.Settings `yaml:"promptkit"`
}
definition := promptkit.Configure(configkit.Definition[root]{}, func(value *root) *promptkit.Settings { return &value.PromptKit })
layers := []configkit.Underlay{{Inputs: []configkit.Input{profile, collection}}, {Inputs: []configkit.Input{item}}}
result, err := configkit.LoadLayered(configkit.Input{Source: configkit.NoSource()}, configkit.LayeredOptions[root]{Definition: definition, Underlays: layers})
if err != nil {
t.Fatal(err)
}
beforeBase, beforeOperator := baseFS.opens, operatorFS.opens
client := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: `{}`}}
engine, err := promptkit.NewFromConfig(result, promptkit.RuntimeConfig{}, promptkit.WithLLMClient(client))
if err != nil {
t.Fatal(err)
}
if baseFS.opens-beforeBase != 1 || operatorFS.opens-beforeOperator != 1 {
t.Fatalf("unexpected construction reads: base=%d operator=%d", baseFS.opens-beforeBase, operatorFS.opens-beforeOperator)
}
standalone, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Underlays: layers}}, promptkit.WithBundledOpenRouter(false), promptkit.WithLLMClient(client))
if err != nil {
t.Fatal(err)
}
wantPrompt, err := standalone.InspectPrompt(context.Background(), "team/child")
if err != nil {
t.Fatal(err)
}
gotPrompt, err := engine.InspectPrompt(context.Background(), "team/child")
if err != nil || !reflect.DeepEqual(gotPrompt, wantPrompt) {
t.Fatalf("resource reports = %+v, %+v, %v", gotPrompt, wantPrompt, err)
}
wantInventory, err := standalone.ListPrompts(context.Background())
if err != nil {
t.Fatal(err)
}
gotInventory, err := engine.ListPrompts(context.Background())
if err != nil || !reflect.DeepEqual(gotInventory, wantInventory) {
t.Fatalf("inventories = %+v, %+v, %v", gotInventory, wantInventory, err)
}
collectionSettings := result.Config.PromptKit.Collections["team"]
*collectionSettings.Prompts[1].Output.SchemaFile = "changed.json"
prepared, err := engine.Prepare(context.Background(), promptkit.RunRequest{PromptID: "team/child"})
if err != nil || prepared.Messages[0].Content != "replacement" {
t.Fatalf("prepared = %+v, %v", prepared, err)
}
standalonePrepared, err := standalone.Prepare(context.Background(), promptkit.RunRequest{PromptID: "team/child"})
if err != nil || !reflect.DeepEqual(prepared.StructuredOutput, standalonePrepared.StructuredOutput) || !reflect.DeepEqual(prepared.OutputContract, standalonePrepared.OutputContract) || !reflect.DeepEqual(prepared.Messages, standalonePrepared.Messages) {
t.Fatalf("prepared details differ: %+v, %+v, %v", prepared, standalonePrepared, err)
}
run, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "team/child"})
if err != nil || run.Validation.Status != promptkit.ValidationPassed {
t.Fatalf("run = %+v, %v", run, err)
}
}
func TestHandoffUsesOnlyBundlesSelectedBeforeLoad(t *testing.T) {
type root struct {
PromptKit promptkit.Settings `yaml:"promptkit"`
}
definition := promptkit.Configure(configkit.Definition[root]{}, func(value *root) *promptkit.Settings { return &value.PromptKit })
result, err := configkit.LoadLayered(configkit.Input{Source: configkit.NoSource()}, configkit.LayeredOptions[root]{Definition: definition, Underlays: []configkit.Underlay{promptkit.BundledOpenRouterUnderlay()}})
if err != nil {
t.Fatal(err)
}
engine, err := promptkit.NewFromConfig(result, promptkit.RuntimeConfig{}, promptkit.WithBundledOpenRouter(false))
if err != nil {
t.Fatal(err)
}
backends, err := engine.ListBackends(context.Background())
if err != nil || len(backends) != 1 || backends[0] != "openrouter" {
t.Fatalf("bundles = %+v, %v", backends, err)
}
}
func TestApplicationFieldsComposeAndInherit(t *testing.T) {
type root struct {
PromptKit promptkit.Settings `yaml:"promptkit"`
}
definition := promptkit.Configure(configkit.Definition[root]{}, func(value *root) *promptkit.Settings { return &value.PromptKit })
underlay := promptkit.SettingsInput(configkit.Mapping("defaults", map[string]any{
"section_data": map[string]any{"first": "section"},
"backends": []any{map[string]any{"id": "base", "endpoint": "https://example.test/v1", "app_data": map[string]any{"first": "backend"}}},
"profiles": []any{map[string]any{"id": "base", "model": "model", "app_data": map[string]any{"first": "profile"}}},
"prompts": []any{map[string]any{"id": "base", "messages": []any{map[string]any{"role": "user", "content": "hello"}}, "output": map[string]any{"format": "text", "validation_mode": "none"}, "app_data": map[string]any{"first": "prompt"}}},
"collections": map[string]any{"team": map[string]any{"app_data": map[string]any{"first": "collection"}}},
}))
main := promptkit.SettingsInput(configkit.Mapping("operator", map[string]any{
"section_data": map[string]any{"second": "section"},
"backends": []any{map[string]any{"id": "child", "extends": map[string]any{"id": "base"}, "app_data": map[string]any{"second": "backend"}}},
"profiles": []any{map[string]any{"id": "child", "extends": map[string]any{"id": "base"}, "app_data": map[string]any{"second": "profile"}}},
"prompts": []any{map[string]any{"id": "child", "extends": map[string]any{"id": "base"}, "app_data": map[string]any{"second": "prompt"}}},
"collections": map[string]any{"team": map[string]any{"app_data": map[string]any{"second": "collection"}}},
}))
result, err := configkit.LoadLayered(main, configkit.LayeredOptions[root]{Definition: definition, Underlays: []configkit.Underlay{{Inputs: []configkit.Input{underlay}}}})
if err != nil {
t.Fatal(err)
}
settings := result.Config.PromptKit
for name, fields := range map[string]map[string]any{"section": settings.Application, "backend": settings.Backends[1].Application, "profile": settings.Profiles[1].Application, "prompt": settings.Prompts[1].Application, "collection": settings.Collections["team"].Application} {
key := "app_data"
if name == "section" {
key = "section_data"
}
data, ok := fields[key].(map[string]any)
if !ok || data["first"] != name || data["second"] != name {
t.Errorf("%s extension = %#v", name, fields[key])
}
}
}
func TestHandoffMissingAndNullSections(t *testing.T) {
type noSection struct {
Prompts []string `yaml:"prompts"`
}
missing, err := configkit.Load[noSection](configkit.Bytes("application-only", []byte("prompts: [application]\n")))
if err != nil {
t.Fatal(err)
}
empty, err := promptkit.NewFromConfig(missing, promptkit.RuntimeConfig{})
if err != nil {
t.Fatal(err)
}
if prompts, err := empty.ListPrompts(context.Background()); err != nil || len(prompts) != 0 {
t.Fatalf("application root resources were consumed: %+v, %v", prompts, err)
}
for _, source := range []string{"{}", "promptkit: null", "promptkit: {}"} {
t.Run(source, func(t *testing.T) {
type root struct {
PromptKit *promptkit.Settings `yaml:"promptkit"`
Prompts []string `yaml:"prompts"`
}
definition := promptkit.Configure(configkit.Definition[root]{}, func(value *root) *promptkit.Settings { return value.PromptKit })
result, err := configkit.Load(configkit.Bytes("missing", []byte(source)), configkit.Options[root]{Definition: definition})
if err != nil {
t.Fatal(err)
}
engine, err := promptkit.NewFromConfig(result, promptkit.RuntimeConfig{})
if err != nil {
t.Fatal(err)
}
prompts, err := engine.ListPrompts(context.Background())
if err != nil || len(prompts) != 0 {
t.Fatalf("prompts = %+v, %v", prompts, err)
}
})
}
if _, err := promptkit.NewFromConfig(nil, promptkit.RuntimeConfig{}); !errors.Is(err, promptkit.ErrInvalidConfig) {
t.Fatalf("nil result = %v", err)
}
}

View File

@@ -0,0 +1,90 @@
package promptkit_test
import (
"context"
"errors"
"testing"
"go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit"
)
func TestApplicationDefaultInputInspection(t *testing.T) {
for _, derived := range []bool{false, true} {
name := "Go defaults"
if derived {
name = "derived collection defaults"
}
t.Run(name, func(t *testing.T) {
settings := promptkit.Settings{
Profiles: []promptkit.ProfileSettings{{ID: "profile", Endpoint: new("https://example.test/v1"), Model: new("model")}},
Prompts: []promptkit.PromptSettings{{ID: "prompt", DefaultProfile: new("profile"),
Inputs: []promptkit.PromptInputSettings{{Name: "doc", Required: true, ContentType: new("text/plain"), Description: new("")}},
Messages: []promptkit.PromptMessageSettings{{Role: new("user"), Content: new("hello")}},
Output: &promptkit.PromptOutputSettings{Format: new("text"), ValidationMode: new("none")},
}},
}
definition := configkit.Definition[handoffApplication]{Defaults: func() handoffApplication { return handoffApplication{PromptKit: settings} }}
source := configkit.NoSource()
id := "prompt"
if derived {
definition.Defaults = nil
source = configkit.Bytes("derived inputs", []byte(`promptkit:
profiles: [{id: profile, endpoint: https://example.test/v1, model: model}]
collections:
team:
prompts:
- id: prompt
default_profile: profile
inputs: [{name: doc}]
messages: [{role: user, content: hello}]
output: {format: text, validation_mode: none}
`))
id = "team/prompt"
definition.Resolve = func(value *handoffApplication, resolution *configkit.Resolution) error {
input := &value.PromptKit.Collections["team"].Prompts[0].Inputs[0]
input.Required, input.ContentType, input.Description = true, new("text/plain"), new("")
base := configkit.Keys("promptkit", "collections", "team", "prompts").Index(0).Key("inputs").Index(0)
for _, field := range []string{"required", "content_type", "description"} {
resolution.RecordDerivedDefault(base.Key(field), "Application input default", base.Key("name"))
}
return nil
}
}
definition = promptkit.Configure(definition, func(value *handoffApplication) *promptkit.Settings { return &value.PromptKit })
result, err := configkit.Load(source, configkit.Options[handoffApplication]{Definition: definition})
if err != nil {
t.Fatal(err)
}
engine, err := promptkit.NewFromConfig(result, promptkit.RuntimeConfig{})
if err != nil {
t.Fatal(err)
}
ctx := context.Background()
resource, err := engine.InspectPrompt(ctx, id)
if err != nil {
t.Fatal(err)
}
input := (*resource.Inputs)[0]
if input.Required == nil || !*input.Required || input.ContentType == nil || *input.ContentType != "text/plain" || input.Description == nil || *input.Description != "" {
t.Fatalf("resource input = %+v", input)
}
requirements, err := engine.InspectRequirements(ctx, promptkit.RequirementInspectionRequest{PromptID: id})
if err != nil {
t.Fatal(err)
}
required := requirements.Inputs[0]
if required.Kind != promptkit.RequirementRequired || required.ContentType == nil || *required.ContentType != "text/plain" || required.Description == nil || *required.Description != "" {
t.Fatalf("requirement = %+v", required)
}
request := promptkit.RunRequest{PromptID: id}
if err := engine.ValidateRequest(ctx, request); !errors.Is(err, promptkit.ErrPromptRender) {
t.Fatalf("missing input: %v", err)
}
request.Inputs = map[string]promptkit.ArtifactRef{"doc": promptkit.Inline("document")}
if err := engine.ValidateRequest(ctx, request); err != nil {
t.Fatal(err)
}
})
}
}

View File

@@ -0,0 +1,43 @@
package promptkit_test
import (
"context"
"fmt"
"sync"
"testing"
"go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit"
)
type independentArtifactReader struct{}
func (independentArtifactReader) Read(_ context.Context, ref promptkit.ArtifactRef) (*promptkit.Artifact, error) {
return &promptkit.Artifact{Body: []byte(ref.Body), Hash: ref.Body}, nil
}
func TestConcurrentPreparationWithTransferredArtifacts(t *testing.T) {
input := configkit.Input{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("ownership", map[string]any{
"profiles": []any{map[string]any{"id": "profile", "endpoint": "https://example.invalid/v1", "model": "model"}},
"prompts": []any{map[string]any{"id": "prompt", "default_profile": "profile", "inputs": []any{map[string]any{"name": "document", "required": true}}, "messages": []any{map[string]any{"role": "user", "content": `{{input "document"}}`}}, "output": map[string]any{"format": "text", "validation_mode": "none"}}},
})}
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &input}}, promptkit.WithArtifactReader(independentArtifactReader{}))
if err != nil {
t.Fatal(err)
}
var group sync.WaitGroup
for i := range 24 {
group.Go(func() {
value := fmt.Sprintf("body-%d", i)
prepared, err := engine.Prepare(context.Background(), promptkit.RunRequest{PromptID: "prompt", Inputs: map[string]promptkit.ArtifactRef{"document": promptkit.Inline(value)}})
if err != nil {
t.Errorf("prepare: %v", err)
return
}
if len(prepared.Messages) != 1 || prepared.Messages[0].Content != value || prepared.InputHashes["document"] != value {
t.Errorf("prepared = %+v", prepared)
}
})
}
group.Wait()
}

View File

@@ -73,6 +73,7 @@ func newAssembledWorkflowFixture(b *testing.B, name string) assembledWorkflowFix
output := map[string]any{"format": "text", "validation_mode": "none"} output := map[string]any{"format": "text", "validation_mode": "none"}
profile := map[string]any{"id": "profile", "endpoint": "http://example.test/v1", "model": "benchmark-model"} profile := map[string]any{"id": "profile", "endpoint": "http://example.test/v1", "model": "benchmark-model"}
options := []promptkit.Option{promptkit.WithLLMClient(assembledWorkflowClient{content: response})} options := []promptkit.Option{promptkit.WithLLMClient(assembledWorkflowClient{content: response})}
var underlays []configkit.Underlay
var expectedMessage string var expectedMessage string
switch name { switch name {
@@ -91,9 +92,9 @@ func newAssembledWorkflowFixture(b *testing.B, name string) assembledWorkflowFix
"format": "json", "validation_mode": "json_schema", "schema_path": "workflow-schema.json", "format": "json", "validation_mode": "json_schema", "schema_path": "workflow-schema.json",
} }
options[0] = promptkit.WithLLMClient(assembledWorkflowClient{content: response}) options[0] = promptkit.WithLLMClient(assembledWorkflowClient{content: response})
options = append(options, promptkit.WithSchemaFS(fstest.MapFS{ underlays = append(underlays, configkit.Underlay{Inputs: []configkit.Input{schemaRootInput(fstest.MapFS{
"workflow-schema.json": &fstest.MapFile{Data: assembledWorkflowSchema(b)}, "workflow-schema.json": &fstest.MapFile{Data: assembledWorkflowSchema(b)},
}, ".")) }, ".")}})
default: default:
b.Fatalf("unknown assembled workflow fixture %q", name) b.Fatalf("unknown assembled workflow fixture %q", name)
} }
@@ -106,8 +107,8 @@ func newAssembledWorkflowFixture(b *testing.B, name string) assembledWorkflowFix
"output": output, "output": output,
}}, }},
} }
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &configkit.Input{ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Underlays: underlays, Main: &configkit.Input{
Source: configkit.Mapping("assembled workflow benchmark", config), Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("assembled workflow benchmark", config),
}}}, options...) }}}, options...)
if err != nil { if err != nil {
b.Fatal(err) b.Fatal(err)

View File

@@ -19,10 +19,13 @@ const BackendLocal = "local"
// explicit zero or blank values. A nil ExtraParams map is omitted; a non-nil // explicit zero or blank values. A nil ExtraParams map is omitted; a non-nil
// empty map supplies an empty mapping. A nil SmokeTests pointer is omitted; a // empty map supplies an empty mapping. A nil SmokeTests pointer is omitted; a
// non-nil pointer supplies a smoke_tests mapping. Use raw ConfigKit Mapping // non-nil pointer supplies a smoke_tests mapping. Use raw ConfigKit Mapping
// inputs for explicit null. BackendConfig has no stable JSON representation. // inputs for explicit null. Extends names a sibling backend in the same list.
// BackendConfig has no stable JSON representation.
type BackendConfig struct { type BackendConfig struct {
// ID is the local backend identity within its resource list. // ID is the local backend identity within its resource list.
ID string ID string
// Extends names a sibling backend to inherit through ConfigKit.
Extends *string
// Endpoint is the required absolute HTTP(S) base URL after all layers merge. // Endpoint is the required absolute HTTP(S) base URL after all layers merge.
Endpoint *string Endpoint *string
// APIKeyEnv optionally names an environment variable, not a credential. // APIKeyEnv optionally names an environment variable, not a credential.
@@ -38,51 +41,20 @@ type BackendConfig struct {
SmokeTests *BackendSmokeTestConfig SmokeTests *BackendSmokeTestConfig
} }
// BackendSmokeTestConfig contains optional diagnostic settings associated with
// a backend. Nil nested fields are omitted by BackendSource.
type BackendSmokeTestConfig struct {
// DefaultProfile is the exact profile ID selected as the backend's
// diagnostic inference default. A supplied value must not be blank; the ID
// is resolved only when that default is selected.
DefaultProfile *string
// Reachability optionally selects a full HTTP destination for reachability
// diagnostics. A nil endpoint uses the effective backend endpoint.
Reachability *ReachabilityProbeConfig
// Authentication optionally declares a named authentication diagnostic.
// A supplied object requires Kind.
Authentication *AuthenticationProbeConfig
}
// ReachabilityProbeConfig configures an optional reachability destination.
type ReachabilityProbeConfig struct {
// Endpoint is an absolute HTTP(S) URL, including any desired path or query.
// Nil selects the effective backend endpoint; an explicit blank value is
// invalid.
Endpoint *string
}
// AuthenticationProbeConfig configures an optional named authentication
// diagnostic.
type AuthenticationProbeConfig struct {
// Kind selects the authentication adapter. Supported values are "none" and
// "openrouter_key"; surrounding whitespace and case are normalized.
Kind *string
// Endpoint is an absolute HTTP(S) URL, including any desired path or query.
// Nil selects the adapter's backend-relative destination; an explicit blank
// value is invalid.
Endpoint *string
}
// BackendSource creates a labeled ConfigKit mapping containing backend records // BackendSource creates a labeled ConfigKit mapping containing backend records
// in caller order. IDs are explicit and duplicates are left for loading to reject. // in caller order. IDs are explicit and duplicates are left for loading to reject.
// It encodes fields without validating them; NewEngine validates the final // It encodes fields without validating them; NewEngine validates the final
// effective configuration after all selected source layers are combined. // effective configuration after all selected source layers are combined.
// Mapping captures a copy of supported input values at creation; unsupported // Mapping captures a copy of supported input values at creation; unsupported
// values fail when the source is loaded. // values fail when the source is loaded. ExtraParams must use ConfigKit Mapping
// input forms; callers must explicitly convert numbers or containers as needed.
func BackendSource(label string, backends []BackendConfig) configkit.Source { func BackendSource(label string, backends []BackendConfig) configkit.Source {
definitions := make([]any, 0, len(backends)) definitions := make([]any, 0, len(backends))
for _, config := range backends { for _, config := range backends {
fields := map[string]any{"id": config.ID} fields := map[string]any{"id": config.ID}
if config.Extends != nil {
fields["extends"] = map[string]any{"id": *config.Extends}
}
if config.Endpoint != nil { if config.Endpoint != nil {
fields["endpoint"] = *config.Endpoint fields["endpoint"] = *config.Endpoint
} }
@@ -96,7 +68,7 @@ func BackendSource(label string, backends []BackendConfig) configkit.Source {
fields["queue_capacity"] = *config.QueueCapacity fields["queue_capacity"] = *config.QueueCapacity
} }
if config.ExtraParams != nil { if config.ExtraParams != nil {
fields["extra_params"] = providerMapForMapping(config.ExtraParams) fields["extra_params"] = config.ExtraParams
} }
if config.SmokeTests != nil { if config.SmokeTests != nil {
smokeTests := make(map[string]any) smokeTests := make(map[string]any)

View File

@@ -0,0 +1,114 @@
package promptkit_test
import (
"context"
"reflect"
"slices"
"testing"
"go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit"
)
func TestBundledCompositionReplacementAndInheritance(t *testing.T) {
underlay := configkit.Input{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("caller underlay", map[string]any{
"profiles": []any{map[string]any{"id": "gpt-sol-latest", "model": "underlay-model"}},
})}
underlayEngine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Underlays: []configkit.Underlay{{Inputs: []configkit.Input{underlay}}}}})
if err != nil {
t.Fatal(err)
}
underlayProfile, err := underlayEngine.InspectProfile(context.Background(), "gpt-sol-latest")
if err != nil || underlayProfile.Model == nil || *underlayProfile.Model != "underlay-model" || underlayProfile.Backend != nil {
t.Fatalf("underlay replacement = %+v, %v", underlayProfile, err)
}
main := configkit.Input{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("caller main", map[string]any{
"backends": []any{map[string]any{"id": "openrouter", "endpoint": "https://local.invalid/v1"}, map[string]any{"id": "other", "endpoint": "https://other.invalid/v1"}},
"profiles": []any{
map[string]any{"id": "gpt-sol-latest", "model": "main-model"},
map[string]any{"id": "custom", "extends": map[string]any{"id": "gpt-luna-latest"}, "model": "custom-model"},
map[string]any{"id": "auto", "backend": "other", "model": "other-model"},
},
"collections": map[string]any{"openrouter": map[string]any{
"backends": []any{map[string]any{"id": "local", "endpoint": "https://collection.invalid/v1"}},
"profiles": []any{map[string]any{"id": "profile", "backend": "openrouter/local", "model": "collection-model"}},
}},
})}
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Underlays: []configkit.Underlay{{Inputs: []configkit.Input{underlay}}}, Main: &main}})
if err != nil {
t.Fatal(err)
}
inspect := func(id string) *promptkit.ProfileInspection {
t.Helper()
profile, err := engine.InspectProfile(context.Background(), id)
if err != nil {
t.Fatal(err)
}
return profile
}
if target := inspect("gpt-sol-latest"); target.Model == nil || *target.Model != "main-model" || target.Backend != nil {
t.Fatalf("replacement = %+v", target)
}
if target := inspect("custom"); target.Model == nil || *target.Model != "custom-model" || target.Backend == nil || *target.Backend != "openrouter" || target.Endpoint != nil {
t.Fatalf("inherited = %+v", target)
}
if backend, err := engine.InspectBackend(context.Background(), "openrouter"); err != nil || backend.Endpoint != "https://local.invalid/v1" || backend.APIKeyEnv != nil {
t.Fatalf("backend replacement = %+v, %v", backend, err)
}
if target := inspect("openrouter/profile"); target.Backend == nil || *target.Backend != "openrouter/local" || target.Model == nil || *target.Model != "collection-model" {
t.Fatalf("collection = %+v", target)
}
target := inspect("auto-low")
wantPlugins := map[string]any{"plugins": []any{map[string]any{"id": "auto-router", "cost_tier": "low"}}}
if target.Backend == nil || *target.Backend != "other" || target.Model == nil || *target.Model != "other-model" || target.ExtraParams == nil || !reflect.DeepEqual(*target.ExtraParams, wantPlugins) {
t.Errorf("inherited tier = %+v", target)
}
}
func TestBundledCompositionSequenceModes(t *testing.T) {
for _, test := range []struct {
name, mode, id string
wantErr bool
hasAuto bool
}{
{"duplicate append", "append", "auto", true, false},
{"distinct append", "append", "custom", false, true},
{"replace", "replace", "custom", false, false},
} {
t.Run(test.name, func(t *testing.T) {
input := configkit.Input{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("sequence", map[string]any{
"config": map[string]any{"sequences": []any{map[string]any{"destination": []any{"promptkit", "profiles"}, "mode": test.mode}}},
"profiles": []any{map[string]any{"id": test.id, "model": "custom-model"}},
})}
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &input}})
if test.wantErr {
if err == nil {
t.Fatal("expected duplicate rejection")
}
return
}
if err != nil {
t.Fatal(err)
}
profiles, err := engine.ListProfiles(context.Background())
if err != nil || !slices.Contains(profiles, "custom") || slices.Contains(profiles, "auto") != test.hasAuto {
t.Fatalf("profiles = %v, %v", profiles, err)
}
})
}
}
func TestBundledCompositionRejectsStrictConflictsAndMissingBackends(t *testing.T) {
profile := configkit.Input{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("profile", map[string]any{"profiles": []any{map[string]any{"id": "auto", "model": "caller"}}})}
_, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Underlays: []configkit.Underlay{{Inputs: []configkit.Input{profile, profile}}}}})
if err == nil {
t.Fatal("strict group accepted duplicate records")
}
input := configkit.Input{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("replace backends", map[string]any{
"config": map[string]any{"sequences": []any{map[string]any{"destination": []any{"promptkit", "backends"}, "mode": "replace"}}},
"backends": []any{map[string]any{"id": "other", "endpoint": "https://other.invalid/v1"}},
})}
if _, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &input}}); err == nil {
t.Fatal("missing bundled backend was accepted")
}
}

View File

@@ -0,0 +1,89 @@
package promptkit_test
import (
"context"
"encoding/json"
"errors"
"io"
"net/http"
"reflect"
"strings"
"testing"
"go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit"
)
func bundledPromptConfig(client *http.Client) promptkit.Config {
input := configkit.Input{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("prompt", map[string]any{"prompts": []any{map[string]any{
"id": "prompt", "messages": []any{map[string]any{"role": "user", "content": "hello"}}, "output": map[string]any{"format": "text", "validation_mode": "none"},
}}})}
return promptkit.Config{HTTPClient: client, Sources: promptkit.ConfigurationSources{Main: &input}}
}
func TestBundledProfilePreparationAndExecution(t *testing.T) {
client := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: "ok"}}
engine, err := promptkit.NewEngine(bundledPromptConfig(nil), promptkit.WithLLMClient(client))
if err != nil {
t.Fatal(err)
}
if _, err := engine.PrepareExecution(context.Background(), promptkit.RunRequest{PromptID: "prompt"}); !errors.Is(err, promptkit.ErrProfileRequired) {
t.Fatalf("missing profile = %v", err)
}
temperature := 0.3
request := promptkit.RunRequest{PromptID: "prompt", ProfileID: "gpt-luna-latest", Execution: &promptkit.ExecutionTargetOverride{Temperature: &temperature}}
prepared, err := engine.PrepareExecution(context.Background(), request)
if err != nil {
t.Fatal(err)
}
details := prepared.Details()
if details.SelectedProfileID != "gpt-luna-latest" || details.EffectiveModelParams.Model != "~openai/gpt-luna-latest" || details.EffectiveModelParams.BackendID != "openrouter" || details.EffectiveModelParams.Temperature != temperature {
t.Fatalf("prepared target = %+v", details.EffectiveModelParams)
}
if _, err := engine.Run(context.Background(), request); err != nil {
t.Fatal(err)
}
if len(client.requests) != 1 || client.requests[0].Target.Model != "~openai/gpt-luna-latest" || client.requests[0].Target.BackendID != "openrouter" {
t.Fatalf("requests = %+v", client.requests)
}
}
func TestBundledProfileHTTPPayloads(t *testing.T) {
t.Setenv("OPENROUTER_API_KEY", "synthetic-key")
for _, test := range []struct{ id, model, tier string }{
{"auto-low", "openrouter/auto", "low"}, {"gpt-luna-latest", "~openai/gpt-luna-latest", ""},
} {
t.Run(test.id, func(t *testing.T) {
var body map[string]any
transport := roundTripFunc(func(request *http.Request) (*http.Response, error) {
if request.Header.Get("Authorization") != "Bearer synthetic-key" {
t.Errorf("authorization = %q", request.Header.Get("Authorization"))
}
if err := json.NewDecoder(request.Body).Decode(&body); err != nil {
t.Fatal(err)
}
return &http.Response{StatusCode: 200, Header: make(http.Header), Body: io.NopCloser(strings.NewReader(`{"choices":[{"message":{"content":"ok"}}]}`))}, nil
})
engine, err := promptkit.NewEngine(bundledPromptConfig(&http.Client{Transport: transport}))
if err != nil {
t.Fatal(err)
}
if _, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt", ProfileID: test.id}); err != nil {
t.Fatal(err)
}
if body["model"] != test.model || body["extra_params"] != nil {
t.Fatalf("request = %+v", body)
}
if test.tier == "" {
if _, exists := body["plugins"]; exists {
t.Fatalf("unexpected plugins = %+v", body)
}
return
}
want := []any{map[string]any{"id": "auto-router", "cost_tier": test.tier}}
if !reflect.DeepEqual(body["plugins"], want) {
t.Fatalf("plugins = %#v, want %#v", body["plugins"], want)
}
})
}
}

View File

@@ -0,0 +1,117 @@
package promptkit_test
import (
"context"
"errors"
"net/http"
"reflect"
"strings"
"testing"
"go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit"
)
type rejectBundleTransport struct{}
func (rejectBundleTransport) RoundTrip(*http.Request) (*http.Response, error) {
return nil, errors.New("unexpected HTTP request")
}
func TestBundledOpenRouterOptionsAndInventory(t *testing.T) {
t.Setenv("OPENROUTER_API_KEY", "")
for _, test := range []struct {
name string
options []promptkit.Option
enabled bool
}{
{"default", nil, true}, {"enabled", []promptkit.Option{promptkit.WithBundledOpenRouter(true)}, true},
{"disabled", []promptkit.Option{promptkit.WithBundledOpenRouter(false)}, false},
{"last disables", []promptkit.Option{promptkit.WithBundledOpenRouter(true), promptkit.WithBundledOpenRouter(false)}, false},
{"last enables", []promptkit.Option{promptkit.WithBundledOpenRouter(false), promptkit.WithBundledOpenRouter(true)}, true},
} {
t.Run(test.name, func(t *testing.T) {
engine, err := promptkit.NewEngine(promptkit.Config{HTTPClient: &http.Client{Transport: rejectBundleTransport{}}}, test.options...)
if err != nil {
t.Fatal(err)
}
profiles, err := engine.ListProfiles(context.Background())
if err != nil {
t.Fatal(err)
}
backends, err := engine.ListBackends(context.Background())
if err != nil {
t.Fatal(err)
}
prompts, err := engine.ListPrompts(context.Background())
if err != nil {
t.Fatal(err)
}
collections, err := engine.ListCollections(context.Background())
if err != nil {
t.Fatal(err)
}
if (len(profiles) > 0) != test.enabled || len(backends) != boolCount(test.enabled) || len(prompts) != 0 || len(collections) != 0 {
t.Fatalf("inventories: %v %v %v %v", profiles, backends, prompts, collections)
}
if test.enabled {
if !reflect.DeepEqual(backends, []string{"openrouter"}) {
t.Fatalf("bundle inventory: %v %v", profiles, backends)
}
profile, err := engine.InspectProfile(context.Background(), "auto")
if err != nil || profile.Backend == nil || *profile.Backend != "openrouter" {
t.Fatalf("bundle profile = %+v, %v", profile, err)
}
view, err := engine.InspectConfiguration()
if err != nil {
t.Fatal(err)
}
if !strings.Contains(fmtSources(view.Sources()), "Promptkit bundled OpenRouter") {
t.Fatal("bundle source missing")
}
if explanation, found := view.Explain(configkit.Keys("promptkit", "backends").Index(0).Key("endpoint")); !found || !strings.Contains(explanation.Effective.YAML, "https://openrouter.ai/api/v1") || len(explanation.History) == 0 {
t.Fatalf("bundle explanation = %+v", explanation)
}
}
})
}
}
func boolCount(value bool) int {
if value {
return 1
}
return 0
}
func fmtSources(sources []configkit.SourceInfo) string {
var b strings.Builder
for _, source := range sources {
b.WriteString(source.Label)
}
return b.String()
}
func TestBundlePreservesCallerUnderlayBackingArray(t *testing.T) {
source := configkit.Mapping("caller", map[string]any{"profiles": []any{map[string]any{"id": "caller", "model": "model"}}})
groups := make([]configkit.Underlay, 1, 2)
groups[0] = configkit.Underlay{Inputs: []configkit.Input{{Destination: configkit.Keys("promptkit"), Source: source}}}
before := append([]configkit.Underlay(nil), groups...)
if _, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Underlays: groups}}); err != nil {
t.Fatal(err)
}
if !reflect.DeepEqual(groups, before) || !reflect.DeepEqual(groups[:2][1], configkit.Underlay{}) {
t.Fatal("caller underlays were modified")
}
}
func TestDisabledBundleKeepsCallerOpenRouter(t *testing.T) {
input := configkit.Input{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("caller", map[string]any{"backends": []any{map[string]any{"id": "openrouter", "endpoint": "https://local.invalid/v1"}}})}
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &input}}, promptkit.WithBundledOpenRouter(false))
if err != nil {
t.Fatal(err)
}
backends, err := engine.ListBackends(context.Background())
if err != nil || !reflect.DeepEqual(backends, []string{"openrouter"}) {
t.Fatalf("backends = %v, %v", backends, err)
}
}

View File

@@ -0,0 +1,181 @@
package promptkit_test
import (
"context"
"encoding/json"
"errors"
"io"
"net/http"
"reflect"
"strings"
"testing"
"go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit"
)
type cancelingAuthenticationBody struct {
remaining int
cancel context.CancelFunc
closed bool
}
func (b *cancelingAuthenticationBody) Read(p []byte) (int, error) {
n := min(len(p), b.remaining)
for i := range p[:n] {
p[i] = 'x'
}
b.remaining -= n
if b.remaining == 0 {
b.cancel()
return n, context.Canceled
}
return n, nil
}
func (b *cancelingAuthenticationBody) Close() error { b.closed = true; return nil }
func TestAuthenticationReadCancellationWinsOverOverflow(t *testing.T) {
for _, size := range []int{1024, (1 << 20) + 1} {
ctx, cancel := context.WithCancel(context.Background())
body := &cancelingAuthenticationBody{remaining: size, cancel: cancel}
engine, err := promptkit.NewEngine(promptkit.Config{HTTPClient: &http.Client{Transport: roundTripFunc(func(*http.Request) (*http.Response, error) {
return &http.Response{StatusCode: 200, Header: make(http.Header), Body: body}, nil
})}})
if err != nil {
cancel()
t.Fatal(err)
}
report, err := engine.SmokeTest(ctx, promptkit.SmokeTestRequest{Kind: promptkit.SmokeTestAuthentication, BackendID: "openrouter"})
cancel()
if !errors.Is(err, context.Canceled) || report == nil || report.Completed || report.Failure != promptkit.ExecutionFailureCanceled || report.Facts.CredentialsAccepted != nil || report.Facts.ResponseReceived == nil || !*report.Facts.ResponseReceived || !body.closed {
t.Fatalf("size %d: report=%+v error=%v closed=%v", size, report, err, body.closed)
}
}
}
func TestBundledSmokeHTTPDefaults(t *testing.T) {
t.Setenv("OPENROUTER_API_KEY", "synthetic-key")
var urls []string
var models []string
transport := roundTripFunc(func(request *http.Request) (*http.Response, error) {
urls = append(urls, request.URL.String())
body := `{"data":{"label":"member","is_free_tier":false,"usage":0}}`
if request.Method == http.MethodPost {
var payload map[string]any
if err := json.NewDecoder(request.Body).Decode(&payload); err != nil {
t.Fatal(err)
}
models = append(models, payload["model"].(string))
content := "PROMPTKIT_OK"
if _, structured := payload["response_format"]; structured {
content = `{"status":"ok"}`
}
encoded, err := json.Marshal(map[string]any{"choices": []any{map[string]any{"message": map[string]any{"content": content}}}})
if err != nil {
t.Fatal(err)
}
body = string(encoded)
} else if strings.HasSuffix(request.URL.Path, "/key") && request.Header.Get("Authorization") != "Bearer synthetic-key" {
t.Errorf("authentication header missing")
}
return &http.Response{StatusCode: 200, Header: make(http.Header), Body: io.NopCloser(strings.NewReader(body))}, nil
})
engine, err := promptkit.NewEngine(promptkit.Config{HTTPClient: &http.Client{Transport: transport}})
if err != nil {
t.Fatal(err)
}
before, _ := engine.ListProfiles(context.Background())
inspectionBefore, err := engine.InspectConfiguration()
if err != nil {
t.Fatal(err)
}
for _, kind := range []promptkit.SmokeTestKind{promptkit.SmokeTestReachability, promptkit.SmokeTestAuthentication, promptkit.SmokeTestTextInference, promptkit.SmokeTestStructuredOutput} {
report, err := engine.SmokeTest(context.Background(), promptkit.SmokeTestRequest{Kind: kind, BackendID: "openrouter"})
if err != nil || report == nil || !report.Completed || report.Failure != "" {
t.Fatalf("%s report = %+v, %v", kind, report, err)
}
if kind == promptkit.SmokeTestAuthentication && (report.Facts.CredentialsAccepted == nil || !*report.Facts.CredentialsAccepted) {
t.Fatalf("authentication facts = %+v", report.Facts)
}
if kind == promptkit.SmokeTestTextInference || kind == promptkit.SmokeTestStructuredOutput {
if report.Target.ProfileID != "gpt-luna-latest" || report.Target.EffectiveSettings.Model != "~openai/gpt-luna-latest" {
t.Fatalf("target = %+v", report.Target)
}
}
}
wantURLs := []string{"https://openrouter.ai/api/v1", "https://openrouter.ai/api/v1/key", "https://openrouter.ai/api/v1/chat/completions", "https://openrouter.ai/api/v1/chat/completions"}
if !reflect.DeepEqual(urls, wantURLs) || !reflect.DeepEqual(models, []string{"~openai/gpt-luna-latest", "~openai/gpt-luna-latest"}) {
t.Fatalf("urls %v, models %v", urls, models)
}
after, _ := engine.ListProfiles(context.Background())
if !reflect.DeepEqual(before, after) {
t.Fatalf("inventory changed: %v %v", before, after)
}
inspectionAfter, err := engine.InspectConfiguration()
if err != nil {
t.Fatal(err)
}
diff, err := inspectionBefore.Diff(inspectionAfter)
if err != nil || len(diff) != 0 {
t.Fatalf("inspection changed: %+v, %v", diff, err)
}
}
func TestBundledSmokeSelectionUsesEffectiveConfiguration(t *testing.T) {
replacement := configkit.Input{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("replacement", map[string]any{"profiles": []any{map[string]any{"id": "gpt-luna-latest", "backend": "openrouter", "model": "replacement"}}})}
replacementClient := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: "PROMPTKIT_OK"}}
replacementEngine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &replacement}}, promptkit.WithLLMClient(replacementClient))
if err != nil {
t.Fatal(err)
}
replacementReport, err := replacementEngine.SmokeTest(context.Background(), promptkit.SmokeTestRequest{Kind: promptkit.SmokeTestTextInference, BackendID: "openrouter"})
if err != nil || replacementReport.Target.ProfileID != "gpt-luna-latest" || replacementReport.Target.EffectiveSettings.Model != "replacement" {
t.Fatalf("replaced default = %+v, %v", replacementReport, err)
}
replacementClient.response.Content = "unexpected"
mismatch, err := replacementEngine.SmokeTest(context.Background(), promptkit.SmokeTestRequest{Kind: promptkit.SmokeTestTextInference, BackendID: "openrouter"})
if err != nil || mismatch == nil || !mismatch.Completed {
t.Fatalf("concluded mismatch = %+v, %v", mismatch, err)
}
input := configkit.Input{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("application", map[string]any{
"backends": []any{map[string]any{"id": "openrouter", "endpoint": "https://example.invalid/v1", "smoke_tests": map[string]any{"default_profile": "custom"}}},
"profiles": []any{map[string]any{"id": "gpt-luna-latest", "backend": "openrouter", "model": "replacement"}, map[string]any{"id": "custom", "backend": "openrouter", "model": "custom-model"}},
"prompts": []any{map[string]any{"id": "prompt", "default_profile": "free", "messages": []any{map[string]any{"role": "user", "content": "private application marker"}}, "output": map[string]any{"format": "text", "validation_mode": "none"}}},
})}
client := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: "PROMPTKIT_OK"}}
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &input}}, promptkit.WithLLMClient(client))
if err != nil {
t.Fatal(err)
}
for _, test := range []struct {
request promptkit.SmokeTestRequest
profile, model string
}{
{promptkit.SmokeTestRequest{BackendID: "openrouter"}, "custom", "custom-model"},
{promptkit.SmokeTestRequest{BackendID: "openrouter", PromptID: "prompt"}, "free", "openrouter/free"},
{promptkit.SmokeTestRequest{BackendID: "openrouter", PromptID: "prompt", ProfileID: "gpt-luna-latest"}, "gpt-luna-latest", "replacement"},
} {
test.request.Kind = promptkit.SmokeTestTextInference
report, err := engine.SmokeTest(context.Background(), test.request)
if err != nil || report.Target.ProfileID != test.profile || report.Target.EffectiveSettings.Model != test.model {
t.Fatalf("report = %+v, %v", report, err)
}
if strings.Contains(client.requests[len(client.requests)-1].Prompt.Messages[0].Content, "private application marker") {
t.Fatal("application content used by diagnostic")
}
}
}
func TestBundledSmokeAuthenticationResponseOverflow(t *testing.T) {
transport := roundTripFunc(func(*http.Request) (*http.Response, error) {
return &http.Response{StatusCode: 200, Header: make(http.Header), Body: io.NopCloser(strings.NewReader(strings.Repeat("x", (1<<20)+1)))}, nil
})
engine, err := promptkit.NewEngine(promptkit.Config{HTTPClient: &http.Client{Transport: transport}})
if err != nil {
t.Fatal(err)
}
report, err := engine.SmokeTest(context.Background(), promptkit.SmokeTestRequest{Kind: promptkit.SmokeTestAuthentication, BackendID: "openrouter", APIKey: "private-key-marker"})
if err != nil || report == nil || !report.Completed || report.Reason != promptkit.SmokeTestReasonAuthenticationResponseLimitExceeded || report.Failure != promptkit.ExecutionFailureOther || report.Facts.CredentialsAccepted != nil {
t.Fatalf("overflow = %+v, %v", report, err)
}
}

View File

@@ -265,7 +265,7 @@ func newBackendCapacityEngine(
options = append(options, promptkit.WithArtifactReader(reader)) options = append(options, promptkit.WithArtifactReader(reader))
} }
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &configkit.Input{ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &configkit.Input{
Source: configkit.Mapping("capacity fixture", map[string]any{ Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("capacity fixture", map[string]any{
"backends": []any{backend}, "backends": []any{backend},
"profiles": []any{map[string]any{"id": "profile", "backend": "limited", "model": "model"}}, "profiles": []any{map[string]any{"id": "profile", "backend": "limited", "model": "model"}},
"prompts": []any{definition}, "prompts": []any{definition},

38
configuration.go Normal file
View File

@@ -0,0 +1,38 @@
package promptkit
import (
"go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit/internal/configuration"
)
// Configure composes PromptKit's settings rules with an application definition.
// selectSettings identifies the application's PromptKit section and must be
// non-nil. A nil selected section is left untouched. Existing resolution and
// validation callbacks run first; callers may wrap the returned callbacks to
// run additional work afterward. Defaults and version policy are preserved.
func Configure[T any](definition configkit.Definition[T], selectSettings func(*T) *Settings) configkit.Definition[T] {
previousResolve := definition.Resolve
previousValidate := definition.Validate
previousSensitive := definition.Sensitive
identities := append([]configkit.SequenceIdentity(nil), definition.SequenceIdentities...)
definition.SequenceIdentities = append(identities, configuration.SequenceIdentities()...)
definition.Resolve = func(value *T, resolution *configkit.Resolution) error {
if previousResolve != nil {
if err := previousResolve(value, resolution); err != nil {
return err
}
}
configuration.Resolve(selectSettings(value), resolution)
return nil
}
definition.Validate = func(value *T, validation *configkit.Validation) {
if previousValidate != nil {
previousValidate(value, validation)
}
configuration.Validate(selectSettings(value), validation)
}
definition.Sensitive = func(path configkit.Path) bool {
return (previousSensitive != nil && previousSensitive(path)) || configuration.Sensitive(path)
}
return definition
}

View File

@@ -21,7 +21,7 @@ func TestConfigurationInspectionDiffsCapturedValues(t *testing.T) {
t.Fatalf("difference count = %d, want 1: %+v", len(differences), differences) t.Fatalf("difference count = %d, want 1: %+v", len(differences), differences)
} }
difference := differences[0] difference := differences[0]
if got, want := difference.Path.String(), "/profiles"; got != want { if got, want := difference.Path.String(), "/promptkit/profiles"; got != want {
t.Fatalf("difference path = %q, want %q", got, want) t.Fatalf("difference path = %q, want %q", got, want)
} }
if !strings.Contains(difference.Left.YAML, "earlier-model") || !strings.Contains(difference.Right.YAML, "later-model") { if !strings.Contains(difference.Left.YAML, "earlier-model") || !strings.Contains(difference.Right.YAML, "later-model") {
@@ -81,6 +81,34 @@ func TestConfigurationInspectionDiffIgnoresSourceHistory(t *testing.T) {
} }
} }
func TestConfigurationInspectionDiffUsesNormalizedValues(t *testing.T) {
inspect := func(endpoint string) promptkit.ConfigurationInspection {
t.Helper()
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Underlays: []configkit.Underlay{{Inputs: []configkit.Input{{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("backend", map[string]any{
"backends": []any{map[string]any{"id": "local", "endpoint": endpoint}},
})}}}}}})
if err != nil {
t.Fatal(err)
}
view, err := engine.InspectConfiguration()
if err != nil {
t.Fatal(err)
}
return view
}
left := inspect(" HTTPS://example.invalid/v1 ")
equal := inspect("https://example.invalid/v1")
differences, err := left.Diff(equal)
if err != nil || len(differences) != 0 {
t.Fatalf("normalized diff = %+v, %v", differences, err)
}
changed := inspect("https://other.invalid/v1")
differences, err = left.Diff(changed)
if err != nil || len(differences) != 1 || differences[0].Path.String() != "/promptkit/backends" {
t.Fatalf("changed diff = %+v, %v", differences, err)
}
}
func TestConfigurationInspectionDiffReportsPresenceAndSequences(t *testing.T) { func TestConfigurationInspectionDiffReportsPresenceAndSequences(t *testing.T) {
left := comparisonInspectionFrom(t, "before", "same-model", map[string]any{ left := comparisonInspectionFrom(t, "before", "same-model", map[string]any{
"sequence": []any{"earlier", "unchanged"}, "sequence": []any{"earlier", "unchanged"},
@@ -94,7 +122,7 @@ func TestConfigurationInspectionDiffReportsPresenceAndSequences(t *testing.T) {
if err != nil { if err != nil {
t.Fatalf("compare presence and sequence values: %v", err) t.Fatalf("compare presence and sequence values: %v", err)
} }
if len(differences) != 1 || differences[0].Path.String() != "/backends" { if len(differences) != 1 || differences[0].Path.String() != "/promptkit/backends" {
t.Fatalf("atomic backend list difference = %+v", differences) t.Fatalf("atomic backend list difference = %+v", differences)
} }
difference := differences[0] difference := differences[0]
@@ -119,7 +147,7 @@ func TestConfigurationInspectionDiffProtectsSensitiveValuesAndReturnedRecords(t
if err != nil { if err != nil {
t.Fatalf("compare sensitive values: %v", err) t.Fatalf("compare sensitive values: %v", err)
} }
if len(differences) != 1 || differences[0].Path.String() != "/backends" { if len(differences) != 1 || differences[0].Path.String() != "/promptkit/backends" {
t.Fatalf("sensitive backend list difference = %+v", differences) t.Fatalf("sensitive backend list difference = %+v", differences)
} }
secret := differences[0] secret := differences[0]
@@ -148,7 +176,7 @@ func TestConfigurationInspectionDiffProtectsSensitiveValuesAndReturnedRecords(t
if _, exists := againByPath["/caller/mutation"]; exists { if _, exists := againByPath["/caller/mutation"]; exists {
t.Fatal("mutating returned path reached retained inspection") t.Fatal("mutating returned path reached retained inspection")
} }
if _, exists := againByPath["/backends"]; !exists { if _, exists := againByPath["/promptkit/backends"]; !exists {
t.Fatalf("original backend path was lost after mutation: %+v", again) t.Fatalf("original backend path was lost after mutation: %+v", again)
} }
for _, difference := range again { for _, difference := range again {
@@ -173,7 +201,7 @@ func TestConfigurationInspectionDiffMasksSensitiveShapeReplacement(t *testing.T)
if err != nil { if err != nil {
t.Fatalf("compare sensitive shape replacement: %v", err) t.Fatalf("compare sensitive shape replacement: %v", err)
} }
if len(differences) != 1 || differences[0].Path.String() != "/backends" { if len(differences) != 1 || differences[0].Path.String() != "/promptkit/backends" {
t.Fatalf("shape replacement differences = %+v", differences) t.Fatalf("shape replacement differences = %+v", differences)
} }
difference := differences[0] difference := differences[0]
@@ -197,7 +225,7 @@ func comparisonInspectionFrom(t *testing.T, source, model string, extraParams ma
backend["extra_params"] = extraParams backend["extra_params"] = extraParams
} }
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
Underlays: []configkit.Underlay{{Inputs: []configkit.Input{{Source: configkit.Mapping(source, map[string]any{ Underlays: []configkit.Underlay{{Inputs: []configkit.Input{{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping(source, map[string]any{
"backends": []any{backend}, "backends": []any{backend},
"profiles": []any{map[string]any{ "profiles": []any{map[string]any{
"id": "offline", "backend": "local", "id": "offline", "backend": "local",

View File

@@ -0,0 +1,76 @@
package promptkit_test
import (
"context"
"errors"
"testing"
"go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit"
)
type handoffApplication struct {
PromptKit promptkit.Settings `yaml:"promptkit"`
}
type handoffWrapper struct {
result *configkit.Result[handoffApplication]
override any
useOverride bool
inspection bool
}
func (w *handoffWrapper) Value(path configkit.Path) (any, bool) {
if w.useOverride {
return w.override, true
}
return w.result.Value(path)
}
func (w *handoffWrapper) InspectSubtree(path configkit.Path) (configkit.Inspection, bool) {
if !w.inspection {
return configkit.Inspection{}, false
}
return w.result.InspectSubtree(path)
}
func (w *handoffWrapper) Diagnostics() []configkit.Diagnostic { return w.result.Diagnostics() }
func TestConstructorsShareCapturedResources(t *testing.T) {
source := configkit.Bytes("prompt", []byte("promptkit:\n prompts:\n - id: example\n messages: [{role: user, content: hello}]\n output: {format: text, validation_mode: basic}\n"))
definition := promptkit.Configure(configkit.Definition[handoffApplication]{}, func(value *handoffApplication) *promptkit.Settings { return &value.PromptKit })
result, err := configkit.Load(source, configkit.Options[handoffApplication]{Definition: definition})
if err != nil {
t.Fatal(err)
}
wrapped := &handoffWrapper{result: result, inspection: true}
applicationEngine, err := promptkit.NewFromConfig(wrapped, promptkit.RuntimeConfig{}, promptkit.WithBundledOpenRouter(true))
if err != nil {
t.Fatal(err)
}
input := configkit.Input{Source: source}
standalone, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &input}}, promptkit.WithBundledOpenRouter(false))
if err != nil {
t.Fatal(err)
}
for _, engine := range []*promptkit.Engine{applicationEngine, standalone} {
prompt, err := engine.InspectPrompt(context.Background(), "example")
if err != nil || prompt.Messages[0].Content != "hello" || prompt.OutputContract.RepairAttempts != 1 {
t.Fatalf("prompt = %+v, %v", prompt, err)
}
backends, err := engine.ListBackends(context.Background())
if err != nil || len(backends) != 0 {
t.Fatalf("backends = %+v, %v", backends, err)
}
}
}
func TestNewFromConfigRejectsInvalidHandoff(t *testing.T) {
var nilResult *handoffWrapper
if _, err := promptkit.NewFromConfig(nilResult, promptkit.RuntimeConfig{}); !errors.Is(err, promptkit.ErrInvalidConfig) {
t.Fatalf("nil result: %v", err)
}
for _, wrapper := range []*handoffWrapper{{inspection: false}, {inspection: true, useOverride: true, override: "wrong"}} {
if _, err := promptkit.NewFromConfig(wrapper, promptkit.RuntimeConfig{}); !errors.Is(err, promptkit.ErrInvalidConfig) {
t.Fatalf("invalid handoff: %v", err)
}
}
}

View File

@@ -9,8 +9,10 @@ import (
) )
// ConfigurationInspection is a read-only view of one Engine's captured // ConfigurationInspection is a read-only view of one Engine's captured
// effective configuration and source history. It does not expose typed // effective configuration and source history focused on /promptkit. Metadata,
// configuration values, filesystem contexts, or runtime credentials. // sources, and explicit path queries retain native whole-load evidence; the
// focus is not an isolation boundary. It does not expose typed configuration
// values, filesystem contexts, or runtime credentials.
// Report content may include application prompt and provider data, so callers // Report content may include application prompt and provider data, so callers
// choose where and to whom to disclose it. // choose where and to whom to disclose it.
type ConfigurationInspection struct { type ConfigurationInspection struct {
@@ -30,7 +32,9 @@ func (e *Engine) InspectConfiguration() (ConfigurationInspection, error) {
// Diff compares this view's captured effective configuration with other. This // Diff compares this view's captured effective configuration with other. This
// view is the earlier (left) value and other is the later (right) value. The // view is the earlier (left) value and other is the later (right) value. The
// comparison delegates to ConfigKit's frozen effective-value semantics, so // comparison delegates to ConfigKit's frozen effective-value semantics, so
// source or resolution history alone does not create differences. It does not // source or resolution history alone does not create differences. Native
// ancestor differences are retained, while sibling-only records are excluded.
// It does not
// reread external sources or compare referenced file contents, resolved // reread external sources or compare referenced file contents, resolved
// credential contents, later execution preparation, or request-specific // credential contents, later execution preparation, or request-specific
// overrides. Either zero view returns ErrInvalidConfig and no records; other // overrides. Either zero view returns ErrInvalidConfig and no records; other
@@ -52,6 +56,14 @@ func (v ConfigurationInspection) Metadata() configkit.Metadata {
return v.source.Metadata() return v.source.Metadata()
} }
// HistoricalPaths lists native historical paths within the PromptKit focus.
func (v ConfigurationInspection) HistoricalPaths() []configkit.Path {
if v.source == nil {
return nil
}
return v.source.HistoricalPaths()
}
// Sources returns a detached list of loaded source descriptions. // Sources returns a detached list of loaded source descriptions.
func (v ConfigurationInspection) Sources() []configkit.SourceInfo { func (v ConfigurationInspection) Sources() []configkit.SourceInfo {
if v.source == nil { if v.source == nil {
@@ -78,7 +90,8 @@ func (v ConfigurationInspection) Explain(path configkit.Path) (configkit.Explana
return v.source.Explain(path) return v.source.Explain(path)
} }
// WriteEffective writes ConfigKit's redacted effective YAML. Writer errors // WriteEffective writes the focused ConfigKit-redacted effective YAML. An absent
// effective focus fails without output; explicit null renders normally. Writer errors
// remain available through errors.Is and errors.As. The zero view returns // remain available through errors.Is and errors.As. The zero view returns
// ErrInvalidConfig. // ErrInvalidConfig.
func (v ConfigurationInspection) WriteEffective(w io.Writer) error { func (v ConfigurationInspection) WriteEffective(w io.Writer) error {

View File

@@ -19,9 +19,9 @@ import (
func TestConfigurationInspectionReportsLiteralKeysAndRedactedHistory(t *testing.T) { func TestConfigurationInspectionReportsLiteralKeysAndRedactedHistory(t *testing.T) {
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
Underlays: []configkit.Underlay{ Underlays: []configkit.Underlay{
{Inputs: []configkit.Input{{Source: configkit.Mapping("defaults", map[string]any{ {Inputs: []configkit.Input{{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("defaults", map[string]any{
"backends": []any{map[string]any{ "backends": []any{map[string]any{
"id": "local", "endpoint": "https://example.invalid/v1", "api_key_env": "OPTIONAL_KEY", "id": "local", "endpoint": " HTTPS://example.invalid/v1 ", "api_key_env": "OPTIONAL_KEY",
"extra_params": map[string]any{"token": "private-sentinel", "visible": "public-value"}, "extra_params": map[string]any{"token": "private-sentinel", "visible": "public-value"},
}}, }},
"collections": map[string]any{"literal": map[string]any{ "collections": map[string]any{"literal": map[string]any{
@@ -35,14 +35,14 @@ func TestConfigurationInspectionReportsLiteralKeysAndRedactedHistory(t *testing.
}, },
}}, }},
})}}}, })}}},
{Inputs: []configkit.Input{{Source: configkit.Mapping("operator", map[string]any{ {Inputs: []configkit.Input{{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("operator", map[string]any{
"collections": map[string]any{"literal": map[string]any{ "collections": map[string]any{"literal": map[string]any{
"backends": []any{map[string]any{"id": "local", "endpoint": "https://example.invalid/v1", "extra_params": map[string]any{"token": "replacement-private-sentinel"}}}, "backends": []any{map[string]any{"id": "local", "endpoint": "https://example.invalid/v1", "extra_params": map[string]any{"token": "replacement-private-sentinel"}}},
"profiles": []any{map[string]any{"id": "base.v1", "backend": "local", "model": "operator-model"}}, "profiles": []any{map[string]any{"id": "base.v1", "backend": "local", "model": "operator-model"}},
}}, }},
})}}}, })}}},
}, },
}}) }}, promptkit.WithBundledOpenRouter(false))
if err != nil { if err != nil {
t.Fatalf("construct layered engine: %v", err) t.Fatalf("construct layered engine: %v", err)
} }
@@ -50,30 +50,35 @@ func TestConfigurationInspectionReportsLiteralKeysAndRedactedHistory(t *testing.
if err != nil { if err != nil {
t.Fatalf("inspect configuration: %v", err) t.Fatalf("inspect configuration: %v", err)
} }
modelPath := configkit.Keys("collections", "literal", "profiles").Index(0).Key("model") modelPath := configkit.Keys("promptkit", "collections", "literal", "profiles").Index(0).Key("model")
model, found := view.Explain(modelPath) model, found := view.Explain(modelPath)
if !found || !strings.Contains(model.Effective.YAML, "operator-model") || len(model.History) == 0 { if !found || !strings.Contains(model.Effective.YAML, "operator-model") || len(model.History) == 0 {
t.Fatalf("layered model explanation = (%+v, %t)", model, found) t.Fatalf("layered model explanation = (%+v, %t)", model, found)
} }
inheritedPath := configkit.Keys("collections", "literal", "profiles").Index(1).Key("model") inheritedPath := configkit.Keys("promptkit", "collections", "literal", "profiles").Index(1).Key("model")
inherited, found := view.Explain(inheritedPath) inherited, found := view.Explain(inheritedPath)
if !found || !strings.Contains(inherited.Effective.YAML, "operator-model") || len(inherited.History) == 0 || !view.Metadata().IsExplicit(inheritedPath) { if !found || !strings.Contains(inherited.Effective.YAML, "operator-model") || len(inherited.History) == 0 || !view.Metadata().IsExplicit(inheritedPath) {
t.Fatalf("literal-key inheritance explanation = (%+v, %t)", inherited, found) t.Fatalf("literal-key inheritance explanation = (%+v, %t)", inherited, found)
} }
if inheritedPath.String() != "/collections/literal/profiles/1/model" { if inheritedPath.String() != "/promptkit/collections/literal/profiles/1/model" {
t.Fatalf("literal-key path = %q", inheritedPath) t.Fatalf("literal-key path = %q", inheritedPath)
} }
secretPath := configkit.Keys("backends").Index(0).Key("extra_params").Key("token") secretPath := configkit.Keys("promptkit", "backends").Index(0).Key("extra_params").Key("token")
endpointPath := configkit.Keys("promptkit", "backends").Index(0).Key("endpoint")
endpoint, found := view.Explain(endpointPath)
if !found || !strings.Contains(endpoint.Effective.YAML, "https://example.invalid/v1") || strings.Contains(fmt.Sprintf("%+v", view.Metadata().History(endpointPath)), "private-sentinel") {
t.Fatalf("normalized endpoint explanation = %+v", endpoint)
}
secret, found := view.Explain(secretPath) secret, found := view.Explain(secretPath)
if !found || !secret.Effective.Redacted || strings.Contains(secret.Effective.YAML, "private-sentinel") { if !found || !secret.Effective.Redacted || strings.Contains(secret.Effective.YAML, "private-sentinel") {
t.Fatalf("credential-shaped provider value was not redacted: (%+v, %t)", secret, found) t.Fatalf("credential-shaped provider value was not redacted: (%+v, %t)", secret, found)
} }
namespacedSecretPath := configkit.Keys("collections", "literal", "backends").Index(0).Key("extra_params").Key("token") namespacedSecretPath := configkit.Keys("promptkit", "collections", "literal", "backends").Index(0).Key("extra_params").Key("token")
namespacedSecret, found := view.Explain(namespacedSecretPath) namespacedSecret, found := view.Explain(namespacedSecretPath)
if !found || !namespacedSecret.Effective.Redacted || strings.Contains(namespacedSecret.Effective.YAML, "namespaced-private-sentinel") { if !found || !namespacedSecret.Effective.Redacted || strings.Contains(namespacedSecret.Effective.YAML, "namespaced-private-sentinel") {
t.Fatalf("namespaced provider value was not redacted: (%+v, %t)", namespacedSecret, found) t.Fatalf("namespaced provider value was not redacted: (%+v, %t)", namespacedSecret, found)
} }
inheritedSecretPath := configkit.Keys("collections", "literal", "backends").Index(1).Key("extra_params").Key("token") inheritedSecretPath := configkit.Keys("promptkit", "collections", "literal", "backends").Index(1).Key("extra_params").Key("token")
inheritedSecret, found := view.Explain(inheritedSecretPath) inheritedSecret, found := view.Explain(inheritedSecretPath)
if !found || !inheritedSecret.Effective.Redacted || strings.Contains(inheritedSecret.Effective.YAML, "replacement-private-sentinel") { if !found || !inheritedSecret.Effective.Redacted || strings.Contains(inheritedSecret.Effective.YAML, "replacement-private-sentinel") {
t.Fatalf("inherited provider value was not redacted: (%+v, %t)", inheritedSecret, found) t.Fatalf("inherited provider value was not redacted: (%+v, %t)", inheritedSecret, found)
@@ -119,7 +124,7 @@ func TestConfigurationInspectionReportsLiteralKeysAndRedactedHistory(t *testing.
func TestConfigurationInspectionFailureDoesNotDiscloseSensitiveValues(t *testing.T) { func TestConfigurationInspectionFailureDoesNotDiscloseSensitiveValues(t *testing.T) {
_, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &configkit.Input{ _, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &configkit.Input{
Source: configkit.Mapping("invalid backend", map[string]any{ Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("invalid backend", map[string]any{
"collections": map[string]any{"private": map[string]any{"backends": []any{ "collections": map[string]any{"private": map[string]any{"backends": []any{
map[string]any{"id": "backend", "endpoint": "invalid endpoint", "extra_params": map[string]any{"token": "failure-private-sentinel"}}, map[string]any{"id": "backend", "endpoint": "invalid endpoint", "extra_params": map[string]any{"token": "failure-private-sentinel"}},
}}}, }}},
@@ -134,9 +139,9 @@ func TestConfigurationInspectionFailureDoesNotDiscloseSensitiveValues(t *testing
} }
} }
func TestConfigurationInspectionRetainsMainSpecificationDeclaration(t *testing.T) { func TestConfigurationInspectionRetainsApplicationVersionField(t *testing.T) {
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &configkit.Input{ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &configkit.Input{
Source: configkit.Mapping("main", map[string]any{"version": "2"}), Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("main", map[string]any{"version": "2"}),
}}}) }}})
if err != nil { if err != nil {
t.Fatalf("construct engine with accepted declaration: %v", err) t.Fatalf("construct engine with accepted declaration: %v", err)
@@ -145,7 +150,7 @@ func TestConfigurationInspectionRetainsMainSpecificationDeclaration(t *testing.T
if err != nil { if err != nil {
t.Fatalf("inspect configuration: %v", err) t.Fatalf("inspect configuration: %v", err)
} }
path := configkit.Keys("version") path := configkit.Keys("promptkit", "version")
field, found := inspection.Explain(path) field, found := inspection.Explain(path)
if !found || !inspection.Metadata().IsExplicit(path) || !strings.Contains(field.Effective.YAML, "2") { if !found || !inspection.Metadata().IsExplicit(path) || !strings.Contains(field.Effective.YAML, "2") {
t.Fatalf("main specification declaration was not retained: (%+v, %t)", field, found) t.Fatalf("main specification declaration was not retained: (%+v, %t)", field, found)
@@ -166,7 +171,7 @@ func TestConfigurationInspectionZeroAndWriterErrors(t *testing.T) {
t.Fatalf("nil engine inspection = %v", err) t.Fatalf("nil engine inspection = %v", err)
} }
var empty promptkit.ConfigurationInspection var empty promptkit.ConfigurationInspection
path := configkit.Keys("profiles", "missing") path := configkit.Keys("promptkit", "profiles", "missing")
if _, found := empty.Metadata().Lookup(path); found || len(empty.Sources()) != 0 || len(empty.Diagnostics()) != 0 { if _, found := empty.Metadata().Lookup(path); found || len(empty.Sources()) != 0 || len(empty.Diagnostics()) != 0 {
t.Fatal("zero inspection reported configuration") t.Fatal("zero inspection reported configuration")
} }
@@ -180,7 +185,7 @@ func TestConfigurationInspectionZeroAndWriterErrors(t *testing.T) {
t.Fatalf("zero explanation writer = %v", err) t.Fatalf("zero explanation writer = %v", err)
} }
profile := configkit.Input{Source: configkit.Mapping("inspection test profile", map[string]any{ profile := configkit.Input{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("inspection test profile", map[string]any{
"profiles": []any{map[string]any{"id": "inspection-profile", "endpoint": "https://configured.example/v1", "model": "configured-model"}}, "profiles": []any{map[string]any{"id": "inspection-profile", "endpoint": "https://configured.example/v1", "model": "configured-model"}},
})} })}
valid, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{ valid, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
@@ -198,7 +203,7 @@ func TestConfigurationInspectionZeroAndWriterErrors(t *testing.T) {
if err := view.WriteEffective(writer); !errors.Is(err, writeErr) { if err := view.WriteEffective(writer); !errors.Is(err, writeErr) {
t.Fatalf("effective writer error = %v", err) t.Fatalf("effective writer error = %v", err)
} }
if err := view.WriteExplanation(writer, configkit.Keys("profiles").Index(0).Key("model")); !errors.Is(err, writeErr) { if err := view.WriteExplanation(writer, configkit.Keys("promptkit", "profiles").Index(0).Key("model")); !errors.Is(err, writeErr) {
t.Fatalf("explanation writer error = %v", err) t.Fatalf("explanation writer error = %v", err)
} }
} }
@@ -225,7 +230,7 @@ func TestConfigurationInspectionUsesCapturedSourcesWithoutProviderCalls(t *testi
client := countingLLMClient{calls: &modelCalls} client := countingLLMClient{calls: &modelCalls}
endpoint, model, otherModel := "https://example.invalid/v1", "offline-model", "comparison-model" endpoint, model, otherModel := "https://example.invalid/v1", "offline-model", "comparison-model"
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
Underlays: []configkit.Underlay{{Inputs: inputs}, {Inputs: []configkit.Input{{Source: configkit.Mapping("offline profile", map[string]any{ Underlays: []configkit.Underlay{{Inputs: mountSettingsInputs(inputs)}, {Inputs: []configkit.Input{{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("offline profile", map[string]any{
"profiles": []any{map[string]any{"id": "offline", "endpoint": endpoint, "model": model}}, "profiles": []any{map[string]any{"id": "offline", "endpoint": endpoint, "model": model}},
})}}}}, })}}}},
}}, promptkit.WithLLMClient(client)) }}, promptkit.WithLLMClient(client))
@@ -233,7 +238,7 @@ func TestConfigurationInspectionUsesCapturedSourcesWithoutProviderCalls(t *testi
t.Fatalf("construct engine: %v", err) t.Fatalf("construct engine: %v", err)
} }
otherEngine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{ otherEngine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
Underlays: []configkit.Underlay{{Inputs: otherInputs}, {Inputs: []configkit.Input{{Source: configkit.Mapping("comparison profile", map[string]any{ Underlays: []configkit.Underlay{{Inputs: mountSettingsInputs(otherInputs)}, {Inputs: []configkit.Input{{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("comparison profile", map[string]any{
"profiles": []any{map[string]any{"id": "offline", "endpoint": endpoint, "model": otherModel}}, "profiles": []any{map[string]any{"id": "offline", "endpoint": endpoint, "model": otherModel}},
})}}}}, })}}}},
}}, promptkit.WithLLMClient(client)) }}, promptkit.WithLLMClient(client))
@@ -256,7 +261,7 @@ func TestConfigurationInspectionUsesCapturedSourcesWithoutProviderCalls(t *testi
if err != nil || len(differences) != 1 { if err != nil || len(differences) != 1 {
t.Fatalf("compare captured configurations = (%+v, %v), want one difference", differences, err) t.Fatalf("compare captured configurations = (%+v, %v), want one difference", differences, err)
} }
if differences[0].Path.String() != "/profiles" || if differences[0].Path.String() != "/promptkit/profiles" ||
!strings.Contains(differences[0].Left.YAML, model) || !strings.Contains(differences[0].Left.YAML, model) ||
!strings.Contains(differences[0].Right.YAML, otherModel) { !strings.Contains(differences[0].Right.YAML, otherModel) {
t.Fatalf("captured difference = %+v", differences[0]) t.Fatalf("captured difference = %+v", differences[0])
@@ -268,7 +273,7 @@ func TestConfigurationInspectionUsesCapturedSourcesWithoutProviderCalls(t *testi
t.Fatalf("captured difference changed between reads: first=%+v next=%+v", expected, differences[0]) t.Fatalf("captured difference changed between reads: first=%+v next=%+v", expected, differences[0])
} }
} }
if _, found := view.Explain(configkit.Keys("prompts").Index(0).Key("messages").Index(0).Key("content_file")); !found { if _, found := view.Explain(configkit.Keys("promptkit", "prompts").Index(0).Key("messages").Index(0).Key("content_file")); !found {
t.Fatal("captured prompt content path was not inspectable") t.Fatal("captured prompt content path was not inspectable")
} }
var report bytes.Buffer var report bytes.Buffer

View File

@@ -0,0 +1,60 @@
package promptkit_test
import (
"io/fs"
"testing"
"testing/fstest"
"go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit"
)
type mountApplication struct {
PromptKit promptkit.Settings `yaml:"promptkit"`
}
type countingMountFS struct {
fs.FS
opens int
}
func (c *countingMountFS) Open(name string) (fs.File, error) { c.opens++; return c.FS.Open(name) }
func TestSettingsAndCollectionMounts(t *testing.T) {
files := &countingMountFS{FS: fstest.MapFS{
"fragments/root.yaml": {Data: []byte("profiles:\n - id: main\n model: model\n")},
"fragments/collection.yaml": {Data: []byte("prompts:\n - id: child\n messages: [{role: user, content_file: body.txt}]\n output: {format: text, validation_mode: basic}\n")},
"fragments/item.yaml": {Data: []byte("id: item\nmessages: [{role: user, content: hello}]\noutput: {format: text, validation_mode: basic}\n")},
"fragments/body.txt": {Data: []byte("message")},
}}
namespace := configkit.NewFilesystem("mounts", files)
root := promptkit.SettingsInput(namespace.File("fragments/root.yaml"))
collection, err := promptkit.CollectionInput("docs", namespace.File("fragments/collection.yaml"))
if err != nil {
t.Fatal(err)
}
item := configkit.Input{Source: namespace.File("fragments/item.yaml"), Destination: configkit.Keys("promptkit", "prompts"), PlacementMode: configkit.PlacementModeSequenceItem}
for _, input := range []*configkit.Input{&root, &collection, &item} {
input.Filesystem = &configkit.FilesystemContext{FS: &namespace, Root: "."}
}
bundled := promptkit.BundledOpenRouterUnderlay()
if files.opens != 0 || root.Imports || !collection.Imports || root.Destination.String() != "/promptkit" || collection.Destination.String() != "/promptkit/collections/docs" {
t.Fatalf("mount construction read files or chose wrong defaults: opens=%d", files.opens)
}
definition := promptkit.Configure(configkit.Definition[mountApplication]{}, func(value *mountApplication) *promptkit.Settings { return &value.PromptKit })
result, err := configkit.LoadLayered(configkit.Input{Source: configkit.NoSource()}, configkit.LayeredOptions[mountApplication]{Definition: definition, Underlays: []configkit.Underlay{bundled, {Inputs: []configkit.Input{root, collection, item}}}})
if err != nil {
t.Fatal(err)
}
if len(result.Config.PromptKit.Profiles) < 1 || len(result.Config.PromptKit.Prompts) != 1 || len(result.Config.PromptKit.Collections["docs"].Prompts) != 1 {
t.Fatalf("mounted settings = %+v", result.Config.PromptKit)
}
info, ok := result.Metadata().Lookup(configkit.Keys("promptkit", "collections", "docs", "prompts").Index(0).Key("messages").Index(0).Key("content_file"))
if !ok || len(info.Origin.Locations) != 1 {
t.Fatalf("origin = %+v", info)
}
context, ok := result.SourceContext(info.Origin.Locations[0].SourceID)
if !ok || context.FS == nil || context.BaseDir != "fragments" {
t.Fatalf("context = %+v", context)
}
}

View File

@@ -0,0 +1,99 @@
package promptkit_test
import (
"errors"
"strings"
"sync"
"testing"
"go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit"
)
type rulesApplication struct {
Version string `yaml:"version"`
Other []rulesOther `yaml:"other"`
PromptKit *promptkit.Settings `yaml:"promptkit"`
}
type rulesOther struct {
ID string `yaml:"id"`
}
func TestConfigureComposesApplicationRules(t *testing.T) {
var events []string
definition := configkit.Definition[rulesApplication]{
Defaults: func() rulesApplication { return rulesApplication{Version: "app"} },
Version: &configkit.VersionRule{Path: configkit.Keys("version"), Check: func(v configkit.Version, validation *configkit.Validation) error {
events = append(events, "version")
return nil
}},
SequenceIdentities: []configkit.SequenceIdentity{configkit.SequenceIdentityAt(configkit.Keys("other"), "id")},
Resolve: func(_ *rulesApplication, _ *configkit.Resolution) error {
events = append(events, "application resolve")
return nil
},
Validate: func(_ *rulesApplication, _ *configkit.Validation) { events = append(events, "application validate") },
Sensitive: func(path configkit.Path) bool { return path.String() == "/application_secret" },
}
configured := promptkit.Configure(definition, func(v *rulesApplication) *promptkit.Settings { return v.PromptKit })
if len(definition.SequenceIdentities) != 1 || len(configured.SequenceIdentities) != 7 {
t.Fatalf("registrations: %d, %d", len(definition.SequenceIdentities), len(configured.SequenceIdentities))
}
priorValidate := configured.Validate
configured.Validate = func(value *rulesApplication, validation *configkit.Validation) {
priorValidate(value, validation)
if value.PromptKit == nil || value.PromptKit.Backends[0].Endpoint == nil || strings.TrimSpace(*value.PromptKit.Backends[0].Endpoint) != *value.PromptKit.Backends[0].Endpoint {
validation.Error(configkit.Keys("promptkit"), "normalization did not run")
}
events = append(events, "after validate")
}
source := configkit.Bytes("rules", []byte("promptkit:\n backends:\n - id: local\n endpoint: ' https://example.test/v1/ '\n extra_params: {api_key: hidden}\nother: [{id: one}]\n"))
result, err := configkit.Load(source, configkit.Options[rulesApplication]{Definition: configured})
if err != nil {
var loadError *configkit.LoadError
if errors.As(err, &loadError) {
t.Fatalf("%v: %+v", err, loadError.Diagnostics())
}
t.Fatal(err)
}
if result.Config.Version != "app" || strings.Join(events, ",") != "version,application resolve,application validate,after validate" {
t.Fatalf("callbacks: %+v, %+v", result.Config, events)
}
if !configured.Sensitive(configkit.Keys("application_secret")) || !configured.Sensitive(configkit.Keys("promptkit", "backends").Index(0).Key("extra_params").Key("api_key")) || configured.Sensitive(configkit.Keys("backends").Index(0).Key("extra_params").Key("api_key")) {
t.Fatal("sensitivity composition failed")
}
var rendered strings.Builder
if err := result.WriteEffective(&rendered); err != nil || strings.Contains(rendered.String(), "hidden") {
t.Fatalf("redaction: %q, %v", rendered.String(), err)
}
}
func TestConfigurePropagatesResolveError(t *testing.T) {
want := errors.New("resolve failed")
definition := promptkit.Configure(configkit.Definition[rulesApplication]{Resolve: func(*rulesApplication, *configkit.Resolution) error { return want }}, func(v *rulesApplication) *promptkit.Settings { return v.PromptKit })
err := definition.Resolve(&rulesApplication{}, nil)
if !errors.Is(err, want) {
t.Fatalf("resolve error = %v", err)
}
}
func TestConfigureReusableAcrossConcurrentLoads(t *testing.T) {
definition := promptkit.Configure(configkit.Definition[rulesApplication]{}, func(v *rulesApplication) *promptkit.Settings { return v.PromptKit })
var group sync.WaitGroup
for range 8 {
group.Go(func() {
for _, source := range []string{"{}", "promptkit: {prompts: [{id: test, output: {format: text, validation_mode: basic}}]}"} {
result, err := configkit.Load(configkit.Bytes("settings", []byte(source)), configkit.Options[rulesApplication]{Definition: definition})
if err != nil {
t.Error(err)
continue
}
if result.Config.PromptKit != nil && *result.Config.PromptKit.Prompts[0].Output.RepairAttempts != 1 {
t.Error("default was not resolved")
}
}
})
}
group.Wait()
}

View File

@@ -8,12 +8,13 @@ import (
"path/filepath" "path/filepath"
"go.maximumdirect.net/configkit" "go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit/internal/configuration"
"go.maximumdirect.net/promptkit/internal/domain" "go.maximumdirect.net/promptkit/internal/domain"
"go.maximumdirect.net/promptkit/internal/filecatalog" "go.maximumdirect.net/promptkit/internal/filecatalog"
) )
// CollectionInput mounts source at the selected collection's configuration // CollectionInput mounts source at the selected collection's configuration
// destination, collections/<namespace>, and enables ConfigKit imports for that // destination, /promptkit/collections/<namespace>, and enables ConfigKit imports for that
// input. An invalid namespace returns an error matching ErrInvalidConfig. The // input. An invalid namespace returns an error matching ErrInvalidConfig. The
// source is not read or validated until ConfigKit loads it. Callers may adjust // source is not read or validated until ConfigKit loads it. Callers may adjust
// the returned input, including its filesystem context or imports setting. // the returned input, including its filesystem context or imports setting.
@@ -23,11 +24,23 @@ func CollectionInput(namespace string, source configkit.Source) (configkit.Input
} }
return configkit.Input{ return configkit.Input{
Source: source, Source: source,
Destination: configkit.Keys("collections", namespace), Destination: configkit.Keys("promptkit", "collections", namespace),
Imports: true, Imports: true,
}, nil }, nil
} }
// SettingsInput mounts a wrapperless PromptKit settings fragment at /promptkit.
// It does not acquire the source; callers may adjust the returned input.
func SettingsInput(source configkit.Source) configkit.Input {
return configkit.Input{Source: source, Destination: configkit.Keys("promptkit")}
}
// BundledOpenRouterUnderlay selects the bundled OpenRouter settings fragment.
// It does not acquire its source until ConfigKit loads it.
func BundledOpenRouterUnderlay() configkit.Underlay {
return configuration.BundledOpenRouterUnderlay()
}
// DirectoryInputs selects the YAML files beneath an explicitly chosen host // DirectoryInputs selects the YAML files beneath an explicitly chosen host
// directory as one strict-composition group. The returned order is deterministic; // directory as one strict-composition group. The returned order is deterministic;
// it does not assign precedence among those files. Discovery is recursive over // it does not assign precedence among those files. Discovery is recursive over

View File

@@ -23,8 +23,8 @@ func TestCollectionInputValidatesWithoutReadingSource(t *testing.T) {
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
if input.Destination.String() != "/collections/docs" || !input.Imports { if input.Destination.String() != "/promptkit/collections/docs" || !input.Imports {
t.Fatalf("collection input = %+v, want destination /collections/docs with imports enabled", input) t.Fatalf("collection input = %+v, want destination /promptkit/collections/docs with imports enabled", input)
} }
for _, namespace := range []string{"", " ", "docs/api"} { for _, namespace := range []string{"", " ", "docs/api"} {
if _, err := promptkit.CollectionInput(namespace, configkit.NoSource()); !errors.Is(err, promptkit.ErrInvalidConfig) { if _, err := promptkit.CollectionInput(namespace, configkit.NoSource()); !errors.Is(err, promptkit.ErrInvalidConfig) {
@@ -51,18 +51,59 @@ func TestCollectionInputLoadsNativeImportsAndLaterOverride(t *testing.T) {
Underlays: []configkit.Underlay{{Inputs: []configkit.Input{input}}}, Underlays: []configkit.Underlay{{Inputs: []configkit.Input{input}}},
Main: &configkit.Input{ Main: &configkit.Input{
Source: configkit.Mapping("later collection override", map[string]any{"prompts": []any{map[string]any{"id": "greeting", "inputs": []any{map[string]any{"name": "overridden"}}, "messages": []any{map[string]any{"role": "user", "content": "Updated"}}, "output": map[string]any{"format": "text", "validation_mode": "none"}}}}), Source: configkit.Mapping("later collection override", map[string]any{"prompts": []any{map[string]any{"id": "greeting", "inputs": []any{map[string]any{"name": "overridden"}}, "messages": []any{map[string]any{"role": "user", "content": "Updated"}}, "output": map[string]any{"format": "text", "validation_mode": "none"}}}}),
Destination: configkit.Keys("collections", "docs"), Destination: configkit.Keys("promptkit", "collections", "docs"),
}, },
}}) }})
if err != nil { if err != nil {
t.Fatalf("construct engine with native collection import: %v", err) t.Fatalf("construct engine with native collection import: %v", err)
} }
inspection, err := engine.InspectPrompt(context.Background(), "docs/greeting") inspection, err := engine.InspectPrompt(context.Background(), "docs/greeting")
if err != nil || inspection.PromptID != "docs/greeting" || len(inspection.Inputs) != 1 || inspection.Inputs[0].Name != "overridden" { if err != nil || inspection.PromptID != "docs/greeting" || inspection.Inputs == nil || len(*inspection.Inputs) != 1 || (*inspection.Inputs)[0].Name != "overridden" {
t.Fatalf("inspect imported and overlaid prompt = (%+v, %v)", inspection, err) t.Fatalf("inspect imported and overlaid prompt = (%+v, %v)", inspection, err)
} }
} }
func TestCollectionInputUsesMainRelativeContext(t *testing.T) {
filesystem := configkit.NewFilesystem("application", fstest.MapFS{
"site/main.yaml": {Data: []byte("version: '2'\n")},
"site/prompts/greeting.yaml": {Data: []byte(`id: greeting
messages:
- role: user
content_file: greeting.txt
output: {format: text, validation_mode: none}
`)},
"site/prompts/greeting.txt": {Data: []byte("Hello from the imported prompt's directory.")},
})
companion, err := promptkit.CollectionInput("support", configkit.Bytes("companion prompts", []byte(`
config:
imports:
- dir: prompts
pattern: ["*.yaml", "*.yml"]
destination: [prompts]
placement: sequence-item
required: true
`)))
if err != nil {
t.Fatal(err)
}
companion.MainRelative = configkit.MainRelative()
main := configkit.Input{Source: filesystem.File("site/main.yaml")}
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
Main: &main,
Underlays: []configkit.Underlay{{Inputs: []configkit.Input{companion}}},
}}, promptkit.WithBundledOpenRouter(false))
if err != nil {
t.Fatalf("construct engine with main-relative imports: %v", err)
}
inspection, err := engine.InspectPrompt(t.Context(), "support/greeting")
if err != nil {
t.Fatal(err)
}
if len(inspection.Messages) != 1 || inspection.Messages[0].Content != "Hello from the imported prompt's directory." {
t.Fatalf("unexpected imported messages: %+v", inspection.Messages)
}
}
func TestCollectionInputLoadsEmbeddedImportsAndLaterOverride(t *testing.T) { func TestCollectionInputLoadsEmbeddedImportsAndLaterOverride(t *testing.T) {
files := fstest.MapFS{ files := fstest.MapFS{
"collection.yaml": {Data: []byte("config:\n imports:\n - path: defaults.yaml\n required: true\n")}, "collection.yaml": {Data: []byte("config:\n imports:\n - path: defaults.yaml\n required: true\n")},
@@ -78,14 +119,14 @@ func TestCollectionInputLoadsEmbeddedImportsAndLaterOverride(t *testing.T) {
Underlays: []configkit.Underlay{{Inputs: []configkit.Input{input}}}, Underlays: []configkit.Underlay{{Inputs: []configkit.Input{input}}},
Main: &configkit.Input{ Main: &configkit.Input{
Source: configkit.Mapping("later collection override", map[string]any{"prompts": []any{map[string]any{"id": "greeting", "inputs": []any{map[string]any{"name": "overridden"}}, "messages": []any{map[string]any{"role": "user", "content": "Updated"}}, "output": map[string]any{"format": "text", "validation_mode": "none"}}}}), Source: configkit.Mapping("later collection override", map[string]any{"prompts": []any{map[string]any{"id": "greeting", "inputs": []any{map[string]any{"name": "overridden"}}, "messages": []any{map[string]any{"role": "user", "content": "Updated"}}, "output": map[string]any{"format": "text", "validation_mode": "none"}}}}),
Destination: configkit.Keys("collections", "docs"), Destination: configkit.Keys("promptkit", "collections", "docs"),
}, },
}}) }})
if err != nil { if err != nil {
t.Fatalf("construct engine with embedded collection import: %v", err) t.Fatalf("construct engine with embedded collection import: %v", err)
} }
inspection, err := engine.InspectPrompt(context.Background(), "docs/greeting") inspection, err := engine.InspectPrompt(context.Background(), "docs/greeting")
if err != nil || inspection.PromptID != "docs/greeting" || len(inspection.Inputs) != 1 || inspection.Inputs[0].Name != "overridden" { if err != nil || inspection.PromptID != "docs/greeting" || inspection.Inputs == nil || len(*inspection.Inputs) != 1 || (*inspection.Inputs)[0].Name != "overridden" {
t.Fatalf("inspect imported and overlaid prompt = (%+v, %v)", inspection, err) t.Fatalf("inspect imported and overlaid prompt = (%+v, %v)", inspection, err)
} }
} }

View File

@@ -41,7 +41,7 @@ func TestProfileEnvironmentLookupAcceptsNonShellNames(t *testing.T) {
endpoint, model := "http://example.test/v1", "model" endpoint, model := "http://example.test/v1", "model"
lowerName := name lowerName := name
cfg.Sources.Underlays = append(cfg.Sources.Underlays, configkit.Underlay{Inputs: []configkit.Input{{ cfg.Sources.Underlays = append(cfg.Sources.Underlays, configkit.Underlay{Inputs: []configkit.Input{{
Source: promptkit.ProfileSource("default credentials", []promptkit.ProfileConfig{ Destination: configkit.Keys("promptkit"), Source: promptkit.ProfileSource("default credentials", []promptkit.ProfileConfig{
{ID: "profile", Endpoint: &endpoint, Model: &model, APIKeyEnv: &lowerName}, {ID: "profile", Endpoint: &endpoint, Model: &model, APIKeyEnv: &lowerName},
}), }),
}}}) }}})
@@ -49,7 +49,7 @@ func TestProfileEnvironmentLookupAcceptsNonShellNames(t *testing.T) {
if test.clear { if test.clear {
env = "" env = ""
} }
cfg.Sources.Main = &configkit.Input{Source: promptkit.ProfileSource("credentials", []promptkit.ProfileConfig{ cfg.Sources.Main = &configkit.Input{Destination: configkit.Keys("promptkit"), Source: promptkit.ProfileSource("credentials", []promptkit.ProfileConfig{
{ID: "profile", Endpoint: &endpoint, Model: &model, APIKeyEnv: &env}, {ID: "profile", Endpoint: &endpoint, Model: &model, APIKeyEnv: &env},
})} })}
var calls int var calls int
@@ -73,7 +73,7 @@ func TestProfileEnvironmentLookupAcceptsNonShellNames(t *testing.T) {
func TestPublicConstructionWithLocalBackendAndProgrammaticProfile(t *testing.T) { func TestPublicConstructionWithLocalBackendAndProgrammaticProfile(t *testing.T) {
backendID, model := "workstation", "offline-model" backendID, model := "workstation", "offline-model"
prompt := configkit.Input{Source: configkit.Mapping("prompt", map[string]any{ prompt := configkit.Input{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("prompt", map[string]any{
"prompts": []any{map[string]any{ "prompts": []any{map[string]any{
"id": "hello", "id": "hello",
"default_profile": "offline", "default_profile": "offline",
@@ -86,10 +86,10 @@ func TestPublicConstructionWithLocalBackendAndProgrammaticProfile(t *testing.T)
localBackend.ID = backendID localBackend.ID = backendID
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
Underlays: []configkit.Underlay{{Inputs: []configkit.Input{ Underlays: []configkit.Underlay{{Inputs: []configkit.Input{
{Source: promptkit.BackendSource("local backend", []promptkit.BackendConfig{ {Destination: configkit.Keys("promptkit"), Source: promptkit.BackendSource("local backend", []promptkit.BackendConfig{
localBackend, localBackend,
})}, })},
{Source: promptkit.ProfileSource("profile", []promptkit.ProfileConfig{ {Destination: configkit.Keys("promptkit"), Source: promptkit.ProfileSource("profile", []promptkit.ProfileConfig{
{ID: "offline", Backend: &backendID, Model: &model}, {ID: "offline", Backend: &backendID, Model: &model},
})}, })},
}}}, }}},
@@ -116,11 +116,11 @@ func TestFormerCatalogBackendIDsWorkWhenConfigured(t *testing.T) {
model := "configured-model" model := "configured-model"
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
Underlays: []configkit.Underlay{{Inputs: []configkit.Input{ Underlays: []configkit.Underlay{{Inputs: []configkit.Input{
{Source: promptkit.BackendSource("application backends", []promptkit.BackendConfig{ {Destination: configkit.Keys("promptkit"), Source: promptkit.BackendSource("application backends", []promptkit.BackendConfig{
{ID: openRouterID, Endpoint: &openRouterEndpoint}, {ID: openRouterID, Endpoint: &openRouterEndpoint},
{ID: rakestrawhomeID, Endpoint: &rakestrawhomeEndpoint}, {ID: rakestrawhomeID, Endpoint: &rakestrawhomeEndpoint},
})}, })},
{Source: promptkit.ProfileSource("application profiles", []promptkit.ProfileConfig{ {Destination: configkit.Keys("promptkit"), Source: promptkit.ProfileSource("application profiles", []promptkit.ProfileConfig{
{ID: "openrouter-profile", Backend: &openRouterID, Model: &model}, {ID: "openrouter-profile", Backend: &openRouterID, Model: &model},
{ID: "rakestrawhome-profile", Backend: &rakestrawhomeID, Model: &model}, {ID: "rakestrawhome-profile", Backend: &rakestrawhomeID, Model: &model},
})}, })},
@@ -134,7 +134,7 @@ func TestFormerCatalogBackendIDsWorkWhenConfigured(t *testing.T) {
"rakestrawhome-profile": rakestrawhomeID, "rakestrawhome-profile": rakestrawhomeID,
} { } {
inspection, err := engine.InspectProfile(context.Background(), profileID) inspection, err := engine.InspectProfile(context.Background(), profileID)
if err != nil || inspection.EffectiveModelParams.BackendID != backendID { if err != nil || inspection.Backend == nil || *inspection.Backend != backendID {
t.Fatalf("inspect %q = (%+v, %v), want backend %q", profileID, inspection, err, backendID) t.Fatalf("inspect %q = (%+v, %v), want backend %q", profileID, inspection, err, backendID)
} }
} }
@@ -170,7 +170,7 @@ prompts:
t.Fatalf("select operator inputs: %v", err) t.Fatalf("select operator inputs: %v", err)
} }
programmaticModel, programmaticEndpoint := "programmatic-model", "https://example.invalid/v1" programmaticModel, programmaticEndpoint := "programmatic-model", "https://example.invalid/v1"
programmatic := configkit.Input{Source: promptkit.ProfileSource("programmatic settings", []promptkit.ProfileConfig{ programmatic := configkit.Input{Destination: configkit.Keys("promptkit"), Source: promptkit.ProfileSource("programmatic settings", []promptkit.ProfileConfig{
{ID: "application", Endpoint: &programmaticEndpoint, Model: &programmaticModel}, {ID: "application", Endpoint: &programmaticEndpoint, Model: &programmaticModel},
})} })}
for _, test := range []struct { for _, test := range []struct {
@@ -183,9 +183,9 @@ prompts:
{name: "programmatic", apply: true, want: "programmatic-model"}, {name: "programmatic", apply: true, want: "programmatic-model"},
} { } {
t.Run(test.name, func(t *testing.T) { t.Run(test.name, func(t *testing.T) {
underlays := []configkit.Underlay{{Inputs: embedded}} underlays := []configkit.Underlay{{Inputs: mountSettingsInputs(embedded)}}
if test.apply { if test.apply {
underlays = append(underlays, configkit.Underlay{Inputs: operator}) underlays = append(underlays, configkit.Underlay{Inputs: mountSettingsInputs(operator)})
} }
if test.name == "programmatic" { if test.name == "programmatic" {
underlays = append(underlays, configkit.Underlay{Inputs: []configkit.Input{programmatic}}) underlays = append(underlays, configkit.Underlay{Inputs: []configkit.Input{programmatic}})
@@ -207,7 +207,7 @@ prompts:
func TestPublicConstructionPreservesConfigKitLoadError(t *testing.T) { func TestPublicConstructionPreservesConfigKitLoadError(t *testing.T) {
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &configkit.Input{ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &configkit.Input{
Source: configkit.Mapping("invalid", map[string]any{"profiles": []any{map[string]any{"id": "broken", "unknown": true}}}), Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("invalid", map[string]any{"prompts": []any{map[string]any{"id": "broken", "output": map[string]any{"unknown": true}}}}),
}}}) }}})
var loadErr *configkit.LoadError var loadErr *configkit.LoadError
if engine != nil || !errors.Is(err, promptkit.ErrInvalidConfig) || !errors.As(err, &loadErr) { if engine != nil || !errors.Is(err, promptkit.ErrInvalidConfig) || !errors.As(err, &loadErr) {
@@ -215,6 +215,36 @@ func TestPublicConstructionPreservesConfigKitLoadError(t *testing.T) {
} }
} }
func TestPublicConstructionPreservesMalformedYAMLCause(t *testing.T) {
const sourceLabel = "prompts.yml"
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &configkit.Input{
Source: configkit.Bytes(sourceLabel, []byte("profiles: [\n")),
}}})
var loadErr *configkit.LoadError
if engine != nil || !errors.Is(err, promptkit.ErrInvalidConfig) || !errors.As(err, &loadErr) {
t.Fatalf("construction = (%#v, %v), want invalid config preserving LoadError", engine, err)
}
cause := errors.Unwrap(err)
var causeLoadErr *configkit.LoadError
if cause == nil || errors.Is(cause, promptkit.ErrInvalidConfig) || !errors.As(cause, &causeLoadErr) || causeLoadErr != loadErr {
t.Fatalf("construction cause = %v, want the original loading failure without a category branch", cause)
}
for _, diagnostic := range loadErr.Diagnostics() {
if diagnostic.Code != "yaml.invalid" {
continue
}
if diagnostic.Primary == nil {
t.Fatal("YAML diagnostic has no source location")
}
source, ok := loadErr.Source(diagnostic.Primary.SourceID)
if !ok || source.Label != sourceLabel {
t.Fatalf("YAML diagnostic source = %+v, found=%v, want %q", source, ok, sourceLabel)
}
return
}
t.Fatalf("loading diagnostics = %+v, want yaml.invalid", loadErr.Diagnostics())
}
func TestPublicConstructionRejectsSuppliedInvalidMainInput(t *testing.T) { func TestPublicConstructionRejectsSuppliedInvalidMainInput(t *testing.T) {
main := configkit.Input{} main := configkit.Input{}
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &main}}) engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &main}})
@@ -227,10 +257,10 @@ func TestPublicConstructionUsesProgrammaticBackendAndProfileSources(t *testing.T
endpoint, backendID, model := "http://application.example/v1", "application-backend", "synthetic-model" endpoint, backendID, model := "http://application.example/v1", "application-backend", "synthetic-model"
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
Underlays: []configkit.Underlay{{Inputs: []configkit.Input{ Underlays: []configkit.Underlay{{Inputs: []configkit.Input{
{Source: promptkit.BackendSource("operator backend", []promptkit.BackendConfig{ {Destination: configkit.Keys("promptkit"), Source: promptkit.BackendSource("operator backend", []promptkit.BackendConfig{
{ID: backendID, Endpoint: &endpoint}, {ID: backendID, Endpoint: &endpoint},
})}, })},
{Source: promptkit.ProfileSource("operator profile", []promptkit.ProfileConfig{ {Destination: configkit.Keys("promptkit"), Source: promptkit.ProfileSource("operator profile", []promptkit.ProfileConfig{
{ID: "application-profile", Backend: &backendID, Model: &model}, {ID: "application-profile", Backend: &backendID, Model: &model},
})}, })},
}}}, }}},
@@ -242,8 +272,9 @@ func TestPublicConstructionUsesProgrammaticBackendAndProfileSources(t *testing.T
if err != nil { if err != nil {
t.Fatalf("inspect configured profile: %v", err) t.Fatalf("inspect configured profile: %v", err)
} }
if inspection.EffectiveModelParams.Endpoint != endpoint { backendInspection, err := engine.InspectBackend(context.Background(), backendID)
t.Fatalf("effective endpoint = %q, want %q", inspection.EffectiveModelParams.Endpoint, endpoint) if err != nil || inspection.Endpoint != nil || backendInspection.Endpoint != endpoint {
t.Fatalf("profile/backend = %+v / %+v, %v", inspection, backendInspection, err)
} }
} }
@@ -273,7 +304,7 @@ func TestBackendSourceOwnsSmokeTestValuesAndDefersDefaultLookup(t *testing.T) {
authenticationEndpoint = "https://mutated.example/key" authenticationEndpoint = "https://mutated.example/key"
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
Underlays: []configkit.Underlay{{Inputs: []configkit.Input{{Source: source}}}}, Underlays: []configkit.Underlay{{Inputs: []configkit.Input{{Destination: configkit.Keys("promptkit"), Source: source}}}},
}}) }})
if err != nil { if err != nil {
t.Fatalf("construct engine with dangling diagnostic default: %v", err) t.Fatalf("construct engine with dangling diagnostic default: %v", err)
@@ -291,7 +322,7 @@ func TestBackendSourceOwnsSmokeTestValuesAndDefersDefaultLookup(t *testing.T) {
"https://backend.example/v1", "https://backend.example/v1",
"not-yet-listed", "not-yet-listed",
"https://probe.example/ready?check=1", "https://probe.example/ready?check=1",
" OPENROUTER_KEY ", "openrouter_key",
"https://probe.example/key?scope=current", "https://probe.example/key?scope=current",
} { } {
if !strings.Contains(contents, want) { if !strings.Contains(contents, want) {

View File

@@ -11,6 +11,14 @@ import (
"go.maximumdirect.net/promptkit" "go.maximumdirect.net/promptkit"
) )
func mountSettingsInputs(inputs []configkit.Input) []configkit.Input {
result := append([]configkit.Input(nil), inputs...)
for i := range result {
result[i].Destination = configkit.Keys("promptkit")
}
return result
}
type fakeLLMClient struct { type fakeLLMClient struct {
response *promptkit.GenerateResponse response *promptkit.GenerateResponse
responses []*promptkit.GenerateResponse responses []*promptkit.GenerateResponse
@@ -59,7 +67,7 @@ func newSimpleEngine(t *testing.T, content string, options ...promptkit.Option)
func simpleConfig(content string) promptkit.Config { func simpleConfig(content string) promptkit.Config {
return promptkit.Config{Sources: promptkit.ConfigurationSources{ return promptkit.Config{Sources: promptkit.ConfigurationSources{
Main: &configkit.Input{Source: configkit.Mapping("consumer test", map[string]any{ Main: &configkit.Input{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("consumer test", map[string]any{
"profiles": []any{map[string]any{"id": "profile", "profiles": []any{map[string]any{"id": "profile",
"endpoint": "http://example.test/v1", "model": "model", "endpoint": "http://example.test/v1", "model": "model",
}}, }},

View File

@@ -0,0 +1,97 @@
package promptkit_test
import (
"context"
"errors"
"net/http"
"net/url"
"strings"
"testing"
"testing/synctest"
"time"
"go.maximumdirect.net/promptkit"
)
type contextIdentityResponseBody struct {
ctx context.Context
cause error
}
func (b contextIdentityResponseBody) Read([]byte) (int, error) { <-b.ctx.Done(); return 0, b.cause }
func (contextIdentityResponseBody) Close() error { return nil }
func TestExecutionContextAndOriginalCauseIdentities(t *testing.T) {
for _, split := range []bool{false, true} {
for _, kind := range []string{"caller cancel", "caller deadline", "body operation deadline"} {
t.Run(map[bool]string{false: "run/", true: "prepared/"}[split]+kind, func(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
cause := errors.New("private original cause")
ctx, cancel := context.WithCancelCause(context.Background())
defer cancel(nil)
want := error(context.Canceled)
failure := promptkit.ExecutionFailureCanceled
if kind == "caller deadline" {
var stop context.CancelFunc
ctx, stop = context.WithTimeoutCause(context.Background(), time.Second, cause)
defer stop()
}
if kind != "caller cancel" {
want = context.DeadlineExceeded
failure = promptkit.ExecutionFailureDeadline
}
cfg := simpleConfig("message")
cfg.HTTPClient = &http.Client{Transport: roundTripFunc(func(r *http.Request) (*http.Response, error) {
if kind == "body operation deadline" {
return &http.Response{StatusCode: http.StatusOK, Body: contextIdentityResponseBody{ctx: r.Context(), cause: cause}}, nil
}
if kind == "caller cancel" {
cancel(cause)
} else {
<-ctx.Done()
}
return nil, context.Cause(ctx)
})}
engine, err := promptkit.NewEngine(cfg)
if err != nil {
t.Fatal(err)
}
var events []promptkit.ExecutionEvent
req := simpleGenerationRequest()
req.OperationTimeout = time.Minute
if kind == "body operation deadline" {
req.OperationTimeout = time.Second
}
req.Observer = func(e promptkit.ExecutionEvent) { events = append(events, e) }
var result *promptkit.RunResult
if split {
handle, prepareErr := engine.PrepareExecution(ctx, req)
if prepareErr != nil {
t.Fatal(prepareErr)
}
result, err = engine.RunPrepared(ctx, handle)
} else {
result, err = engine.Run(ctx, req)
}
for _, identity := range []error{want, cause, promptkit.ErrLLMGenerate, promptkit.ErrTransport} {
if !errors.Is(err, identity) {
t.Errorf("error=%v, want identity %v", err, identity)
}
}
if kind != "body operation deadline" {
var transportErr *url.Error
if !errors.As(err, &transportErr) {
t.Errorf("lost HTTP transport error: %v", err)
}
}
if result != nil || errors.Is(err, promptkit.ErrOperationTimeout) != (kind == "body operation deadline") || strings.Contains(err.Error(), cause.Error()) {
t.Fatalf("incorrect outcome or disclosure: result=%v error=%v", result, err)
}
if len(events) != 3 || events[1].Failure != failure || events[2].Kind != promptkit.ExecutionFinished || events[2].Failure != failure {
t.Fatalf("attempt/terminal classification = %+v", events)
}
})
})
}
}
}

View File

@@ -0,0 +1,141 @@
package promptkit_test
import (
"context"
"encoding/json"
"os"
"path/filepath"
"reflect"
"sync"
"testing"
"go.maximumdirect.net/promptkit"
)
func TestContinuationBranchesAcrossEngines(t *testing.T) {
ctx := context.Background()
firstClient := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: "first answer"}}
firstEngine := newSimpleEngine(t, "first question", promptkit.WithLLMClient(firstClient))
initial, err := firstEngine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true, APIKey: "synthetic-first-key"})
if err != nil {
t.Fatal(err)
}
original := initial.Exchange.Details()
secondClient := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: "follow-up answer"}}
secondEngine := newNamespacedContractEngine(t, secondClient)
scoped, err := secondEngine.WithNamespaces(promptkit.Namespaces{Prompt: "alpha", Profile: "alpha"})
if err != nil {
t.Fatal(err)
}
request := promptkit.RunRequest{PromptID: "shared", ProfileID: "shared", ContinueFrom: initial.Exchange, SessionID: "new-session", Validation: &promptkit.OutputContractInput{Format: promptkit.FormatJSON, ValidationMode: promptkit.ValidationJSON}, AppendedMessages: []promptkit.RenderedMessage{{Role: promptkit.RoleUser, Content: "branch one"}}}
branch, err := scoped.Run(ctx, request)
if err != nil {
t.Fatal(err)
}
request.AppendedMessages[0].Content = "branch two"
sibling, err := scoped.Run(ctx, request)
if err != nil {
t.Fatal(err)
}
next, err := scoped.Run(ctx, promptkit.RunRequest{PromptID: "shared", ProfileID: "shared", ContinueFrom: branch.Exchange})
if err != nil {
t.Fatal(err)
}
if branch.ParentRunID != initial.RunID || sibling.ParentRunID != initial.RunID || next.ParentRunID != branch.RunID || branch.PromptID != "alpha/shared" || branch.SelectedProfileID != "alpha/shared" || branch.SessionID != "new-session" {
t.Fatalf("ancestry or selections incorrect: %#v %#v %#v", branch, sibling, next)
}
if !reflect.DeepEqual(initial.Exchange.Details(), original) {
t.Fatal("branching changed source")
}
if branch.Validation.Status != promptkit.ValidationFailed || branch.ModelName != "alpha-model" {
t.Fatalf("follow-up settings: %#v", branch)
}
branchMessages := branch.Exchange.Details().Messages
siblingMessages := sibling.Exchange.Details().Messages
if len(branchMessages) != 5 || len(siblingMessages) != 5 || branchMessages[3].Content != "branch one" || siblingMessages[3].Content != "branch two" || len(next.Exchange.Details().Messages) != 7 {
t.Fatalf("branch histories = %#v / %#v", branchMessages, siblingMessages)
}
encoded, err := json.Marshal(branch.Exchange.Details())
if err != nil || !json.Valid(encoded) {
t.Fatalf("details JSON = %s, %v", encoded, err)
}
branchMessages[0].Content = "changed details"
branch.RawOutput = "changed result"
if initial.Exchange.Details().Messages[0].Content != "first question" || branch.Exchange.Details().Messages[4].Content != "follow-up answer" {
t.Fatal("caller mutation changed captured history")
}
for _, request := range secondClient.requests {
if request.APIKey == "synthetic-first-key" || request.Prompt.Messages[0].Content != "first question" || request.Prompt.Messages[1].Content != "first answer" {
t.Fatalf("unexpected follow-up request: %#v", request)
}
}
}
func TestContinuationDoesNotReloadCapturedInputs(t *testing.T) {
ctx := context.Background()
path := filepath.Join(t.TempDir(), "input.txt")
if err := os.WriteFile(path, []byte("original input"), 0600); err != nil {
t.Fatal(err)
}
client := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: ""}}
engine := newSimpleEngine(t, `{{input "source"}}`, promptkit.WithLLMClient(client))
initial, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", Inputs: map[string]promptkit.ArtifactRef{"source": promptkit.File(path)}, CaptureExchange: true})
if err != nil {
t.Fatal(err)
}
if err := os.WriteFile(path, []byte("new input"), 0600); err != nil {
t.Fatal(err)
}
request := promptkit.RunRequest{PromptID: "prompt", Inputs: map[string]promptkit.ArtifactRef{"source": promptkit.File(path)}, ContinueFrom: initial.Exchange}
prepared, err := engine.PrepareExecution(ctx, request)
if err != nil {
t.Fatal(err)
}
if err := os.Remove(path); err != nil {
t.Fatal(err)
}
result, err := engine.RunPrepared(ctx, prepared)
if err != nil {
t.Fatal(err)
}
want := []promptkit.RenderedMessage{{Role: promptkit.RoleUser, Content: "original input"}, {Role: promptkit.RoleAssistant, Content: ""}, {Role: promptkit.RoleUser, Content: "new input"}, {Role: promptkit.RoleAssistant, Content: ""}}
if !reflect.DeepEqual(result.Exchange.Details().Messages, want) {
t.Fatalf("history = %#v", result.Exchange.Details().Messages)
}
if result, err := engine.Run(ctx, request); err == nil || result != nil {
t.Fatalf("missing new input = %#v, %v", result, err)
}
}
func TestContinuationSourceSupportsConcurrentRuns(t *testing.T) {
client := &preparedRecordingClient{response: &promptkit.GenerateResponse{Content: "answer"}}
engine := newSimpleEngine(t, "question", promptkit.WithLLMClient(client))
initial, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true})
if err != nil {
t.Fatal(err)
}
const count = 8
var wg sync.WaitGroup
errors := make(chan error, count)
for i := 0; i < count; i++ {
wg.Add(1)
go func() {
defer wg.Done()
result, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt", ContinueFrom: initial.Exchange})
if err == nil && (result.Exchange == nil || result.ParentRunID != initial.RunID) {
t.Error("missing continuation ancestry")
}
errors <- err
}()
}
wg.Wait()
close(errors)
for err := range errors {
if err != nil {
t.Fatal(err)
}
}
if len(initial.Exchange.Details().Messages) != 2 {
t.Fatal("source changed")
}
}

View File

@@ -1,15 +1,23 @@
package promptkit package promptkit
import ( import (
"context"
"fmt" "fmt"
"reflect"
"unicode/utf8" "unicode/utf8"
"go.maximumdirect.net/promptkit/internal/configuration"
"go.maximumdirect.net/promptkit/internal/domain" "go.maximumdirect.net/promptkit/internal/domain"
"go.maximumdirect.net/promptkit/internal/jsonvalue" "go.maximumdirect.net/promptkit/internal/jsonvalue"
) )
func toDomainRunRequest(req RunRequest) (domain.RunRequest, error) { func toDomainRunRequest(req RunRequest) (domain.RunRequest, error) {
var history *domain.Exchange
if req.ContinueFrom != nil && req.ContinueFrom.internal == nil {
return domain.RunRequest{}, fmt.Errorf("continuation exchange is empty")
}
if req.ContinueFrom != nil {
history = req.ContinueFrom.internal
}
execution, err := toDomainExecutionTargetOverride(req.Execution) execution, err := toDomainExecutionTargetOverride(req.Execution)
if err != nil { if err != nil {
return domain.RunRequest{}, err return domain.RunRequest{}, err
@@ -23,7 +31,16 @@ func toDomainRunRequest(req RunRequest) (domain.RunRequest, error) {
callback := req.Observer callback := req.Observer
observer = func(event domain.ExecutionEvent) { callback(fromDomainExecutionEvent(event)) } observer = func(event domain.ExecutionEvent) { callback(fromDomainExecutionEvent(event)) }
} }
var onPrepared func(context.Context, *domain.PreparedRun)
if callback := req.OnPrepared; callback != nil {
onPrepared = func(ctx context.Context, prepared *domain.PreparedRun) {
callback(ctx, *fromDomainPreparedRun(prepared))
}
}
return domain.RunRequest{ return domain.RunRequest{
OperationTimeout: req.OperationTimeout,
OnPrepared: onPrepared,
CaptureExchange: req.CaptureExchange || req.ContinueFrom != nil,
PromptID: req.PromptID, PromptID: req.PromptID,
ProfileID: req.ProfileID, ProfileID: req.ProfileID,
SessionID: req.SessionID, SessionID: req.SessionID,
@@ -35,6 +52,7 @@ func toDomainRunRequest(req RunRequest) (domain.RunRequest, error) {
AppendedMessages: appendedMessages, AppendedMessages: appendedMessages,
TransientRetries: copyIntPtr(req.TransientRetries), TransientRetries: copyIntPtr(req.TransientRetries),
Observer: observer, Observer: observer,
ContinueFrom: history,
}, nil }, nil
} }
@@ -71,6 +89,7 @@ func fromDomainPreparedRun(prepared *domain.PreparedRun) *PreparedRun {
return nil return nil
} }
return &PreparedRun{ return &PreparedRun{
ParentRunID: prepared.ParentRunID,
PromptID: prepared.PromptID, PromptID: prepared.PromptID,
PromptHash: prepared.PromptHash, PromptHash: prepared.PromptHash,
SelectedProfileID: prepared.SelectedProfileID, SelectedProfileID: prepared.SelectedProfileID,
@@ -81,6 +100,7 @@ func fromDomainPreparedRun(prepared *domain.PreparedRun) *PreparedRun {
InputHashes: copyStringMap(prepared.InputHashes), InputHashes: copyStringMap(prepared.InputHashes),
SessionID: prepared.SessionID, SessionID: prepared.SessionID,
RenderedPromptHash: prepared.RenderedPromptHash, RenderedPromptHash: prepared.RenderedPromptHash,
MessageContentHash: prepared.MessageContentHash,
Messages: fromDomainRenderedMessages(prepared.Messages), Messages: fromDomainRenderedMessages(prepared.Messages),
StartTime: prepared.StartTime, StartTime: prepared.StartTime,
EndTime: prepared.EndTime, EndTime: prepared.EndTime,
@@ -92,7 +112,13 @@ func fromDomainRunResult(result *domain.RunResult) *RunResult {
if result == nil { if result == nil {
return nil return nil
} }
var exchange *Exchange
if result.Exchange != nil {
exchange = &Exchange{internal: result.Exchange}
}
return &RunResult{ return &RunResult{
Exchange: exchange,
ParentRunID: result.ParentRunID,
RunID: result.RunID, RunID: result.RunID,
Artifact: fromDomainArtifact(result.Artifact), Artifact: fromDomainArtifact(result.Artifact),
RawOutput: result.RawOutput, RawOutput: result.RawOutput,
@@ -229,37 +255,58 @@ func fromDomainExecutionTargetWithOwnedParams(target domain.ExecutionTarget, ext
} }
} }
func fromDomainProfileInspection(inspection *domain.ProfileInspection) *ProfileInspection { func fromProfileResource(v *configuration.ProfileResource) *ProfileInspection {
if inspection == nil { out := &ProfileInspection{ProfileID: v.ID, Backend: copyStringPtr(v.Backend), Endpoint: copyStringPtr(v.Endpoint), Model: copyStringPtr(v.Model), Temperature: copyFloat64Ptr(v.Temperature), TopP: copyFloat64Ptr(v.TopP), MaxTokens: copyIntPtr(v.MaxTokens), TimeoutSeconds: copyIntPtr(v.TimeoutSeconds), ServiceTier: copyStringPtr(v.ServiceTier), ReasoningEffort: copyStringPtr(v.ReasoningEffort), APIKeyEnv: copyStringPtr(v.APIKeyEnv), APIKeyRequired: copyBoolPtr(v.APIKeyRequired)}
return nil if v.ExtraParams != nil {
} data := copyAnyMap(v.ExtraParams)
return &ProfileInspection{ out.ExtraParams = &data
ProfileID: inspection.ProfileID,
EffectiveModelParams: fromDomainExecutionTarget(inspection.EffectiveModelParams),
APIKeyRequired: inspection.APIKeyRequired,
} }
return out
} }
func fromDomainPromptInspection(inspection *domain.PromptInspection) *PromptInspection { func fromBackendResource(v *configuration.BackendResource) *BackendInspection {
if inspection == nil { out := &BackendInspection{BackendID: v.ID, Endpoint: v.Endpoint, APIKeyEnv: copyStringPtr(v.APIKeyEnv), ConcurrencyLimit: copyIntPtr(v.ConcurrencyLimit), QueueCapacity: copyIntPtr(v.QueueCapacity)}
return nil if v.ExtraParams != nil {
data := copyAnyMap(v.ExtraParams)
out.ExtraParams = &data
} }
inputs := make([]PromptInputDefinition, len(inspection.Inputs)) if v.SmokeTests != nil {
for i, input := range inspection.Inputs { out.SmokeTests = &BackendSmokeTestConfig{DefaultProfile: copyStringPtr(v.SmokeTests.DefaultProfile)}
inputs[i] = PromptInputDefinition{ if v.SmokeTests.Reachability != nil {
Name: input.Name, out.SmokeTests.Reachability = &ReachabilityProbeConfig{Endpoint: copyStringPtr(v.SmokeTests.Reachability.Endpoint)}
Required: input.Required, }
ContentType: input.ContentType, if v.SmokeTests.Authentication != nil {
Description: input.Description, out.SmokeTests.Authentication = &AuthenticationProbeConfig{Kind: copyStringPtr(v.SmokeTests.Authentication.Kind), Endpoint: copyStringPtr(v.SmokeTests.Authentication.Endpoint)}
} }
} }
return &PromptInspection{ return out
PromptID: inspection.PromptID, }
PromptHash: inspection.PromptHash,
DefaultProfileID: inspection.DefaultProfileID, func fromPromptResource(v *configuration.PromptResource) *PromptInspection {
Inputs: inputs, out := &PromptInspection{PromptID: v.ID, PromptHash: v.Hash, DefaultProfileID: copyStringPtr(v.DefaultProfile), Description: copyStringPtr(v.Description), SessionID: copyStringPtr(v.SessionID), OutputContract: PromptOutputInspection{Format: OutputFormat(v.Output.Format), ValidationMode: ValidationMode(v.Output.ValidationMode), Schema: jsonvalue.CloneTrusted(v.Output.Schema), SchemaFile: copyStringPtr(v.Output.SchemaFile), SchemaPath: copyStringPtr(v.Output.SchemaPath), RepairAttempts: v.Output.RepairAttempts}}
OutputContract: fromDomainOutputContract(inspection.OutputContract), if v.Inputs != nil {
inputs := make([]PromptInputInspection, len(*v.Inputs))
for i, input := range *v.Inputs {
inputs[i] = PromptInputInspection{Name: input.Name, Required: copyBoolPtr(input.Required), ContentType: copyStringPtr(input.ContentType), Description: copyStringPtr(input.Description)}
}
out.Inputs = &inputs
} }
out.Messages = make([]PromptMessageInspection, len(v.Messages))
for i, message := range v.Messages {
out.Messages[i] = PromptMessageInspection{Role: message.Role, Content: message.Content}
if message.CacheControl != nil {
out.Messages[i].CacheControl = &PromptCacheControlInspection{Type: CacheControlType(message.CacheControl.Type), TTL: copyStringPtr(message.CacheControl.TTL)}
}
}
return out
}
func copyBoolPtr(src *bool) *bool {
if src == nil {
return nil
}
value := *src
return &value
} }
func fromDomainExecutionTargetPresence(presence domain.ExecutionTargetPresence) ExecutionTargetPresence { func fromDomainExecutionTargetPresence(presence domain.ExecutionTargetPresence) ExecutionTargetPresence {
@@ -271,20 +318,16 @@ func fromDomainExecutionTargetPresence(presence domain.ExecutionTargetPresence)
} }
} }
func toDomainOutputContractPtr(contract *OutputContract) *domain.OutputContract { func toDomainOutputContractPtr(contract *OutputContractInput) *domain.OutputContractInput {
if contract == nil { if contract == nil {
return nil return nil
} }
out := toDomainOutputContract(*contract) return &domain.OutputContractInput{
return &out
}
func toDomainOutputContract(contract OutputContract) domain.OutputContract {
return domain.OutputContract{
Format: domain.OutputFormat(contract.Format), Format: domain.OutputFormat(contract.Format),
ValidationMode: domain.ValidationMode(contract.ValidationMode), ValidationMode: domain.ValidationMode(contract.ValidationMode),
Schema: contract.Schema,
SchemaPath: contract.SchemaPath, SchemaPath: contract.SchemaPath,
RepairAttempts: contract.RepairAttempts, RepairAttempts: copyIntPtr(contract.RepairAttempts),
} }
} }
@@ -292,6 +335,7 @@ func fromDomainOutputContract(contract domain.OutputContract) OutputContract {
return OutputContract{ return OutputContract{
Format: OutputFormat(contract.Format), Format: OutputFormat(contract.Format),
ValidationMode: ValidationMode(contract.ValidationMode), ValidationMode: ValidationMode(contract.ValidationMode),
Schema: jsonvalue.CloneTrusted(contract.Schema),
SchemaPath: contract.SchemaPath, SchemaPath: contract.SchemaPath,
RepairAttempts: contract.RepairAttempts, RepairAttempts: contract.RepairAttempts,
} }
@@ -395,88 +439,10 @@ func copyAnyMap(src map[string]any) map[string]any {
if src == nil { if src == nil {
return nil return nil
} }
out := make(map[string]any, len(src)) return jsonvalue.CloneTrusted(src).(map[string]any)
for k, v := range src {
out[k] = copyAny(v)
}
return out
} }
func copyAny(value any) any { func copyAny(value any) any { return jsonvalue.CloneTrusted(value) }
if value == nil {
return nil
}
switch v := value.(type) {
case map[string]any:
return copyAnyMap(v)
case []any:
out := make([]any, len(v))
for i, item := range v {
out[i] = copyAny(item)
}
return out
case []string:
return copyStringSlice(v)
case []byte:
return copyBytes(v)
default:
return copyReflectValue(reflect.ValueOf(value)).Interface()
}
}
func copyReflectValue(value reflect.Value) reflect.Value {
if !value.IsValid() {
return value
}
switch value.Kind() {
case reflect.Interface:
if value.IsNil() {
return reflect.Zero(value.Type())
}
copied := copyReflectValue(value.Elem())
if copied.IsValid() && copied.Type().AssignableTo(value.Type()) {
return copied
}
out := reflect.New(value.Type()).Elem()
out.Set(copied)
return out
case reflect.Pointer:
if value.IsNil() {
return reflect.Zero(value.Type())
}
out := reflect.New(value.Type().Elem())
out.Elem().Set(copyReflectValue(value.Elem()))
return out
case reflect.Map:
if value.IsNil() {
return reflect.Zero(value.Type())
}
out := reflect.MakeMapWithSize(value.Type(), value.Len())
iter := value.MapRange()
for iter.Next() {
out.SetMapIndex(copyReflectValue(iter.Key()), copyReflectValue(iter.Value()))
}
return out
case reflect.Slice:
if value.IsNil() {
return reflect.Zero(value.Type())
}
out := reflect.MakeSlice(value.Type(), value.Len(), value.Cap())
for i := 0; i < value.Len(); i++ {
out.Index(i).Set(copyReflectValue(value.Index(i)))
}
return out
case reflect.Array:
out := reflect.New(value.Type()).Elem()
for i := 0; i < value.Len(); i++ {
out.Index(i).Set(copyReflectValue(value.Index(i)))
}
return out
default:
return value
}
}
func copyStringSlice(src []string) []string { func copyStringSlice(src []string) []string {
if src == nil { if src == nil {

82
coordinate.go Normal file
View File

@@ -0,0 +1,82 @@
package promptkit
import (
"context"
"fmt"
"go.maximumdirect.net/promptkit/internal/domain"
"go.maximumdirect.net/promptkit/internal/usecase"
)
// ConversationTurn is delivered before each requested continuation. Result is
// nil on the initial callback; RemainingTurns counts additional runs allowed.
type ConversationTurn struct {
Exchange *Exchange
Result *RunResult
RemainingTurns int
}
// CoordinateRequest selects a starting exchange, turn budget and callback.
// Observer selects continuation runs only, independently of callback calls.
type CoordinateRequest struct {
Start *Exchange
AdditionalTurns int
Next func(context.Context, ConversationTurn) (*RunRequest, error)
Observer ExecutionObserver
}
// CoordinationOutcome distinguishes caller-directed stopping from exhaustion.
type CoordinationOutcome string
const (
// CoordinationStopped means the callback chose not to continue.
CoordinationStopped CoordinationOutcome = "stopped"
// CoordinationExhausted means another run was requested without budget.
CoordinationExhausted CoordinationOutcome = "exhausted"
)
// CoordinationResult contains the normal outcome or a response withheld by
// cancellation before delivery to the callback.
type CoordinationResult struct {
Outcome CoordinationOutcome `json:"outcome,omitempty"`
Undelivered *RunResult `json:"undelivered,omitempty"`
}
// Coordinate advances one branch from Start using ordinary runs. Next is called
// initially even for a zero budget and receives each completed result once,
// including the final permitted response. The initial exchange is not itself a
// newly delivered result. Each started continuation consumes one turn; retries
// and structural repairs do not. Callback work runs after admission is released.
// A nil request stops; requesting a run without remaining budget exhausts it.
// Callback and execution errors retain their identities. Cancellation observed
// before delivery returns the single completed response as Undelivered;
// cancellation after callback invocation does not replay that response. Context
// cancellation takes precedence over a callback's returned action or error.
// Start must be a captured exchange, Next non-nil, and AdditionalTurns
// nonnegative; invalid values match ErrInvalidRequest. A nil engine returns
// ErrInvalidConfig. The selected request is copied and its ContinueFrom is
// replaced by the coordinator's current exchange. The request observer runs
// before the coordinator observer for each continuation.
func (e *Engine) Coordinate(ctx context.Context, req CoordinateRequest) (CoordinationResult, error) {
if e == nil || e.internal == nil {
return CoordinationResult{}, fmt.Errorf("%w: engine is nil", ErrInvalidConfig)
}
if req.Start == nil || req.Start.internal == nil || req.Next == nil || req.AdditionalTurns < 0 {
return CoordinationResult{}, fmt.Errorf("%w: invalid coordination request", ErrInvalidRequest)
}
outcome, err := usecase.Coordinate(ctx, req.Start.internal, req.AdditionalTurns,
func(ctx context.Context, turn usecase.ConversationTurn[RunResult]) (*RunRequest, error) {
return req.Next(ctx, ConversationTurn{Exchange: &Exchange{internal: turn.Exchange}, Result: turn.Result, RemainingTurns: turn.RemainingTurns})
},
func(ctx context.Context, current *domain.Exchange, selected *RunRequest) (*RunResult, *domain.Exchange, error) {
request := *selected
request.ContinueFrom = &Exchange{internal: current}
request.Observer = ComposeObservers(request.Observer, req.Observer)
result, runErr := e.Run(ctx, request)
if runErr != nil {
return nil, nil, runErr
}
return result, result.Exchange.internal, nil
})
return CoordinationResult{Outcome: CoordinationOutcome(outcome.Outcome), Undelivered: outcome.Undelivered}, err
}

View File

@@ -0,0 +1,61 @@
package promptkit_test
import (
"context"
"errors"
"testing"
"go.maximumdirect.net/promptkit"
)
func TestCoordinatePublicBoundary(t *testing.T) {
ctx := context.Background()
engine := newSimpleEngine(t, "question", promptkit.WithLLMClient(&fakeLLMClient{response: &promptkit.GenerateResponse{Content: "answer"}}))
initial, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true})
if err != nil {
t.Fatal(err)
}
for _, request := range []promptkit.CoordinateRequest{{}, {Start: &promptkit.Exchange{}, Next: func(context.Context, promptkit.ConversationTurn) (*promptkit.RunRequest, error) { return nil, nil }}, {Start: initial.Exchange, AdditionalTurns: -1, Next: func(context.Context, promptkit.ConversationTurn) (*promptkit.RunRequest, error) { return nil, nil }}} {
if _, err := engine.Coordinate(ctx, request); !errors.Is(err, promptkit.ErrInvalidRequest) {
t.Fatalf("invalid request error = %v", err)
}
}
if _, err := (*promptkit.Engine)(nil).Coordinate(ctx, promptkit.CoordinateRequest{}); !errors.Is(err, promptkit.ErrInvalidConfig) {
t.Fatalf("nil engine error = %v", err)
}
callbacks := 0
selected := &promptkit.RunRequest{PromptID: "prompt", ContinueFrom: &promptkit.Exchange{}}
result, err := engine.Coordinate(ctx, promptkit.CoordinateRequest{Start: initial.Exchange, AdditionalTurns: 2, Next: func(_ context.Context, turn promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
callbacks++
if callbacks == 1 {
if turn.Result != nil || turn.RemainingTurns != 2 {
t.Fatalf("initial turn = %#v", turn)
}
return selected, nil
}
if turn.Result == nil || turn.Exchange.Details().ParentRunID == "" {
t.Fatalf("delivered turn = %#v", turn)
}
if callbacks == 2 {
turn.Result.Exchange = nil
*turn.Exchange = promptkit.Exchange{}
return selected, nil
}
if turn.Result.ParentRunID == initial.RunID {
t.Fatal("callback mutation redirected branch")
}
return nil, nil
}})
if err != nil || result.Outcome != promptkit.CoordinationStopped || callbacks != 3 || selected.ContinueFrom == nil {
t.Fatalf("result=%#v error=%v callbacks=%d", result, err, callbacks)
}
stop, err := engine.Coordinate(ctx, promptkit.CoordinateRequest{Start: initial.Exchange, Next: func(context.Context, promptkit.ConversationTurn) (*promptkit.RunRequest, error) { return nil, nil }})
if err != nil || stop.Outcome != promptkit.CoordinationStopped {
t.Fatalf("zero budget = %#v, %v", stop, err)
}
sentinel := errors.New("callback")
_, err = engine.Coordinate(ctx, promptkit.CoordinateRequest{Start: initial.Exchange, Next: func(context.Context, promptkit.ConversationTurn) (*promptkit.RunRequest, error) { return nil, sentinel }})
if !errors.Is(err, sentinel) {
t.Fatalf("callback error = %v", err)
}
}

View File

@@ -0,0 +1,86 @@
package promptkit_test
import (
"context"
"errors"
"testing"
"testing/synctest"
"time"
"go.maximumdirect.net/promptkit"
)
func TestCoordinateUsesIndependentOperationBudgets(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
engine := newBackendCapacityEngine(t, observationClientFunc(func(_ context.Context, req promptkit.GenerateRequest) (*promptkit.GenerateResponse, error) {
if req.Prompt.SessionID == "continuation" {
time.Sleep(2 * time.Second)
}
return &promptkit.GenerateResponse{Content: "answer", Usage: promptkit.TokenUsage{PromptTokens: 2}}, nil
}), 1, capacityInt(0), nil)
parent := context.Background()
initial, err := engine.Run(parent, promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true})
if err != nil {
t.Fatal(err)
}
var group promptkit.ExecutionGroup
var child context.Context
var details []promptkit.PreparedRun
var requestEvents, coordinatorEvents []promptkit.ExecutionEvent
start := time.Now()
result, err := engine.Coordinate(parent, promptkit.CoordinateRequest{Start: initial.Exchange, AdditionalTurns: 2, Observer: promptkit.ComposeObservers(group.Observe, func(e promptkit.ExecutionEvent) { coordinatorEvents = append(coordinatorEvents, e) }), Next: func(ctx context.Context, turn promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
if turn.Result != nil {
if ctx.Err() != nil || child == nil || !errors.Is(child.Err(), context.Canceled) {
t.Fatal("child was not cleaned up before callback")
}
if _, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", OperationTimeout: time.Second}); err != nil {
t.Fatalf("assessment: %v", err)
}
}
if turn.RemainingTurns == 0 {
return nil, nil
}
return &promptkit.RunRequest{PromptID: "prompt", SessionID: "continuation", OperationTimeout: 3 * time.Second, Observer: func(e promptkit.ExecutionEvent) { requestEvents = append(requestEvents, e) }, OnPrepared: func(ctx context.Context, prepared promptkit.PreparedRun) {
child = ctx
details = append(details, prepared)
}}, nil
}})
if err != nil || result.Outcome != promptkit.CoordinationStopped || time.Since(start) != 4*time.Second || len(details) != 2 || len(details[0].Messages) != 3 || len(details[1].Messages) != 5 {
t.Fatalf("result=%+v error=%v details=%v elapsed=%v", result, err, len(details), time.Since(start))
}
summary := group.Snapshot()
if len(requestEvents) != 6 || len(coordinatorEvents) != 6 || len(summary.Members) != 2 || summary.Attempts != 2 || summary.Usage.PromptTokens.Value != 4 || !summary.TerminalCoverageComplete {
t.Fatalf("summary=%+v request=%d coordinator=%d", summary, len(requestEvents), len(coordinatorEvents))
}
})
}
func TestCoordinateTimeoutLeavesParentAndSiblingUsable(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
engine := newBackendCapacityEngine(t, observationClientFunc(func(ctx context.Context, req promptkit.GenerateRequest) (*promptkit.GenerateResponse, error) {
if req.Prompt.SessionID == "slow" {
<-ctx.Done()
return nil, ctx.Err()
}
return &promptkit.GenerateResponse{Content: "answer"}, nil
}), 1, capacityInt(0), nil)
parent := context.Background()
initial, err := engine.Run(parent, promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true})
if err != nil {
t.Fatal(err)
}
var details promptkit.PreparedRun
calls := 0
result, err := engine.Coordinate(parent, promptkit.CoordinateRequest{Start: initial.Exchange, AdditionalTurns: 1, Next: func(context.Context, promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
calls++
return &promptkit.RunRequest{PromptID: "prompt", SessionID: "slow", OperationTimeout: time.Second, OnPrepared: func(_ context.Context, p promptkit.PreparedRun) { details = p }}, nil
}})
if !errors.Is(err, promptkit.ErrOperationTimeout) || !errors.Is(err, context.DeadlineExceeded) || calls != 1 || result.Outcome != "" || result.Undelivered != nil || details.ParentRunID != initial.RunID || parent.Err() != nil {
t.Fatalf("result=%+v error=%v callbacks=%d", result, err, calls)
}
sibling, err := engine.Run(parent, promptkit.RunRequest{PromptID: "prompt", ContinueFrom: initial.Exchange})
if err != nil || sibling.Exchange == nil {
t.Fatalf("sibling=%v error=%v", sibling, err)
}
})
}

View File

@@ -0,0 +1,170 @@
package promptkit_test
import (
"context"
"errors"
"net/http"
"testing"
"time"
"go.maximumdirect.net/promptkit"
)
func TestCoordinateReleasesCapacityBeforeCallback(t *testing.T) {
client := &preparedRecordingClient{responses: []*promptkit.GenerateResponse{{Content: "initial"}, {Content: ""}, {Content: " "}, {Content: "validator"}}}
engine := newBackendCapacityEngine(t, client, 1, capacityInt(0), nil)
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
initial, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true})
if err != nil {
t.Fatal(err)
}
calls := 0
result, err := engine.Coordinate(ctx, promptkit.CoordinateRequest{Start: initial.Exchange, AdditionalTurns: 1, Next: func(ctx context.Context, turn promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
calls++
if turn.Result == nil {
return &promptkit.RunRequest{PromptID: "prompt", Validation: &promptkit.OutputContractInput{Format: promptkit.FormatText, ValidationMode: promptkit.ValidationBasic, RepairAttempts: new(1)}}, nil
}
if turn.Result.Validation.Status != promptkit.ValidationFailed || turn.Result.Validation.RepairAttempts != 1 || turn.RemainingTurns != 0 {
t.Fatalf("delivered turn = %#v", turn)
}
_, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt"})
return nil, err
}})
if err != nil || result.Outcome != promptkit.CoordinationStopped || calls != 2 || len(client.snapshot()) != 4 {
t.Fatalf("outcome=%#v error=%v callbacks=%d calls=%d", result, err, calls, len(client.snapshot()))
}
}
func TestCoordinatePreservesCapacityAndProviderErrors(t *testing.T) {
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
sourceEngine := newSimpleEngine(t, "source", promptkit.WithLLMClient(&fakeLLMClient{response: &promptkit.GenerateResponse{Content: "answer"}}))
initial, err := sourceEngine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true})
if err != nil {
t.Fatal(err)
}
release := make(chan struct{})
client := newCapacityGateClient(release, 1)
engine := newBackendCapacityEngine(t, client, 1, capacityInt(0), nil)
done := make(chan capacityRunResult, 1)
go runCapacityRequest(engine, ctx, promptkit.RunRequest{PromptID: "prompt"}, done)
select {
case <-client.started:
case <-ctx.Done():
t.Fatal(ctx.Err())
}
request := promptkit.CoordinateRequest{Start: initial.Exchange, AdditionalTurns: 1, Next: func(context.Context, promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
return &promptkit.RunRequest{PromptID: "prompt"}, nil
}}
_, err = engine.Coordinate(ctx, request)
var capacity *promptkit.CapacityError
if !errors.As(err, &capacity) || capacity.BackendID != "limited" {
t.Fatalf("capacity error = %v", err)
}
close(release)
if finished := awaitCapacityRun(t, done); finished.err != nil {
t.Fatal(finished.err)
}
providerEngine := newBuiltInGenerationErrorEngine(t, http.StatusUnprocessableEntity, `{"error":{"message":"synthetic"}}`)
_, err = providerEngine.Coordinate(ctx, request)
var provider *promptkit.GenerationError
if !errors.As(err, &provider) || provider.StatusCode() != http.StatusUnprocessableEntity {
t.Fatalf("provider error = %v", err)
}
}
func TestCoordinateRetriesAndRepairsDeliverOnce(t *testing.T) {
initialEngine := newSimpleEngine(t, "source", promptkit.WithLLMClient(&fakeLLMClient{response: &promptkit.GenerateResponse{Content: "answer"}}))
initial, err := initialEngine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true})
if err != nil {
t.Fatal(err)
}
responses := []*http.Response{observedHTTPResponse(503, `{}`), observedHTTPResponse(200, `{"choices":[{"message":{"content":"invalid"}}]}`), observedHTTPResponse(200, `{"choices":[{"message":{"content":"{\"ok\":true}"}}]}`)}
calls := 0
one := 1
config := simpleConfig("follow-up")
config.TransientRetries = &one
config.HTTPClient = &http.Client{Transport: roundTripFunc(func(*http.Request) (*http.Response, error) {
response := responses[calls]
calls++
return response, nil
})}
engine, err := promptkit.NewEngine(config)
if err != nil {
t.Fatal(err)
}
callbacks := 0
result, err := engine.Coordinate(context.Background(), promptkit.CoordinateRequest{Start: initial.Exchange, AdditionalTurns: 1, Next: func(_ context.Context, turn promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
callbacks++
if turn.Result == nil {
return &promptkit.RunRequest{PromptID: "prompt", Validation: &promptkit.OutputContractInput{Format: promptkit.FormatJSON, ValidationMode: promptkit.ValidationJSON, RepairAttempts: new(1)}}, nil
}
if !turn.Result.Validation.IsValid || turn.RemainingTurns != 0 {
t.Fatalf("turn = %#v", turn)
}
return nil, nil
}})
if err != nil || result.Outcome != promptkit.CoordinationStopped || callbacks != 2 || calls != 3 {
t.Fatalf("result=%#v error=%v callbacks=%d attempts=%d", result, err, callbacks, calls)
}
}
func TestCoordinateCancellationDelivery(t *testing.T) {
for _, cancelInside := range []bool{false, true} {
t.Run(map[bool]string{false: "before delivery", true: "inside callback"}[cancelInside], func(t *testing.T) {
client := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: "answer"}}
engine := newSimpleEngine(t, "question", promptkit.WithLLMClient(client))
initial, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true})
if err != nil {
t.Fatal(err)
}
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
calls := 0
result, err := engine.Coordinate(ctx, promptkit.CoordinateRequest{Start: initial.Exchange, AdditionalTurns: 1, Next: func(_ context.Context, turn promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
calls++
if turn.Result != nil {
cancel()
return nil, nil
}
return &promptkit.RunRequest{PromptID: "prompt", Observer: func(event promptkit.ExecutionEvent) {
if !cancelInside && event.Kind == promptkit.ExecutionFinished {
cancel()
}
}}, nil
}})
if !errors.Is(err, context.Canceled) || result.Outcome != "" || (result.Undelivered != nil) == cancelInside {
t.Fatalf("result=%#v error=%v callbacks=%d", result, err, calls)
}
if !cancelInside && (calls != 1 || result.Undelivered.Exchange == nil) {
t.Fatalf("undelivered result=%#v calls=%d", result, calls)
}
if cancelInside && calls != 2 {
t.Fatalf("callbacks=%d", calls)
}
})
}
}
func TestCoordinateFailureLeavesSiblingUsable(t *testing.T) {
client := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: "answer"}}
engine := newSimpleEngine(t, "question", promptkit.WithLLMClient(client))
ctx := context.Background()
initial, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true})
if err != nil {
t.Fatal(err)
}
sentinel := errors.New("provider")
client.err = sentinel
result, err := engine.Coordinate(ctx, promptkit.CoordinateRequest{Start: initial.Exchange, AdditionalTurns: 1, Next: func(context.Context, promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
return &promptkit.RunRequest{PromptID: "prompt"}, nil
}})
if !errors.Is(err, sentinel) || !errors.Is(err, promptkit.ErrLLMGenerate) || result.Outcome != "" || result.Undelivered != nil {
t.Fatalf("failure=%#v, %v", result, err)
}
client.err = nil
if sibling, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", ContinueFrom: initial.Exchange}); err != nil || sibling.Exchange == nil {
t.Fatalf("sibling=%#v, %v", sibling, err)
}
}

39
doc.go
View File

@@ -1,29 +1,33 @@
// Package promptkit provides an embeddable engine for preparing and executing // Package promptkit provides an embeddable engine for preparing and executing
// prompt-defined LLM workflows. // prompt-defined LLM workflows.
// //
// Applications construct an [Engine] with [NewEngine], select ConfigKit // Applications construct an [Engine] with [NewEngine] or register [Configure]
// sources for prompts, profiles, and backends, and // during their own ConfigKit load and pass its result to [NewFromConfig], then
// call [Engine.InspectConfiguration], [Engine.InspectPrompt], // call [Engine.InspectConfiguration], [Engine.InspectPrompt],
// [Engine.InspectRequirements], [Engine.InspectProfile], [Engine.ListProfiles], // [Engine.InspectRequirements], [Engine.InspectProfile], [Engine.InspectBackend],
// [Engine.ListPrompts], [Engine.Prepare], [Engine.PrepareExecution], // [Engine.ListProfiles], [Engine.ListPrompts], [Engine.Prepare],
// [Engine.Run], or [Engine.RunPrepared]. Concrete registries, repositories, // [Engine.PrepareExecution], [Engine.Run], [Engine.RunPrepared], or
// [Engine.Coordinate]. Captured [Exchange] values support independent
// continuations, and [ExecutionGroup] collects explicitly observed runs.
// Concrete registries, repositories,
// validators, and the built-in OpenAI-compatible client remain internal // validators, and the built-in OpenAI-compatible client remain internal
// implementation details. // implementation details.
// //
// # Concurrency and ownership // # Concurrency and ownership
// //
// An Engine supports concurrent InspectConfiguration, InspectPrompt, // An Engine supports concurrent InspectConfiguration, InspectPrompt,
// InspectRequirements, InspectProfile, ListProfiles, ListPrompts, Prepare, // InspectRequirements, InspectProfile, InspectBackend, ListProfiles, ListPrompts,
// PrepareExecution, Run, and RunPrepared calls. Engine-local backend policies // Prepare, PrepareExecution, Run, RunPrepared, and Coordinate calls. Engine-local backend policies
// bound admitted Run and RunPrepared calls and model generations where // bound admitted Run and RunPrepared calls and model generations where
// configured, while different backend pools and unlimited backends continue // configured, while different backend pools and unlimited backends continue
// independently. An injected [LLMClient] or [ArtifactReader] can therefore // independently. An injected [LLMClient] or [ArtifactReader] can therefore
// still receive concurrent calls and must be safe for that use. // still receive concurrent calls and must be safe for that use.
// //
// ConfigKit captures the selected configuration when NewEngine constructs an // An Engine captures selected configuration resources during construction.
// Engine. Prepare, PrepareExecution, and Run copy request maps, slices, pointer // Prepare, PrepareExecution, and Run copy request maps, slices, pointer
// values, and JSON-compatible extra parameters before using them. InspectPrompt, // values, and JSON-compatible extra parameters before using them. InspectPrompt,
// InspectRequirements, and InspectProfile return copied inspection values. // InspectRequirements, InspectProfile, and InspectBackend return copied
// inspection values.
// ListProfiles and ListPrompts return caller-owned identity slices. // ListProfiles and ListPrompts return caller-owned identity slices.
// Returned values and values passed to extension interfaces are likewise // Returned values and values passed to extension interfaces are likewise
// isolated from engine state. // isolated from engine state.
@@ -42,7 +46,7 @@
// that enforces their filesystem, authorization, and size policies. // that enforces their filesystem, authorization, and size policies.
// //
// Rendered messages, input and output [Artifact] bodies, [RunResult.RawOutput], // Rendered messages, input and output [Artifact] bodies, [RunResult.RawOutput],
// and [ValidationResult.Errors] may contain sensitive data. Credential // [ValidationResult.Errors], and [ExchangeDetails.Messages] may contain sensitive data. Credential
// exclusion and redaction do not sanitize those values. Applications and // exclusion and redaction do not sanitize those values. Applications and
// injected collaborators are responsible for access control, retention, // injected collaborators are responsible for access control, retention,
// logging, and secret handling appropriate to their data. // logging, and secret handling appropriate to their data.
@@ -56,17 +60,22 @@
// [GenerateResponse], [ExecutionTargetPresence], and the string value types // [GenerateResponse], [ExecutionTargetPresence], and the string value types
// used by those values. // used by those values.
// //
// Construction, inspection, handle, and error values, including [Config], // Resource inspection results support direct JSON and YAML encoding.
// [PromptInspection], [ProfileInspection], [BackendInspection], and their nested
// values define field names, nesting, and omission of absent optional fields.
// Formatting belongs to the selected encoder.
//
// Other construction, inspection, handle, and error values, including [Config],
// [ConfigurationSources], [ConfigurationInspection], [BackendConfig], // [ConfigurationSources], [ConfigurationInspection], [BackendConfig],
// [ProfileConfig], [RunRequest], // [ProfileConfig], [RunRequest],
// [ArtifactRef], [ExecutionTargetOverride], [ProfileInspection], // [ArtifactRef], [ExecutionTargetOverride],
// [PromptIdentity], [PromptInputDefinition], [PromptInspection], // [PromptIdentity], [PromptInputDefinition],
// [RequirementInspectionRequest], [RequirementInspection], [PreparedExecution], // [RequirementInspectionRequest], [RequirementInspection], [PreparedExecution],
// [CapacityError], [GenerationError], [RetryExhaustedError], and // [CapacityError], [GenerationError], [RetryExhaustedError], and
// [ExecutionEvent], do not have stable JSON // [ExecutionEvent], do not have stable JSON
// representations. Direct API keys are nevertheless excluded from JSON for // representations. Direct API keys are nevertheless excluded from JSON for
// every public value. // every public value.
//
// Provider-derived [GenerationError] accessor values are untrusted and can // Provider-derived [GenerationError] accessor values are untrusted and can
// contain sensitive request or schema fragments. Applications must apply their // contain sensitive request or schema fragments. Applications must apply their
// own disclosure policy before logging, displaying, or returning them. // own disclosure policy before logging, displaying, or returning them.

View File

@@ -0,0 +1,63 @@
# 0002: Share Configuration Data Types Across The Public Boundary
## Status
Accepted.
## Date
2026-11-21
## Context
PromptKit is a Go library that applications use to configure and run prompts.
Applications use its public package; the code that loads settings and runs
prompts lives in internal packages. This separation normally lets the internal
code change without changing the application's interface.
The planned configuration API lets applications load settings once and pass them to
PromptKit. Keeping separate public and internal definitions of those same
settings would require duplicate declarations and code to copy each field back
and forth. That adds maintenance work and opportunities to miss a field without
changing what the settings mean.
## Decision
Define the public configuration settings and their nested data types once in a
small, data-only internal package. Expose them through type aliases in the public
`promptkit` package. An alias gives the same Go type a public name; consumers
continue to use names such as `promptkit.Settings` without importing an internal
package. The internal configuration code uses the same definitions directly.
This is a narrow exception to the rule against exposing internal types. These
settings are intentional public contracts, including their fields and
configuration tags. Their location does not make them private implementation
details that can be changed independently of the public API.
The rest of the public boundary stays in place. Loading, validation, and
execution logic remain internal. Source-building helpers, execution requests,
runtime objects, and inspection reports keep separate representations where
they serve different purposes. Internal packages still do not import the public
package. Sharing declarations does not remove the copies needed to keep a
running engine independent of later application changes.
## Alternatives Considered
- Keep separate public and internal settings. This allows independent changes,
but provides little benefit when both describe the same configuration and
must stay synchronized.
- Put the shared types in a new public settings package. This also removes
duplication, but adds a public package when the existing `promptkit` package
already provides a suitable consumer interface.
## Consequences
Configuration fields have one declaration, and registration and resource
capture no longer need conversions between equivalent settings types. The
primary benefit is simpler maintenance; performance improvements are secondary.
Conversion into execution-ready objects and necessary ownership copies remain.
This decision leaves the [ConfigKit ownership boundary](0001-configuration-boundary.md)
unchanged. The [shared-loading guide](../consumers/pkg-promptkit.md#reuse-an-application-owned-load)
describes the implemented consumer workflow; the
[internal overview](../internal/overview.md) describes its implementation.

View File

@@ -1,15 +1,43 @@
# Configuration And API Migration # Configuration And API Migration
This breaking pre-v1 minor migration adopts Promptkit's vanity module path, This guide covers the current configuration and repair-input migration, as well
changes the v0.16-era keyed resources to identified lists, adopts ConfigKit v1, as earlier module-path and keyed-resource migrations.
and removes prompt-version labels. Promptkit installs a default-on bundled OpenRouter catalog; applications choose configuration
Promptkit does not install a provider catalog; applications choose configuration
inputs explicitly, and execution selects one prompt by its exact ID. This guide inputs explicitly, and execution selects one prompt by its exact ID. This guide
explains the shape changes and decisions an application must make; it is not an explains the shape changes and decisions an application must make; it is not an
automatic migration tool. The [framework format reference](../formats.md) owns automatic migration tool. The [framework format reference](../formats.md) owns
the current configuration contract, and the root package GoDoc owns exact API the current configuration contract, and the root package GoDoc owns exact API
signatures and errors. signatures and errors.
## Move Settings Under Promptkit
Place complete-document resources under `promptkit`, including `schema_root`,
`backends`, `profiles`, `prompts`, and `collections`. Old root-level resource
keys are application data and are not loaded as Promptkit resources; selecting
an unmigrated resource can therefore fail with a not-found error. IDs and
references do not acquire a `promptkit` prefix. `SettingsInput` mounts a
wrapperless fragment at `/promptkit`, and `CollectionInput` mounts one at
`/promptkit/collections/<namespace>`. Discovery helpers still return root-mounted
inputs; set destinations explicitly for discovered fragments. Standalone
id-first files use native sequence-item placement. The [format reference](../formats.md)
describes extensions and strict nested objects.
Applications can instead embed `promptkit.Settings`, call `Configure` on their
ConfigKit definition, load once, and pass the whole result to `NewFromConfig`.
The [preparation example](../../examples/go-library/prepare/main.go) shows the
workflow. Select `BundledOpenRouterUnderlay` before an application-owned load;
the default bundle option applies only to `NewEngine`. The application owns
version policy; Promptkit no longer validates resource-version fields.
Replace `RunRequest.Validation` values of type `*OutputContract` with
`*OutputContractInput` and use a pointer for `RepairAttempts`. Nil or null
selects zero for `none` and one for `basic`, `json`, and `json_schema`;
explicit zero disables correction. Runtime replacements resolve independently
of configured budgets. Inspection reports resolved integer budgets, including
zero. A failed validation can cause an additional generation and associated
latency and usage; a correction is not guaranteed. See the
[output contract](../formats.md#output-contract).
## Update Go Imports ## Update Go Imports
Promptkit's module and public package import path is now Promptkit's module and public package import path is now
@@ -37,8 +65,8 @@ The configuration and API adaptations below are also part of this migration.
## Select And Compose The Configuration You Need ## Select And Compose The Configuration You Need
Supply each backend, profile, prompt, schema, and collection source that the Supply each backend, profile, prompt, schema, and collection source that the
application intends to use. Promptkit does not discover directories or install application intends to use. `NewEngine` selects the bundled OpenRouter underlay
provider definitions implicitly. An application may select an explicit by default; callers explicitly select other sources. An application may select an explicit
collection input with `CollectionInput`: collection input with `CollectionInput`:
```go ```go
@@ -64,11 +92,10 @@ Promptkit does not rewrite references when a source is mounted. `extends`
selects a sibling ID in the same list, not a cross-collection reference. selects a sibling ID in the same list, not a cross-collection reference.
Review collisions and all references when moving resources into collections. Review collisions and all references when moving resources into collections.
The optional source-root `version` header now accepts `2` or `"2"`; headerless A collection's `release_version` remains an optional label. Prompt-version
sources remain valid. A collection's `release_version` remains an optional selection has been removed: choose one definition per ID. Application-specific
label. Prompt-version labels have been removed: choose one definition per ID version data may be retained at supported extension points, but is not used by
and store any application-specific release label outside Promptkit. In a Promptkit for resource selection.
standalone imported prompt item, root `version` is a source header, not a label.
For example, change a version-keyed definition: For example, change a version-keyed definition:
@@ -84,7 +111,7 @@ prompts:
validation_mode: none validation_mode: none
``` ```
to one identified definition: to one identified wrapperless fragment, mounted with `SettingsInput`:
```yaml ```yaml
prompts: prompts:
@@ -98,7 +125,8 @@ prompts:
``` ```
The same conversion applies to backends and profiles. Sibling inheritance uses The same conversion applies to backends and profiles. Sibling inheritance uses
identity references, and a later same-ID record replaces the whole lower record: identity references, and a later same-ID record replaces the whole lower record.
This is another wrapperless fragment to mount with `SettingsInput`:
```yaml ```yaml
profiles: profiles:
@@ -113,11 +141,11 @@ profiles:
To customize a lower resource's fields, add a distinct inherited variant as To customize a lower resource's fields, add a distinct inherited variant as
above; when replacing the same ID, supply the complete intended record. Generic above; when replacing the same ID, supply the complete intended record. Generic
append, replace, imports, and inheritance are described by the append, replace, imports, and inheritance are described by the
[ConfigKit v1 configuration contract](https://forge.maximumdirect.net/base-layer/configkit/src/tag/v1.0.0/docs/config.md). [ConfigKit v1 configuration contract](https://forge.maximumdirect.net/base-layer/configkit/src/tag/v1.4.0/docs/config.md).
Source-root controls are now `config:`, not `configkit:`. Imports are optional Source-root controls are now `config:`, not `configkit:`. Imports are optional
unless the selected import declares `required: true`. unless the selected import declares `required: true`.
Update ConfigKit imports to `go.maximumdirect.net/configkit` at v1.0.0: its Go Update ConfigKit imports to `go.maximumdirect.net/configkit` at v1.4.0: its Go
types are incompatible with the old module path. Promptkit now requires Go types are incompatible with the old module path. Promptkit now requires Go
1.27.0 or newer. Source builders take ordered slices with explicit IDs: 1.27.0 or newer. Source builders take ordered slices with explicit IDs:
@@ -133,14 +161,18 @@ source := promptkit.ProfileSource("application", []promptkit.ProfileConfig{
Replace former maps passed to `BackendSource` in the same way, setting each Replace former maps passed to `BackendSource` in the same way, setting each
`BackendConfig.ID`. Duplicates are retained for loading to reject. Resource `BackendConfig.ID`. Duplicates are retained for loading to reject. Resource
inspection paths now address list indexes, for example inspection paths now address list indexes, for example
`configkit.Keys("profiles").Index(0).Key("model")`. Configuration inspection `configkit.Keys("promptkit", "profiles").Index(0).Key("model")`. Configuration inspection
and diffs expose composed values and provenance, while `InspectProfile` and and diffs expose resolved values, recorded normalization and provenance.
other interpreted views expose Promptkit normalization and defaults; the views `InspectProfile` reports the selected resolved profile without backend completion.
are not promised to be identical. Use `Prepare` or `PreparedExecution.Details` for completed request settings,
including backend defaults and request overrides; see the
[resource inspection guidance](pkg-promptkit.md#inspect-a-profile-before-prompt-work).
Remove `PromptVersion` and `PromptIdentity.Version` uses and the corresponding Remove `PromptVersion` and `PromptIdentity.Version` uses and the corresponding
`prompt_version` JSON members. Prompt hashes are opaque equality values and may `prompt_version` JSON members. Prompt definition hashes are opaque equality
change when the removed field no longer participates in hashing. values and may also change because redundant captured fields were removed.
Invalidate persisted caches or comparisons that span this change; hashes from
inspection and preparation of the same captured definition still agree.
## Identity Mapping ## Identity Mapping
@@ -161,7 +193,8 @@ These are conversion examples, not aliases or bundled resources. Mounting old
defaults, or diagnostic defaults inside them. Edit those values to the exact defaults, or diagnostic defaults inside them. Edit those values to the exact
qualified IDs you chose. qualified IDs you chose.
For example, an optional caller-owned standard profile may extend a selected For example, the following wrapperless fragments can be mounted with
`SettingsInput`. An optional caller-owned standard profile may extend a selected
Gemini profile in the same `openrouter` collection: Gemini profile in the same `openrouter` collection:
```yaml ```yaml

View File

@@ -18,7 +18,7 @@ import "go.maximumdirect.net/promptkit"
For migration from the former module path, see For migration from the former module path, see
[Update Go Imports](configuration-migration.md#update-go-imports). [Update Go Imports](configuration-migration.md#update-go-imports).
Configuration APIs use `go.maximumdirect.net/configkit` v1.0.0. Its types are Configuration APIs use `go.maximumdirect.net/configkit` v1.4.0. Its types are
not interchangeable with the old ConfigKit module path. not interchangeable with the old ConfigKit module path.
The maintained [offline preparation example](../../examples/go-library/prepare/main.go) The maintained [offline preparation example](../../examples/go-library/prepare/main.go)
@@ -52,30 +52,123 @@ Create an engine with [`NewEngine`](../../engine.go). Select each directory
explicitly and place its inputs in an ordered ConfigKit underlay. A group explicitly and place its inputs in an ordered ConfigKit underlay. A group
strictly composes its files; later groups can overlay earlier ones: strictly composes its files; later groups can overlay earlier ones:
The bundled OpenRouter backend and profiles are installed in a lower underlay
by default. Select a profile explicitly, for example:
```go
input := promptkit.SettingsInput(configkit.Mapping("application", map[string]any{
"prompts": []any{map[string]any{
"id": "my-prompt",
"messages": []any{map[string]any{"role": "user", "content": "Hello"}},
"output": map[string]any{"format": "text", "validation_mode": "none"},
}},
}))
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{Main: &input}})
if err != nil { return err }
result, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "my-prompt", ProfileID: "gpt-luna-latest"})
```
The backend looks up
`OPENROUTER_API_KEY` when available; credentials are not required at engine
construction. For the complete inventory and diagnostic defaults see the
[format reference](../formats.md#bundled-openrouter-configuration). Use
`promptkit.WithBundledOpenRouter(false)` for application-only inventories.
Caller records replace matching bundled records in full; use `extends` with a
distinct ID to inherit. Replacing `auto` also changes its cost-tier children's
inherited target, while their plugins remain. External collections remain
supported.
```go ```go
promptInputs, err := promptkit.DirectoryInputs("prompts") promptInputs, err := promptkit.DirectoryInputs("prompts")
if err != nil { return err } if err != nil { return err }
profileInputs, err := promptkit.DirectoryInputs("profiles") profileInputs, err := promptkit.DirectoryInputs("profiles")
if err != nil { return err } if err != nil { return err }
// These selected files are wrapperless Promptkit fragments.
for i := range promptInputs { promptInputs[i].Destination = configkit.Keys("promptkit") }
for i := range profileInputs { profileInputs[i].Destination = configkit.Keys("promptkit") }
filesystem := configkit.NewFilesystem("application schemas", os.DirFS("."))
schemaInput := configkit.Input{
Source: promptkit.SchemaRootSource("application schemas", "schemas"),
Destination: configkit.Keys("promptkit"),
Filesystem: &configkit.FilesystemContext{FS: &filesystem, Root: ".", BaseDir: "."},
}
engine, err := promptkit.NewEngine(promptkit.Config{ engine, err := promptkit.NewEngine(promptkit.Config{
Sources: promptkit.ConfigurationSources{Underlays: []configkit.Underlay{ Sources: promptkit.ConfigurationSources{Underlays: []configkit.Underlay{
{Inputs: promptInputs}, {Inputs: promptInputs},
{Inputs: profileInputs}, {Inputs: profileInputs},
{Inputs: []configkit.Input{schemaInput}},
}}, }},
SchemaDir: "schemas",
}) })
``` ```
For a single highest-priority document, set `Sources.Main` to a For a single highest-priority document, set `Sources.Main` to a
`*configkit.Input`; omit it for ordinary layered construction. A single file `*configkit.Input`; omit it for ordinary layered construction. A single file
can use `configkit.File` directly. `ProfileSource`, `BackendSource`, and raw can use `configkit.File` directly. `PromptSource`, `ProfileSource`, `BackendSource`, and raw
`configkit.Mapping` supply programmatic definitions. Schema and runtime `configkit.Mapping` supply programmatic definitions. Runtime collaborator
collaborator options remain available. Construction validates the final options remain available. Construction validates the final
effective configuration and captures referenced prompt content; construct a effective configuration and captures referenced prompt content; construct a
new engine to observe configuration edits. See the [constructor GoDoc](../../engine.go) new engine to observe configuration edits. See the [constructor GoDoc](../../engine.go)
for the exact public contract and the [framework format reference](../formats.md) for the exact public contract and the [framework format reference](../formats.md)
for definition fields. for definition fields.
### Reuse An Application-Owned Load
An application that already loads ConfigKit configuration can embed
`promptkit.Settings` at `yaml:"promptkit"`, compose its definition with
`promptkit.Configure`, and pass the successful whole result to
`promptkit.NewFromConfig`. The [preparation example](../../examples/go-library/prepare/main.go)
shows a complete single-load workflow. Existing application resolve and
validate callbacks run before Promptkit's, and its defaults, version rule,
sequence registrations, and sensitivity predicate are preserved. Callers can
wrap the returned callbacks for work that must run afterward.
`NewFromConfig` accepts a completed result or a wrapper exposing the three
methods in [`Configuration`](../../engine.go). The selected `/promptkit`
value can be `Settings`, `*Settings`, absent, or nil; an incompatible value,
nil result, or unavailable inspection view returns `ErrInvalidConfig`. It
does not load sources, rerun callbacks, or add a bundle. Select
`BundledOpenRouterUnderlay()` before the application load if desired;
`WithBundledOpenRouter` affects only `NewEngine`. `RuntimeConfig` supplies only
client timeout, HTTP client, and transient-retry settings. Keep resolved settings
consistent with the load's frozen inspection evidence at handoff; subsequent mutation does not change captured
resources. Applications may register equivalent callbacks explicitly instead
of using `Configure`. Construction errors retain their underlying causes;
on the standalone path, a ConfigKit `LoadError` remains available through
`errors.As`.
### Load Companion Prompts Relative To The Main Configuration
A non-file underlay can use the selected main configuration's directory for
caller-declared imports. For example, this loads standalone prompt definitions
from the adjacent `prompts/` directory into the `support` collection:
```go
main := configkit.Input{Source: configkit.File(configPath)}
companion, err := promptkit.CollectionInput("support", configkit.Bytes("companion prompts", []byte(`
config:
imports:
- dir: prompts
pattern: ["*.yaml", "*.yml"]
destination: [prompts]
placement: sequence-item
required: true
`)))
if err != nil { return err }
companion.MainRelative = configkit.MainRelative()
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
Main: &main,
Underlays: []configkit.Underlay{{Inputs: []configkit.Input{companion}}},
}})
```
`CollectionInput` enables imports; for a raw `configkit.Input`, enable them with
`Imports: true`. The binding itself does not enable imports, and an explicit
`Input.Filesystem` takes precedence. Imported prompt `content_file` references
keep their source-local context. See ConfigKit's
[layered-loading contract](https://forge.maximumdirect.net/base-layer/configkit/src/tag/v1.4.0/docs/config.md#layered-loading)
for eligible inputs, fallback contexts, and behavior when the main has no usable
filesystem context.
### Use Scoped Engine Views ### Use Scoped Engine Views
Create a view when an application wants short request IDs to resolve within a Create a view when an application wants short request IDs to resolve within a
@@ -115,6 +208,8 @@ defaults, err := promptkit.FilesystemInputs("embedded defaults", applicationFS,
if err != nil { return err } if err != nil { return err }
operator, err := promptkit.DirectoryInputs(operatorDirectory) operator, err := promptkit.DirectoryInputs(operatorDirectory)
if err != nil { return err } if err != nil { return err }
for i := range defaults { defaults[i].Destination = configkit.Keys("promptkit") }
for i := range operator { operator[i].Destination = configkit.Keys("promptkit") }
profile := promptkit.ProfileSource("application choices", []promptkit.ProfileConfig{ profile := promptkit.ProfileSource("application choices", []promptkit.ProfileConfig{
{ID: "summary", Endpoint: &selectedEndpoint, Model: &selectedModel}, {ID: "summary", Endpoint: &selectedEndpoint, Model: &selectedModel},
}) })
@@ -122,7 +217,7 @@ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.Configura
Underlays: []configkit.Underlay{ Underlays: []configkit.Underlay{
{Inputs: defaults}, {Inputs: defaults},
{Inputs: operator}, {Inputs: operator},
{Inputs: []configkit.Input{{Source: profile}}}, {Inputs: []configkit.Input{promptkit.SettingsInput(profile)}},
}, },
}}) }})
``` ```
@@ -147,12 +242,23 @@ document := configkit.Mapping("application configuration", map[string]any{
}, },
}, },
}) })
input := configkit.Input{Source: document} input := promptkit.SettingsInput(document)
// Add input to a later underlay, or set Sources.Main = &input. // Add input to a later underlay, or set Sources.Main = &input.
``` ```
`ProfileSource` and `BackendSource` are smaller typed conveniences when null `PromptSource`, `ProfileSource`, and `BackendSource` are typed conveniences when
is not needed. Backend and profile definitions are supplied by the caller; null is not needed. Prompt and backend helpers, like profiles, support sibling
inheritance via `Extends`; partial records can be completed by other layers.
For relative prompt `content_file` references, supply the ordinary ConfigKit
`Input.Filesystem` context for that source, or use a
[main-relative underlay](#load-companion-prompts-relative-to-the-main-configuration).
The helpers do not read files or
validate final definitions. Profile and backend provider maps use ConfigKit Mapping input forms: built-in
scalars, `[]any`, and `map[string]any` recursively. Convert `json.Number` or
typed containers explicitly to an appropriate supported value, choosing numeric
precision for your application. Mapping captures supported values immediately;
unsupported or cyclic values are reported when loaded. Runtime override values
follow their separate rules. Backend and profile definitions are supplied by the caller;
there is no registration-only backend namespace. there is no registration-only backend namespace.
When configuration or message files change, build and validate a new engine, When configuration or message files change, build and validate a new engine,
@@ -211,7 +317,7 @@ sensitive application data.
## Inspect The Captured Configuration ## Inspect The Captured Configuration
Use [`Engine.InspectConfiguration`](../../configuration_inspection.go) for an Use [`Engine.InspectConfiguration`](../../configuration_inspection.go) for an
operator-selected report of composed configuration, source descriptions, operator-selected report of resolved configuration, source descriptions,
diagnostics, and path history. Configuration inspection uses list indexes in diagnostics, and path history. Configuration inspection uses list indexes in
the collection's nested path rather than qualified runtime IDs. For a profile the collection's nested path rather than qualified runtime IDs. For a profile
at index `i` in collection `team`, construct the path with `configkit.Keys`: at index `i` in collection `team`, construct the path with `configkit.Keys`:
@@ -219,7 +325,7 @@ at index `i` in collection `team`, construct the path with `configkit.Keys`:
```go ```go
view, err := engine.InspectConfiguration() view, err := engine.InspectConfiguration()
if err != nil { return err } if err != nil { return err }
path := configkit.Keys("collections", "team", "profiles").Index(i).Key("model") path := configkit.Keys("promptkit", "collections", "team", "profiles").Index(i).Key("model")
explanation, found := view.Explain(path) explanation, found := view.Explain(path)
if found && explanation.Effective.Redacted { if found && explanation.Effective.Redacted {
// Apply the application's disclosure policy before sharing the report. // Apply the application's disclosure policy before sharing the report.
@@ -230,11 +336,17 @@ if found && explanation.Effective.Redacted {
the application decides whether those reports may be disclosed. Credential- the application decides whether those reports may be disclosed. Credential-
shaped provider keys are redacted, but arbitrary prompt and provider content shaped provider keys are redacted, but arbitrary prompt and provider content
is not necessarily safe to share. Inspection uses the engine's captured result is not necessarily safe to share. Inspection uses the engine's captured result
and does not reopen sources or resolve credentials. It retains composed values and does not reopen sources or resolve credentials. It retains effective values,
and provenance rather than Promptkit's normalized/defaulted interpretation. including recorded field normalization and queue defaulting, alongside source
`InspectProfile` instead shows backend-completed execution settings for one profile, and `Prepare` or provenance. The view focuses effective output and `HistoricalPaths` on
`PreparedExecution.Details` shows request-specific rendered messages, inputs, `/promptkit`; it is not an isolation boundary. Metadata and source catalogs
and validation resources. None of those views is a substitute for the others. remain whole-load evidence, and explicit `Explain` queries may address paths
outside the focus. The empty path explains the whole application. Diagnostics
come from the original load and are returned as detached records.
`InspectPrompt`, `InspectProfile`, and `InspectBackend` instead show individual
resolved resources without backend-to-profile execution completion or provenance.
`Prepare` or `PreparedExecution.Details` shows completed request-specific settings,
rendered messages, inputs, and validation resources. None substitutes for the others.
To review changes between two independently captured configurations, compare To review changes between two independently captured configurations, compare
their inspection views: their inspection views:
@@ -263,11 +375,15 @@ for _, difference := range differences {
``` ```
See the [`ConfigurationInspection.Diff` GoDoc](../../configuration_inspection.go) See the [`ConfigurationInspection.Diff` GoDoc](../../configuration_inspection.go)
and ConfigKit's [v1.0.0 effective-configuration diff guidance](https://forge.maximumdirect.net/base-layer/configkit/src/tag/v1.0.0/docs/config.md#effective-configuration-diffs) and ConfigKit's [effective-configuration diff guidance](https://forge.maximumdirect.net/base-layer/configkit/src/tag/v1.4.0/docs/config.md#effective-configuration-diffs)
for the exact comparison contract. The operation compares only frozen effective for the exact comparison contract. The operation compares only frozen effective
configuration values: it does not reread external files or compare their configuration values, including native ancestor differences; sibling-only
records are excluded, but an ancestor record may contain sibling changes. It
does not reread external files or compare their
contents, resolved credential contents, later execution preparation, or contents, resolved credential contents, later execution preparation, or
request-specific overrides. An empty diff is not proof of equivalent execution. request-specific overrides. Inputs normalizing to the same effective value do
not create a value diff; lists are compared atomically. An empty diff is not
proof of equivalent execution.
Records can contain ordinary application data, so use the application's Records can contain ordinary application data, so use the application's
disclosure policy when presenting them. disclosure policy when presenting them.
@@ -346,8 +462,10 @@ if err != nil {
return err return err
} }
for _, input := range inspection.Inputs { if inspection.Inputs != nil {
// Compare the declared input with application configuration. for _, input := range *inspection.Inputs {
// Compare the declared input with application configuration.
}
} }
``` ```
@@ -483,12 +601,17 @@ if err != nil {
defer preparedExecution.Discard() defer preparedExecution.Discard()
details := preparedExecution.Details() details := preparedExecution.Details()
messageContentHash := details.MessageContentHash
_ = messageContentHash // An opaque value available before model execution.
// Inspect or persist an application-selected safe subset of details. // Inspect or persist an application-selected safe subset of details.
result, err := engine.RunPrepared(ctx, preparedExecution) result, err := engine.RunPrepared(ctx, preparedExecution)
``` ```
Preparation does not call the model or reserve backend capacity. Preparation does not call the model or reserve backend capacity.
`MessageContentHash` compares the ordered message roles and text, whereas
`RenderedPromptHash` also includes the session and cache-control hints. See
the [`PreparedRun` GoDoc](../../types.go) for their contracts and compatibility.
`RunPrepared` executes from the retained snapshot rather than reloading `RunPrepared` executes from the retained snapshot rather than reloading
consumer sources. The handle is opaque in-process state, while `Details` consumer sources. The handle is opaque in-process state, while `Details`
contains rendered content and remains subject to the application's data contains rendered content and remains subject to the application's data
@@ -505,6 +628,59 @@ retained callback without invoking it. The function's closure is retained by
identity, so synchronize any shared state it uses. Prepared start events can identity, so synchronize any shared state it uses. Prepared start events can
include identities already resolved in the frozen snapshot. include identities already resolved in the frozen snapshot.
## Bound Execution And Retain Prepared Details
Select the optional controls on [RunRequest](../../types.go). The operation
budget is separate from the provider timeout and cancellation is cooperative.
Notification is conditional on reaching the prepared point, synchronous, and
runs while admission is held; its caller-owned details can be retained even if
generation later fails. Do not rely on notification for preparation or admission
failures. The application owns any subsequent diagnostic handling.
```go
var details *promptkit.PreparedRun
result, err := engine.Run(ctx, promptkit.RunRequest{
PromptID: "answer", OperationTimeout: 30 * time.Second,
OnPrepared: func(_ context.Context, prepared promptkit.PreparedRun) { details = &prepared },
})
if errors.Is(err, promptkit.ErrOperationTimeout) && errors.Is(err, context.DeadlineExceeded) {
// details is available only if the notification was reached.
}
```
For coordinated turns, put the controls on each returned request. Assessment
in the next-turn callback runs after the previous admission and child scope
are released, so it can select its own budget:
```go
next := func(ctx context.Context, turn promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
if turn.Result != nil {
_, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "assessment", OperationTimeout: 10 * time.Second})
return nil, err
}
return &promptkit.RunRequest{PromptID: "answer", OperationTimeout: 30 * time.Second,
OnPrepared: func(_ context.Context, prepared promptkit.PreparedRun) { details = &prepared }}, nil
}
```
For split execution, [PrepareExecution](../../engine.go) retains the controls
without starting their timer; idle time does not count. Use the same caller
deadline for both calls if preparation and execution need one shared bound.
```go
handle, err := engine.PrepareExecution(ctx, promptkit.RunRequest{
PromptID: "answer", OperationTimeout: 30 * time.Second,
OnPrepared: func(_ context.Context, prepared promptkit.PreparedRun) { details = &prepared },
})
if err != nil { return err }
defer handle.Discard()
result, err := engine.RunPrepared(ctx, handle)
```
See the [observation example](../../examples/go-library/observe/main.go) for an
offline coordinated workflow and [ErrOperationTimeout](../../engine.go) for
timeout attribution.
## Execute And Validate ## Execute And Validate
[`Engine.Run`](../../engine.go) performs the same preparation, invokes the [`Engine.Run`](../../engine.go) performs the same preparation, invokes the
@@ -568,22 +744,26 @@ Set a small additional-call budget when a structurally invalid result can be
corrected automatically: corrected automatically:
```go ```go
request.Validation = &promptkit.OutputContract{ request.Validation = &promptkit.OutputContractInput{
Format: promptkit.FormatJSON, Format: promptkit.FormatJSON,
ValidationMode: promptkit.ValidationJSONSchema, ValidationMode: promptkit.ValidationJSONSchema,
SchemaPath: "events.schema.json", Schema: map[string]any{"type": "object"},
RepairAttempts: 1, RepairAttempts: new(1),
} }
``` ```
Each repair attempt is another model generation, so it can increase latency and Omission or null selects one repair for active validation and zero for `none`;
an explicit zero selects single-pass validation. A replacement resolves its
own budget and does not inherit the configured one. Each repair attempt is
another model generation, so it can increase latency and
usage; `RunResult.Usage` is cumulative and `Validation.RepairAttempts` reports usage; `RunResult.Usage` is cumulative and `Validation.RepairAttempts` reports
repair generations rather than transient HTTP attempts. Exhaustion still repair generations rather than transient HTTP attempts. Exhaustion still
returns the final failed validation result. `basic` validation can also repair returns the final failed validation result. `basic` validation can also repair
an empty candidate, but structural an empty candidate, but structural
validity is not evidence of factual or domain correctness. See the validity is not evidence of factual or domain correctness and a correction is
not guaranteed. See the
[output-contract format reference](../formats.md#output-contract) and [output-contract format reference](../formats.md#output-contract) and
[`OutputContract` GoDoc](../../types.go) for the exact budget and eligibility [`OutputContractInput` GoDoc](../../types.go) for the exact budget and eligibility
rules. rules.
The built-in client also retries eligible temporary HTTP statuses. By default, The built-in client also retries eligible temporary HTTP statuses. By default,
@@ -619,6 +799,116 @@ application owns any semantic retry budget. When a
pre-execution equality check is required, use `PrepareExecution` and compare pre-execution equality check is required, use `PrepareExecution` and compare
its opaque rendered-prompt hash before invoking `RunPrepared`. its opaque rendered-prompt hash before invoking `RunPrepared`.
### Capture And Continue A Conversation
The [request GoDoc](../../types.go), [exchange GoDoc](../../exchange.go), and
[prepared-execution GoDoc](../../prepared_execution.go) own the exact capture,
continuation, inspection, and lifetime contracts.
Capture records the messages actually executed and the final assistant content.
An exchange is an immutable in-process capability; `Details()` returns detached,
encodable history that may contain sensitive content. Ordinary result encoding
does not include the exchange. Continuations imply capture and may branch or use
a different engine, profile, output contract, credential, or session:
```go
initial, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "proposal", CaptureExchange: true})
if err != nil { return err }
left, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "refine", ContinueFrom: initial.Exchange})
if err != nil { return err }
right, err := anotherEngine.Run(ctx, promptkit.RunRequest{PromptID: "alternative", ContinueFrom: initial.Exchange})
if err != nil { return err }
_, _ = left, right // Independent branches; each result has its own exchange.
```
For a frozen execution, set `CaptureExchange` in the request passed to
`PrepareExecution`, then obtain the exchange from `RunPrepared`'s result.
Preparation and validation alone do not produce an exchange. Captured history
precedes newly rendered definition messages and appended messages. A completed
structurally invalid result still has an exchange; an operational error returns
no result. The caller chooses what constitutes acceptance.
```go
prepared, err := engine.PrepareExecution(ctx, promptkit.RunRequest{PromptID: "proposal", CaptureExchange: true})
if err != nil { return err }
result, err := engine.RunPrepared(ctx, prepared)
if err != nil { return err }
details := result.Exchange.Details()
_ = details // Explicitly encode or inspect when appropriate.
```
### Coordinate Application-Directed Turns
The following sketches use illustrative application functions. The first records
each newly delivered response before selecting another question:
```go
outcome, err := engine.Coordinate(ctx, promptkit.CoordinateRequest{
Start: initial.Exchange, AdditionalTurns: 3,
Next: func(ctx context.Context, turn promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
if turn.Result != nil {
if !turn.Result.Validation.IsValid { return nil, fmt.Errorf("invalid response") }
analysis.Record(turn.Result)
}
question, ok := analysis.NextQuestion()
if !ok { return nil, nil }
return &promptkit.RunRequest{PromptID: "research/investigate", Vars: map[string]string{"question": question}}, nil
},
})
```
An application can instead start with existing findings, assess each new
candidate, and keep its decisions in a closure:
```go
outcome, err := engine.Coordinate(ctx, promptkit.CoordinateRequest{
Start: proposal.Exchange, AdditionalTurns: 3, Observer: branchGroup.Observe,
Next: func(ctx context.Context, turn promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
if turn.Result != nil {
if !turn.Result.Validation.IsValid { return nil, fmt.Errorf("invalid revision") }
candidate, withdrawn, err := decodeRevision(turn.Result.RawOutput)
if err != nil { return nil, err }
if withdrawn { branch.MarkWithdrawn(); return nil, nil }
findings, err := assess(ctx, engine, candidate, validatorGroup.Observe)
if err != nil { return nil, err }
branch.Update(candidate, findings)
if branch.Done() { return nil, nil }
}
return &promptkit.RunRequest{PromptID: branch.FollowUpPrompt(), Vars: branch.Variables()}, nil
},
})
```
`assess` is application code and explicitly selects its observer on any nested
run. A completed turn releases its admission lease before the callback, so that
assessment may use the same backend. `stopped` and `exhausted` describe control
flow, not content acceptance. Cancellation before callback delivery can return one
completed response in `Undelivered`; errors retain their ordinary identities.
### Collect Selected Execution Facts
Groups are independent of continuation and enrollment is explicit:
```go
var group, validatorGroup promptkit.ExecutionGroup
request.Observer = promptkit.ComposeObservers(diagnosticObserver, group.Observe)
prepared, err := engine.PrepareExecution(ctx, request) // no event yet
if err != nil { return err }
result, err := engine.RunPrepared(ctx, prepared)
if err != nil { return err }
_ = result
// Set CoordinateRequest.Observer to group.Observe for its continuation runs.
// A nested validator selects validatorGroup.Observe on its own request.
summary, validatorSummary := group.Snapshot(), validatorGroup.Snapshot()
_, _ = summary, validatorSummary
```
Only the first terminal event per run contributes counts and usage; coverage of
currently observed members is separate from token-usage completeness. Empty
groups report complete, known zero totals, while members without terminals do
not invent totals. Overlapping groups are not generally additive. See the
[offline example](../../examples/go-library/observe/main.go) for complete setup.
## Inputs, Profiles, And Overrides ## Inputs, Profiles, And Overrides
Use `File`, `Inline`, or `InlineWithURI` to construct request inputs. A request Use `File`, `Inline`, or `InlineWithURI` to construct request inputs. A request
@@ -662,9 +952,9 @@ the engine to pick up changed definitions or message files.
### Choose A Profile Ownership Model ### Choose A Profile Ownership Model
Supply backend and profile definitions through the configuration sources your Supply backend and profile definitions through the configuration sources your
application selects. PromptKit does not install provider catalogs or built-in application selects, alongside the default-on bundled OpenRouter inventory.
profile inventories; conventional backend ID constants are names only. Define Conventional backend ID constants are names only. Define or replace the model,
the model, endpoint, credentials lookup, and settings that your application endpoint, credentials lookup, and settings that your application
intends to support. A profile may inherit a sibling profile by its local key in intends to support. A profile may inherit a sibling profile by its local key in
the same root or collection list; its backend reference uses the exact the same root or collection list; its backend reference uses the exact
configured identity. Caller-selected source layers determine the final values. The configured identity. Caller-selected source layers determine the final values. The
@@ -673,10 +963,9 @@ composition rules.
### Inspect A Profile Before Prompt Work ### Inspect A Profile Before Prompt Work
Use [`Engine.InspectProfile`](../../engine.go) to validate one configured Use [`Engine.InspectProfile`](../../engine.go) to inspect a resolved profile
profile without constructing a synthetic prompt or placeholder inputs. It without constructing a synthetic prompt or placeholder inputs. It reports the
resolves the profile's effective target but does not prepare or execute a profile's own fields, not a backend-completed target:
prompt:
```go ```go
inspection, err := engine.InspectProfile(ctx, profileID) inspection, err := engine.InspectProfile(ctx, profileID)
@@ -684,23 +973,48 @@ if err != nil {
return err return err
} }
target := inspection.EffectiveModelParams if inspection.APIKeyRequired != nil && *inspection.APIKeyRequired {
if target.APIKeyEnv != "" { // Arrange a usable direct or configured environment credential.
// This is a configured optional environment lookup source. }
} else if inspection.APIKeyRequired { if inspection.APIKeyEnv != nil {
// Arrange a direct credential before later execution. // This is the configured environment lookup source, not its value.
} }
``` ```
Use this configuration-time boundary when only the profile and its target need Use [`InspectBackend`](../../engine.go) to inspect the referenced backend
checking. Use `Prepare` when the application also needs prompt, input, schema, separately. Its diagnostics are configured metadata, not live checks or derived
or rendering work; use prepared execution when that work must remain tied to a destinations; `ErrBackendNotFound` identifies an absent backend. References in
later execution. A reported `APIKeyEnv` is a configured optional source, while all three resource reports remain as authored. Use `Prepare` for completed
`APIKeyRequired` is the explicit local requirement. The request settings and rendering, or prepared execution when that work must remain
tied to later execution. A reported `APIKeyEnv` identifies a configured source,
not its value; `APIKeyRequired` can also be present. The
[credential format reference](../formats.md#credentials) and the method's [credential format reference](../formats.md#credentials) and the method's
[GoDoc](../../engine.go) own the exact precedence, timing, result, and error [GoDoc](../../engine.go) own the exact precedence, timing, result, and error
contracts. contracts.
Resource inspection pointers represent optional fields: nil is absent, while a
non-nil pointer preserves zero, false, or an empty string. Pointer-to-collection
fields distinguish omission from an empty map or list. Results and nested mutable
values belong to the caller. Ordinary generic encoders need no per-field adapter:
```go
jsonBytes, err := json.Marshal(inspection)
if err != nil { return err }
yamlBytes, err := yaml.Marshal(inspection) // go.yaml.in/yaml/v4
if err != nil { return err }
_, _ = jsonBytes, yamlBytes
```
For migration, replace `ProfileInspection.EffectiveModelParams` accesses with
the optional resource fields, and use `Prepare` or `PreparedExecution.Details`
where backend completion and request overrides are needed. Prompt inspection's
`DefaultProfileID`, `Inputs`, nested input metadata, and output fields now use
optional pointers where presence matters; check for nil before dereferencing.
Consumers such as CommandKit can render the results directly instead of copying
each field into a parallel report. See the result types' [GoDoc](../../resource_inspection.go)
for keys and ownership; resource inspection does not replace `InspectConfiguration`
provenance, `InspectRequirements`, or concrete-request `ValidateRequest`.
### Set A Per-Run Session And Reasoning ### Set A Per-Run Session And Reasoning
Supply a direct session ID when one prompt should be correlated with a Supply a direct session ID when one prompt should be correlated with a
@@ -742,7 +1056,7 @@ endpoint, model := "http://localhost:8000/v1", "example-model"
profile := promptkit.ProfileSource("local profile", []promptkit.ProfileConfig{ profile := promptkit.ProfileSource("local profile", []promptkit.ProfileConfig{
{ID: "local-summary", Endpoint: &endpoint, Model: &model}, {ID: "local-summary", Endpoint: &endpoint, Model: &model},
}) })
// Add configkit.Input{Source: profile} to a caller underlay. // Add promptkit.SettingsInput(profile) to a caller underlay.
``` ```
Endpoint-only profiles have an empty backend ID and remain unrestricted by Endpoint-only profiles have an empty backend ID and remain unrestricted by
@@ -850,6 +1164,22 @@ secret-handling policies.
## Extension Interfaces ## Extension Interfaces
JSON Schema validation and native structured output use a self-contained root
document. See [schema selection and resource context](../formats.md#schemas).
Prepared executions retain their captured schema even if the source later changes.
### Migrating Schema Sources
`Config.SchemaDir`, `WithSchemaFS`, and `WithSchemaFile` have been removed.
Supply a context-backed `SchemaRootSource` at the desired ConfigKit layer, as
shown above, or declare `schema_root` in application or collection mappings.
For a former single-file source, configure `schema_file` relative to its
contributing source, or supply its directory as a root and use its basename in
`schema_path`. To substitute an unrelated file per request, read and decode it
in the caller and supply its value as `OutputContractInput.Schema`. There is no
implicit working-directory root. Public structs gained fields; change any
positional struct literals to keyed literals when migrating.
Inject an [`LLMClient` or `ArtifactReader`](../../types.go) when the built-in Inject an [`LLMClient` or `ArtifactReader`](../../types.go) when the built-in
behavior does not fit the application. Their GoDoc defines concurrent use, behavior does not fit the application. Their GoDoc defines concurrent use,
context handling, ownership of copied values, nil responses, and preservation context handling, ownership of copied values, nil responses, and preservation
@@ -857,6 +1187,11 @@ of collaborator errors. Implementations must honor cancellation, safely manage
copies they retain, avoid unsafe logging of content or credentials, and enforce copies they retain, avoid unsafe logging of content or credentials, and enforce
the application policy that motivated the injection. the application policy that motivated the injection.
A successful `ArtifactReader.Read` transfers the returned artifact and mutable
body storage to Promptkit. Readers that previously reused a returned buffer
must instead return independent storage and stop mutating it after return;
private scratch buffers may still be reused when the returned bytes are copied.
## Handle Errors ## Handle Errors
Use `errors.Is` with the Use `errors.Is` with the

View File

@@ -58,6 +58,14 @@ exist.
## Maintainer Validation ## Maintainer Validation
Use the newest Go installation already available locally for ordinary
development checks. Put its `bin` directory first on `PATH` and set
`GOTOOLCHAIN=local` to keep that toolchain selected. For repository-isolated
checks, set `GOWORK=off`; consumers may use their own workspaces. The selected
validation patch does not by itself change the module's consumer minimum. See
the [release toolchain policy](policy/release.md#consumer-minimum-and-release-toolchain)
and [release procedure](release.md#prepare-the-candidate) for release selection.
For ordinary work, select validation proportionate to the change. Do not repeat For ordinary work, select validation proportionate to the change. Do not repeat
successful checks for unchanged content merely to create a commit. During successful checks for unchanged content merely to create a commit. During
development, run relevant Go tests and vet, format changed Go files, and execute development, run relevant Go tests and vet, format changed Go files, and execute

View File

@@ -6,71 +6,58 @@ This is the canonical reference for Promptkit's backend, profile, prompt, and
schema definitions. The [Go consumer guide](consumers/pkg-promptkit.md) shows schema definitions. The [Go consumer guide](consumers/pkg-promptkit.md) shows
how to select sources and use an engine; the how to select sources and use an engine; the
[OpenAI-compatible integration contract](integrations/openai-compatible-chat.md) [OpenAI-compatible integration contract](integrations/openai-compatible-chat.md)
owns outbound wire behavior. [ConfigKit](https://forge.maximumdirect.net/base-layer/configkit/src/tag/v1.0.0/docs/config.md) owns outbound wire behavior. [ConfigKit](https://forge.maximumdirect.net/base-layer/configkit/src/tag/v1.4.0/docs/config.md)
owns the generic YAML grammar, imports, layering, inheritance, and diagnostics. owns the generic YAML grammar, imports, layering, inheritance, and diagnostics.
This reference describes what Promptkit does with the final effective values. This reference describes what Promptkit does with the final effective values.
A configuration document may contain top-level `backends`, `profiles`, and A complete configuration document places Promptkit settings beneath `promptkit`.
`prompts` lists, plus a `collections` mapping of namespaced resource groups. The `backends`, `profiles`, and `prompts` lists contain explicit literal
Each resource has an explicit literal `id`, not a filename-derived identity. `id` values; `collections` groups namespaced resources. IDs remain exact and
IDs and references must be nonblank and remain exact: Promptkit does not trim or do not include the `promptkit` path. Standalone resource fragments may omit
re-key them. Unknown fields are errors. Promptkit installs no provider or the wrapper when explicitly mounted as described below.
profile catalog; callers select the definitions they need.
```yaml ```yaml
backends: promptkit:
- id: local profiles:
endpoint: http://localhost:8000/v1 - id: local-summary
concurrency_limit: 2 endpoint: http://localhost:8000/v1
profiles: model: example-model
- id: local-summary prompts:
backend: local - id: meeting.summary
model: example-model default_profile: local-summary
prompts: messages:
- id: meeting.summary - role: user
default_profile: local-summary content: Summarize this note.
messages: output:
- role: user format: text
content: Summarize this note. validation_mode: none
output: collections:
format: text support:
validation_mode: none release_version: "1.4.0"
collections: backends:
support: - id: offline
release_version: "1.4.0" endpoint: https://example.invalid/v1
backends: profiles:
- id: offline - id: fast
endpoint: https://example.invalid/v1 backend: support/offline
profiles: model: offline-model
- id: base
backend: support/offline
model: offline-model
- id: fast
extends: {id: base}
timeout_seconds: 30
prompts:
- id: answer
default_profile: support/fast
messages:
- role: user
content: Answer the selected caller input.
output:
format: text
validation_mode: none
``` ```
## Source Headers, Collections, And Resource IDs ## Source Headers, Collections, And Resource IDs
An input may omit a source header. When present, the source-root `version` is Applications own ConfigKit's version policy and source controls. Promptkit adds
reserved for ConfigKit's specification header and accepts only the exact string no resource-version rule, including for collection, prompt, or profile records.
`"2"` or strict decimal integer `2`; null and other values are invalid. This An application's root `version` is distinct from a collection's descriptive
header is distinct from a collection's `release_version` and ConfigKit's own `release_version`. Old root-level `prompts` or `profiles` keys are application
software version. ConfigKit checks and consumes accepted headers on non-main data, not Promptkit resources, and are neither consumed nor specially rejected.
inputs; a declaration on the main input remains visible in explicit
configuration inspection. Headerless inputs still undergo ordinary final-value The `promptkit` section, each collection, backend, profile, and prompt retain
validation. ConfigKit owns generic import and source-history behavior. unrecognized fields as application data. Such fields compose normally and are
For a standalone prompt item, a root `version` is only a source header, not visible to application validation and configuration inspection, but do not
prompt metadata. become provider parameters or appear in resource reports. Grouping them under
an application-specific key reduces collisions; a misspelled optional Promptkit
field can also be accepted there. Nested Promptkit-owned objects, including
message, input, output, and smoke-test objects, remain strict.
The root resource lists contain unqualified IDs. Each `collections` entry The root resource lists contain unqualified IDs. Each `collections` entry
has a namespace key and may contain `release_version` plus the same three has a namespace key and may contain `release_version` plus the same three
@@ -86,9 +73,10 @@ list. It is not a qualified or cross-collection reference.
no label; an explicitly empty string is retained. It does not select a resource no label; an explicitly empty string is retained. It does not select a resource
or change execution. Runtime lookups or change execution. Runtime lookups
use the captured effective resource indexes. `InspectConfiguration` is an use the captured effective resource indexes. `InspectConfiguration` is an
explicit operator view of composed settings and provenance. Normalization and explicit operator view of resolved settings and provenance. Selected field
defaults appear in Promptkit's interpreted execution views, not as changes to normalizations and the derived queue default are recorded in that tree, with
that composed tree. Source history is not a second runtime resolver, and the source history retained. Request-time completion remains separate. Source
history is not a second runtime resolver, and the
retained effective shape is not flattened or rewritten into those indexes. retained effective shape is not flattened or rewritten into those indexes.
For examples of explicitly selecting native collection inputs and inspecting For examples of explicitly selecting native collection inputs and inspecting
@@ -97,8 +85,9 @@ inventories, see the [consumer guide](consumers/pkg-promptkit.md) and its
breaking identity and prompt-selection changes, see the breaking identity and prompt-selection changes, see the
[configuration migration guide](consumers/configuration-migration.md). [configuration migration guide](consumers/configuration-migration.md).
Every effective backend, profile, and prompt definition is validated when Every effective backend, profile, and prompt definition is validated during
`NewEngine` constructs the engine, even if it is never selected. A profile may the registered configuration load and engine capture, even if never selected.
A profile may
be partial: an endpoint and model need only be complete after request overrides be partial: an endpoint and model need only be complete after request overrides
are applied. A prompt's default profile similarly need not be runnable without are applied. A prompt's default profile similarly need not be runnable without
an overriding request. Missing request data, artifacts, schemas, credentials, an overriding request. Missing request data, artifacts, schemas, credentials,
@@ -109,13 +98,15 @@ and static template syntax at construction. It does not reread configuration or
message files on ordinary runs; construct a new engine to observe edits. message files on ordinary runs; construct a new engine to observe edits.
Schema and request artifact reads still happen during preparation. Use Schema and request artifact reads still happen during preparation. Use
`InspectConfiguration` for an explicit effective-configuration or provenance `InspectConfiguration` for an explicit effective-configuration or provenance
report, `InspectProfile` for backend-completed profile settings, and preparation report, resource inspection for resolved prompt, profile, and backend fields,
for request-specific rendered details. For identity discovery before inspection, and preparation for completed request-specific details. For identity discovery before inspection,
see the consumer guide's [resource and collection listing](consumers/pkg-promptkit.md#discover-available-resources-and-collections). see the consumer guide's [resource and collection listing](consumers/pkg-promptkit.md#discover-available-resources-and-collections).
## Prompt Definitions ## Prompt Definitions
The `prompts` list contains definitions selected by their literal IDs: The `prompts` list contains definitions selected by their literal IDs. The
following examples in this reference are wrapperless `promptkit` fragments
unless explicitly shown as complete documents; mount them with `SettingsInput`.
```yaml ```yaml
prompts: prompts:
@@ -146,8 +137,9 @@ prompts:
`default_profile`, `description`, `session_id`, `inputs`, and cache-control `default_profile`, `description`, `session_id`, `inputs`, and cache-control
fields may be absent or cleared with null where appropriate. A null whole fields may be absent or cleared with null where appropriate. A null whole
definition, required role, output format, or validation mode is invalid. definition, required role, output format, or validation mode is invalid.
`required` and `repair_attempts` default to false and zero, respectively, and `required` defaults to false and is not nullable. An omitted or null
are not nullable. `InspectPrompt` selects the same captured definition without `repair_attempts` resolves according to `validation_mode`; see the output
contract below. `InspectPrompt` selects the same captured definition without
rendering or resolving a profile. rendering or resolving a profile.
### Inputs ### Inputs
@@ -200,18 +192,30 @@ messages; the integration contract defines its wire representation.
| --- | --- | --- | | --- | --- | --- |
| `format` | yes | `text`, `markdown`, or `json` | | `format` | yes | `text`, `markdown`, or `json` |
| `validation_mode` | yes | `none`, `basic`, `json`, or `json_schema` | | `validation_mode` | yes | `none`, `basic`, `json`, or `json_schema` |
| `schema_path` | for `json_schema` | Path in the selected schema source | | `schema` | one selection for `json_schema` | Inline JSON Schema value |
| `repair_attempts` | no | Integer 0–3; default 0. Positive values need `basic`, `json`, or `json_schema` | | `schema_file` | one selection for `json_schema` | File relative to the contributing field's source context |
| `schema_path` | one selection for `json_schema` | Path within the applicable explicit schema root |
| `repair_attempts` | no | Integer 0–3; omitted or null defaults to 0 for `none`, otherwise 1. Positive values need `basic`, `json`, or `json_schema` |
`none` skips content validation; `basic` requires nonempty content; `json` `none` skips content validation; `basic` requires nonempty content; `json`
accepts any valid JSON value; `json_schema` checks the selected schema. accepts any valid JSON value; `json_schema` checks the selected schema.
In `json_schema` mode exactly one nonnull `schema`, `schema_file`, or
`schema_path` must be selected. Null clears a selection; `false`, `true`, and
`{}` are values. Blank references are invalid. Other modes ignore these fields
for schema selection and loading. ConfigKit composes inline mappings normally;
switching forms can require explicitly clearing an inherited alternative.
Replacing an identified prompt record replaces its complete earlier record.
`format` determines output artifact metadata. JSON Schema mode also supplies `format` determines output artifact metadata. JSON Schema mode also supplies
structured-output metadata to compatible clients. Plain `json` does not structured-output metadata to compatible clients. Plain `json` does not
request a provider-native JSON-object constraint. A positive repair budget request a provider-native JSON-object constraint. A positive repair budget
counts additional generations after failed validation; exhaustion returns the counts additional generations after failed validation; exhaustion returns the
last candidate and its failed validation result. Operational errors remain last candidate and its failed validation result. Operational errors remain
errors. A request-level `OutputContract` replaces the whole configured one, errors. A request-level `OutputContractInput` replaces the whole configured
not individual fields; an empty request format defaults to `text`. contract, not individual fields; an empty request format defaults to `text`.
Its omitted budget resolves from its own validation mode, not the prompt's
budget. Explicit zero disables repair. A failed validation can incur one extra
generation by default, with corresponding latency and usage; correction is not
guaranteed.
## Backend Definitions ## Backend Definitions
@@ -228,7 +232,10 @@ negative value is invalid. For positive concurrency, omitted/null queue
capacity defaults to 1024. An explicit queue capacity, including zero, capacity defaults to 1024. An explicit queue capacity, including zero,
requires positive concurrency and must be nonnegative. Unlimited concurrency requires positive concurrency and must be nonnegative. Unlimited concurrency
has no queue. Provider maps contain JSON-compatible values; null entries are has no queue. Provider maps contain JSON-compatible values; null entries are
ordinary JSON null. Unknown fields, including a raw `api_key`, fail. ordinary JSON null. A raw `api_key` is not a Promptkit credential field: an
unrecognized backend field is application data and is not automatically redacted
by Promptkit. Applications own sensitivity declarations and disclosure policy
for their extension fields.
`smoke_tests` is optional backend metadata for explicit diagnostic operations. `smoke_tests` is optional backend metadata for explicit diagnostic operations.
It may contain `default_profile`, `reachability`, and `authentication`: It may contain `default_profile`, `reachability`, and `authentication`:
@@ -301,7 +308,9 @@ Nonblank endpoints follow the backend URL rules; temperature is 0–2, top_p
is 0–1, max_tokens and timeout_seconds are nonnegative. Provider maps accept is 0–1, max_tokens and timeout_seconds are nonnegative. Provider maps accept
JSON-compatible values with nonempty keys and cannot collide with the built-in JSON-compatible values with nonempty keys and cannot collide with the built-in
client's [standard request fields](integrations/openai-compatible-chat.md#request-body). client's [standard request fields](integrations/openai-compatible-chat.md#request-body).
Credential values do not belong in profiles: raw `api_key` is invalid. A raw `api_key` is not a Promptkit credential setting; an unrecognized profile
field is application data. Applications own its sensitivity declarations and
disclosure policy, just as for backend extension fields.
### Defaults And Overrides ### Defaults And Overrides
@@ -314,26 +323,92 @@ only a model or backend does not clear unrelated explicitly inherited fields.
| Profile field | Absent | Explicit value | Explicit null | | Profile field | Absent | Explicit value | Explicit null |
| --- | --- | --- | --- | | --- | --- | --- | --- |
| `backend` | No backend | Exact backend ID; blank means none | No backend | | `backend` | No backend | Exact backend ID; blank means none | No backend |
| `endpoint` | Selected backend endpoint | Valid nonblank URL; empty clears | Clear, without backend fallback | | `endpoint` | Selected backend endpoint | Valid nonblank URL; empty clears | Selected backend endpoint |
| `model` | Unspecified | Nonblank model; empty clears | Unspecified | | `model` | Unspecified | Nonblank model; empty clears | Unspecified |
| `temperature`, `top_p`, `max_tokens` | Unspecified | Valid value, including outbound zero | Suppress outbound control | | `temperature`, `top_p`, `max_tokens` | Unspecified | Valid value, including outbound zero | Suppress outbound control |
| `service_tier`, `reasoning_effort` | Unspecified | Normalized value; blank suppresses | Suppress outbound control | | `service_tier`, `reasoning_effort` | Unspecified | Normalized value; blank suppresses | Suppress outbound control |
| `timeout_seconds` | 600 seconds | Nonnegative; zero disables generation deadline | Reset to 600 seconds | | `timeout_seconds` | 600 seconds | Nonnegative; zero disables generation deadline | Reset to 600 seconds |
| `api_key_env` | Backend lookup name, if any | Trimmed name; blank disables | Disable backend lookup | | `api_key_env` | Backend lookup name, if any | Trimmed name; blank disables | Backend lookup name, if any |
| `api_key_required` | False | Boolean, including false | False | | `api_key_required` | False | Boolean, including false | False |
| `extra_params` | Backend defaults | Recursively overlay profile map on backend map | Suppress configured provider map | | `extra_params` | Backend defaults | Recursively overlay profile map on backend map | Backend defaults |
An empty profile mapping does not clear backend map entries. After ConfigKit
composition and inheritance, a whole-map null and omission both permit backend
defaults. Nested maps merge recursively, while scalar and array values replace;
nulls within provider maps or arrays remain literal data.
When `api_key_required` is true and the profile source is absent or null,
execution suppresses the backend credential source; an explicitly configured
profile source remains in use.
For example, if the backend provides an endpoint, API-key environment name,
and provider map, a profile containing `endpoint: null`, `api_key_env: null`,
and `extra_params: null` now inherits those backend values after handoff (unless
the required-key rule suppresses the credential source). Earlier versions used
those nulls to suppress the corresponding fallback. Use `endpoint: ""` or
`api_key_env: ""` to clear those individual strings. There is no profile-level
whole-map suppression replacement; change the backend map through ordinary
ConfigKit composition when different provider defaults are needed. ConfigKit's
retained configuration and provenance continue to show the nulls.
An empty mapping does not clear inherited map entries; null clears the whole
map. Nested maps merge recursively, while scalar and array values replace.
Runtime nonempty `ExtraParams` overrides replace the entire configured map. Runtime nonempty `ExtraParams` overrides replace the entire configured map.
Request overrides otherwise follow the [public value GoDoc](../types.go). Request overrides otherwise follow the [public value GoDoc](../types.go).
For example, a runtime `extra_params["extends"]` key is ordinary provider For example, a runtime `extra_params["extends"]` key is ordinary provider
data, not a configuration inheritance instruction. An explicit zero timeout data, not a configuration inheritance instruction. An explicit zero timeout
does not disable the caller context or HTTP transport timeout. does not disable the caller context or HTTP transport timeout.
### Bundled OpenRouter Configuration
The default bundle installs root backend `openrouter`, endpoint
`https://openrouter.ai/api/v1`, optional credential environment name
`OPENROUTER_API_KEY`, and smoke metadata selecting `gpt-luna-latest` with the
`openrouter_key` authentication adapter. Existing diagnostic endpoint fallback
applies. No prompt, implicit profile selection, or tuning defaults are installed.
Disable the bundle for `NewEngine` with `WithBundledOpenRouter(false)`; an
application-owned load selects `BundledOpenRouterUnderlay` explicitly.
The root profile IDs and their models are:
| Profile | Model |
| --- | --- |
| `gpt-astra-latest` | `~openai/gpt-astra-latest` |
| `gpt-sol-latest` | `~openai/gpt-sol-latest` |
| `gpt-terra-latest` | `~openai/gpt-terra-latest` |
| `gpt-luna-latest` | `~openai/gpt-luna-latest` |
| `claude-fable-latest` | `~anthropic/claude-fable-latest` |
| `claude-opus-latest` | `~anthropic/claude-opus-latest` |
| `claude-sonnet-latest` | `~anthropic/claude-sonnet-latest` |
| `claude-haiku-latest` | `~anthropic/claude-haiku-latest` |
| `gemini-pro-latest` | `~google/gemini-pro-latest` |
| `gemini-flash-latest` | `~google/gemini-flash-latest` |
| `auto` | `openrouter/auto` |
| `free` | `openrouter/free` |
The profiles `auto-low`, `auto-medium`, `auto-high`, `auto-xhigh`, and
`auto-max` extend `auto` and add an `auto-router` plugin with the corresponding
`cost_tier`. Replacing `auto` changes their inherited backend and model while
retaining their plugins. These are root IDs even when a model contains `/`.
For example, a caller can replace `auto` as a complete record and extend another
bundled profile under a new ID:
```yaml
profiles:
- id: auto
backend: openrouter
model: openrouter/free
- id: my-luna
extends: {id: gpt-luna-latest}
timeout_seconds: 60
```
See OpenRouter's [alias examples](https://openrouter.ai/~openai/gpt-luna-latest),
[auto router and cost tiers](https://openrouter.ai/docs/guides/routing/routers/auto-router),
and [free router](https://openrouter.ai/docs/guides/routing/routers/free-router)
for provider behavior and availability.
### Source And Profile Precedence ### Source And Profile Precedence
`NewEngine` composes caller underlays in their given order, followed by an `NewEngine` composes the bundled OpenRouter underlay first by default, then
caller underlays in their given order, followed by an
optional main input. Later layers overlay earlier ones. Inputs within one optional main input. Later layers overlay earlier ones. Inputs within one
underlay group are strictly composed: conflicting contributions are errors, underlay group are strictly composed: conflicting contributions are errors,
not priority signals. The directory helpers discover `.yaml` and `.yml` files recursively not priority signals. The directory helpers discover `.yaml` and `.yml` files recursively
@@ -344,8 +419,12 @@ on an input. Single files can use ConfigKit inputs directly.
Root backends, profiles, and prompts, as well as collection declarations and Root backends, profiles, and prompts, as well as collection declarations and
their resource lists, follow these same source rules. `CollectionInput` their resource lists, follow these same source rules. `CollectionInput`
mounts one explicitly selected source at a collection destination and enables mounts one explicitly selected source at `/promptkit/collections/<namespace>` and enables
ConfigKit imports for that input; it does not discover or select sources. ConfigKit imports for that input; it does not discover or select sources.
`SettingsInput` mounts wrapperless settings fragments at `/promptkit`; an
id-first resource file uses ConfigKit sequence-item placement at the matching
resource list. Directory discovery retains root destinations, so callers mount
discovered wrapperless fragments explicitly. No document shape is inferred.
A later same-ID record replaces the lower definition in full. Use a distinct A later same-ID record replaces the lower definition in full. Use a distinct
inherited sibling when a variant should retain its parent's fields. inherited sibling when a variant should retain its parent's fields.
A request `ProfileID` takes precedence over a prompt `default_profile`; if A request `ProfileID` takes precedence over a prompt `default_profile`; if
@@ -363,16 +442,39 @@ references are exact IDs and use their qualified form when they select a
collection resource. There is no Promptkit-specific depth limit or second collection resource. There is no Promptkit-specific depth limit or second
base-chain resolver. Missing references and cycles fail at construction. For base-chain resolver. Missing references and cycles fail at construction. For
generic inheritance and layer mechanics, consult generic inheritance and layer mechanics, consult
[ConfigKit](https://forge.maximumdirect.net/base-layer/configkit/src/tag/v1.0.0/docs/config.md). [ConfigKit](https://forge.maximumdirect.net/base-layer/configkit/src/tag/v1.4.0/docs/config.md).
## Schemas ## Schemas
Schemas are JSON documents selected by a prompt or request `schema_path`. Schemas can be inline, source-relative files, or files under an explicit root.
Directory and `fs.FS` schema paths resolve within the configured root; An optional `schema_root` may be set at the `promptkit` root and in each
referenced nested schemas resolve relative to the owning schema document. collection. A collection's nonnull root takes precedence over the application
`WithSchemaFile` exposes one schema under its base name. An unreadable, root for its prompts; an absent or null collection root falls back to the
invalid, or unresolvable schema causes an operational validation error. application root. A nonnull blank root is invalid. There is no implicit
working-directory root. Roots and `schema_file` references resolve using the
effective field's contributing ConfigKit filesystem and base directory;
missing usable context is a configuration error. An explicit `.` is valid
with usable context. A `schema_path` requires an applicable root. Root choice
is made before reading: a missing file never triggers another-root search.
Schema selection and context are captured during construction without reading
schema files. Files are read and compiled during preparation; a missing,
malformed, or unresolvable file is an operational validation error. Inline
configuration values follow native ConfigKit composition and null handling.
For request-level replacement, inline `Schema` needs no root and `SchemaPath`
uses the selected prompt's applicable root. A replacement does not inherit the
configured schema selection or `schema_file` context.
Runtime inline values are encoded as JSON before preparation; file-backed JSON
uses the existing number-preserving decoder. No additional numerical equivalence
between input representations is implied.
Each schema must be self-contained: in-document references, IDs, and anchors are
supported, but references requiring another file or network resource fail.
References must also work without the source filename as a base; use fragment
references or explicit schema IDs rather than relying on that filename.
Valid JSON that fails the schema instead yields a failed validation result. Valid JSON that fails the schema instead yields a failed validation result.
Schema documents are detached without applying the provider extra-parameter
copy limits; this does not change the schema decoder, compiler, or
self-contained-resource rules.
## Credentials ## Credentials

View File

@@ -16,14 +16,13 @@ Generation sends an HTTP `POST` with `Content-Type: application/json`.
Before the client is called, the engine resolves framework, backend, profile, Before the client is called, the engine resolves framework, backend, profile,
and request values into one execution target. Endpoint configuration is trimmed and request values into one execution target. Endpoint configuration is trimmed
and must be an absolute HTTP or HTTPS URL with a host and without user and must be an absolute HTTP or HTTPS URL with a host and without user
information, a query, or a fragment. A non-empty endpoint from the target information, a query, or a fragment. The target's endpoint is validated again
overrides the client's configured base URL. The final selected endpoint is before transport.
validated again before transport.
The completion URL is composed through parsed URL path operations. Nested base The completion URL is composed through parsed URL path operations. Nested base
paths are retained, repeated trailing slashes are normalized, and the result paths are retained, repeated trailing slashes are normalized, and the result
has exactly one appended `/chat/completions` suffix. Generation fails before has exactly one appended `/chat/completions` suffix. Generation fails before
sending when neither source supplies a valid endpoint. sending when the target does not supply a valid endpoint.
The target's backend ID is routing metadata for prepared values, results, and The target's backend ID is routing metadata for prepared values, results, and
injected clients. The built-in client does not derive the URL from that ID and injected clients. The built-in client does not derive the URL from that ID and
@@ -70,7 +69,9 @@ and numeric `usage`; unrelated fields and account limits are ignored. This is
only a minimal identifying subset of OpenRouter's documented only a minimal identifying subset of OpenRouter's documented
[current-key response](https://github.com/OpenRouterTeam/docs/blob/main/api_reference/limits.mdx#checking-your-limits), [current-key response](https://github.com/OpenRouterTeam/docs/blob/main/api_reference/limits.mdx#checking-your-limits),
not validation of the full account document. Other statuses and unrecognized not validation of the full account document. Other statuses and unrecognized
200 responses do not establish acceptance. 200 responses do not establish acceptance. A 200 response exceeding the bounded
recognition allowance reports `authentication_response_limit_exceeded`, leaving
credential acceptance unknown; see the [public reason](../../smoke_diagnostics.go).
Diagnostic HTTP evidence describes the last request observed at Promptkit's Diagnostic HTTP evidence describes the last request observed at Promptkit's
RoundTripper boundary and any response associated with it. It includes the RoundTripper boundary and any response associated with it. It includes the
@@ -87,12 +88,13 @@ transport for HTTP-only diagnostics.
## Request Body ## Request Body
The request body always contains `model` and `messages`. The execution The request body always contains `model` and `messages`. The execution
target's model takes precedence over the client's configured model, and one target's model must be available.
must be available.
Each ordinary message contains its `role` and string `content`. A Each ordinary message contains its `role` and string `content`. A
cache-controlled message instead uses a text content block containing `type`, cache-controlled message instead uses a text content block containing `type`,
`text`, and `cache_control`; an empty cache-control TTL is omitted. `text`, and `cache_control`; an empty cache-control TTL is omitted.
For a continuation, captured history precedes the new definition and appended
messages. Ancestry is local metadata and is not an upstream request field.
Promptkit sends only `developer`, `system`, `user`, and `assistant` roles and Promptkit sends only `developer`, `system`, `user`, and `assistant` roles and
does so without provider-specific translation. Tool and deprecated function does so without provider-specific translation. Tool and deprecated function
@@ -115,9 +117,10 @@ The client conditionally includes:
strict flag, and schema document. Plain JSON validation does not add an strict flag, and schema document. Plain JSON validation does not add an
object-only response constraint. object-only response constraint.
The engine resolves backend, profile, and request extra-parameter maps by Configured profile maps recursively overlay backend defaults, while a nonempty
whole-map replacement rather than key merging. The resulting effective map is runtime override replaces the complete configured map. The [format reference](../formats.md#defaults-and-overrides)
merged into the top-level body. Empty keys, values that cannot be encoded, and owns these resolution rules. The resulting effective map is merged into the
top-level body. Empty keys, values that cannot be encoded, and
collisions with these reserved fields are rejected before any provider call: collisions with these reserved fields are rejected before any provider call:
- `model` - `model`

View File

@@ -68,9 +68,8 @@ Preparation and validation never hold an active permit.
The manager's `Acquire` operation owns that active-permit acquisition and The manager's `Acquire` operation owns that active-permit acquisition and
returns an idempotent release function. The coordinator uses it around returns an idempotent release function. The coordinator uses it around
individual attempts. The earlier one-shot wrapper still delegates to the same individual attempts. There is one queue and one active limit per limited backend
operation for focused internal callers and tests, but engine assembly no pool.
longer uses it. There is one queue and one active limit per limited backend pool.
When all active permits are occupied, calls join a mutex-protected FIFO waiter When all active permits are occupied, calls join a mutex-protected FIFO waiter
list. Releasing a permit transfers it directly to the oldest remaining waiter list. Releasing a permit transfers it directly to the oldest remaining waiter
@@ -121,7 +120,7 @@ validation, bounded admission, idempotent release, context handling, and
unlimited admission. The unlimited admission. The
[client tests](../../internal/capacity/client_test.go) own peak enforcement, [client tests](../../internal/capacity/client_test.go) own peak enforcement,
FIFO transfer, canceled-waiter removal, grant/cancel races, independent pools, FIFO transfer, canceled-waiter removal, grant/cancel races, independent pools,
unlimited calls, passthrough behavior, and panic release. unlimited calls, and panic release using test-only one-shot calls.
The [runner tests](../../internal/usecase/runner_test.go) own ordinary early The [runner tests](../../internal/usecase/runner_test.go) own ordinary early
admission, lease lifetime, failure release, and shared initial/repair admission, lease lifetime, failure release, and shared initial/repair

View File

@@ -26,10 +26,9 @@ client uses its endpoint, credential metadata, generation fields, and extra
parameters. `BackendID` remains routing metadata parameters. `BackendID` remains routing metadata
for the generation boundary and is not mapped into the provider payload. for the generation boundary and is not mapped into the provider payload.
Construction trims and validates a nonempty configured base URL and clones any Construction clones any supplied `http.Client` so Promptkit can apply its
supplied `http.Client` so Promptkit can apply its timeout default without timeout default without mutating the caller's client. Generation preparation
mutating the caller's client. An empty configured base remains valid because a then:
resolved request target may supply the endpoint. Generation preparation then:
1. validates shared execution-setting invariants and the final selected base 1. validates shared execution-setting invariants and the final selected base
endpoint; endpoint;
@@ -42,6 +41,10 @@ resolved request target may supply the endpoint. Generation preparation then:
the complete-request failure remains a request failure; the complete-request failure remains a request failure;
6. resolves authentication and freezes the URL, body, and optional credential. 6. resolves authentication and freezes the URL, body, and optional credential.
Provider values are detached with trusted copying before encoding. Custom JSON
encoding methods and per-value encoding-failure diagnostics operate on those
private copies rather than retained prepared or diagnostic metadata.
Each execution of that prepared request creates a fresh HTTP request, body Each execution of that prepared request creates a fresh HTTP request, body
reader, and attempt timeout context. It closes the response body and returns reader, and attempt timeout context. It closes the response body and returns
the HTTP status, decoded response or error, reported token values and presence, the HTTP status, decoded response or error, reported token values and presence,
@@ -77,7 +80,8 @@ not retain headers, credentials, raw request or response bodies, or a redirect
trace. These are observations at Promptkit's RoundTripper boundary, not proof trace. These are observations at Promptkit's RoundTripper boundary, not proof
that bytes reached a peer or that a custom transport did no hidden dispatch. that bytes reached a peer or that a custom transport did no hidden dispatch.
Reachability and named authentication probes use the same configured built-in Reachability and named authentication probes use the same configured built-in
client but discard their response bodies after local recognition. The client. The OpenRouter adapter owns endpoint derivation, bounded account-response
recognition, and body closure; the smoke coordinator interprets its evidence. The
[integration contract](../integrations/openai-compatible-chat.md#explicit-smoke-diagnostics) [integration contract](../integrations/openai-compatible-chat.md#explicit-smoke-diagnostics)
owns their observable HTTP semantics. owns their observable HTTP semantics.
@@ -105,14 +109,13 @@ public ownership and redaction semantics belong to the
## Failure Categories ## Failure Categories
The package preserves distinct error identities for invalid client The package preserves distinct error identities for invalid generation
configuration, invalid generation requests, request execution failures, requests, request execution failures, non-success provider statuses, and
non-success provider statuses, and malformed successful responses. Provider malformed successful responses. Provider response bodies are never exposed in
response bodies are never exposed in raw form through non-success errors. raw form through non-success errors.
Invalid nonempty configured endpoints are configuration failures. A missing or A missing or invalid final selected endpoint is an invalid generation request
invalid final selected endpoint is an invalid generation request and is and is rejected before transport.
rejected before transport.
Authentication resolves a trimmed direct key before a trimmed configured Authentication resolves a trimmed direct key before a trimmed configured
environment value. Optional missing, empty, or whitespace-only sources do not environment value. Optional missing, empty, or whitespace-only sources do not
@@ -156,6 +159,13 @@ caller deadlines, generation deadlines, and whole-request client timeouts
retain `context.DeadlineExceeded`. The runner adds its generation category retain `context.DeadlineExceeded`. The runner adds its generation category
without discarding those identities or depending on HTTP status mapping. without discarding those identities or depending on HTTP status mapping.
When a transport failure carries a custom context cancellation cause, the
client also retains the standard context identity. It correlates that cause
with the returned failure rather than labeling unrelated transport errors from
context state alone. Cancellation while reading a successful response body
retains the original read failure alongside the context identity within the
redacting wrapper.
## Test Ownership ## Test Ownership
The The

View File

@@ -10,6 +10,13 @@ lifecycle instrumentation spans the facade and runner boundaries.
Prepared execution starts only after its private payload is atomically claimed; Prepared execution starts only after its private payload is atomically claimed;
the facade receives that invocation's recorder privately and emits the terminal the facade receives that invocation's recorder privately and emits the terminal
event after public error mapping. Failed claims and preparation emit nothing. event after public error mapping. Failed claims and preparation emit nothing.
The public projection retains private operation-owner metadata separately from
inspectable child causes. Terminal classification uses that owner, preserving
context precedence and generation-specific transport/provider details rather
than mistaking a collaborator's nested lookup error for the parent's operation.
Timeout attribution is finalized and the child scope released before terminal
delivery, so observer latency cannot change the already selected outcome.
Prepared-content notification is a separate opt-in callback, not an event.
The facade masks inherited recorder and prepared-claim state when entering The facade masks inherited recorder and prepared-claim state when entering
preparation or execution. Delegated calls from an injected collaborator retain preparation or execution. Delegated calls from an injected collaborator retain
@@ -23,6 +30,8 @@ callback. The caller supplies normalized failure metadata; the recorder never
parses an error string or receives provider bodies, credentials, headers, or parses an error string or receives provider bodies, credentials, headers, or
other content. Events are values with no mutable containers. Callback panics other content. Events are values with no mutable containers. Callback panics
propagate to the caller. propagate to the caller.
The immediate parent run ID is set before start delivery and is retained as
selection metadata changes. It describes ancestry, not implicit group membership.
Resolved prompt and profile IDs are the exact runtime identities selected for Resolved prompt and profile IDs are the exact runtime identities selected for
the invocation, including namespace qualification when supplied by an engine the invocation, including namespace qualification when supplied by an engine
view. view.
@@ -44,6 +53,13 @@ attempts above the first in a generation, and repair generations with at least
one dispatch. A terminal event is an aggregate and must not be added to its one dispatch. A terminal event is an aggregate and must not be added to its
attempt events. attempt events.
An execution group is a zero-ready synchronized collector of explicitly selected
events. It enrolls by run ID and aggregates each run's first terminal counters and
presence-aware usage exactly once. Snapshots copy members and totals under one
lock; terminal coverage and token completeness are distinct. Observer fan-out
copies the callback slice and invokes callbacks synchronously without holding a
group lock. No group history, registry, or automatic nested membership exists.
Smoke diagnostics mask inherited execution-recorder state while invoking Smoke diagnostics mask inherited execution-recorder state while invoking
generation or HTTP collaborators. Their private attempt evidence is assembled generation or HTTP collaborators. Their private attempt evidence is assembled
into a `SmokeTestReport`; it does not create ordinary run events or enter an into a `SmokeTestReport`; it does not create ordinary run events or enter an

View File

@@ -11,27 +11,28 @@ contributor workflow and validation.
| Component | Implemented responsibility | References | | Component | Implemented responsibility | References |
| --- | --- | --- | | --- | --- | --- |
| Root `promptkit` package | Provides the supported engine facade, ConfigKit source helpers, collection and resource mapping conveniences, injection options, public requests and results, prompt/profile/requirement inspection and identity inventories, concrete-request validation, explicit smoke diagnostics, observation values, opaque prepared-execution handles, value conversion, redacted formatting, and typed capacity and generation error mapping. | [Package GoDoc](../../doc.go), [observations](../../observations.go), [prepared execution](../../prepared_execution.go), [backend API](../../backends.go), [engine facade](../../engine.go), [smoke diagnostics](../../smoke_diagnostics.go), [identity listing](../../listing.go), [requirement facade](../../requirements.go) | | Root `promptkit` package | Provides the supported engine facade, ConfigKit source helpers, collection and resource mapping conveniences, injection options, public requests and results, resource and requirement inspection and identity inventories, concrete-request validation, explicit smoke diagnostics, observation values, opaque prepared-execution handles, value conversion, redacted formatting, and typed capacity and generation error mapping. | [Package GoDoc](../../doc.go), [observations](../../observations.go), [prepared execution](../../prepared_execution.go), [backend API](../../backends.go), [engine facade](../../engine.go), [smoke diagnostics](../../smoke_diagnostics.go), [identity listing](../../listing.go), [requirement facade](../../requirements.go) |
| `examples/go-library/prepare` | Demonstrates offline preparation from two explicitly selected synthetic collections, source-relative imported content, a scoped view, filtered listings, and operator inventory. It is not a public library package. | [Example program](../../examples/go-library/prepare/main.go), [collection inputs](../../examples/go-library/prepare/collections/support/collection.yaml) | | `examples/go-library/prepare` | Demonstrates offline preparation from two explicitly selected synthetic collections, source-relative imported content, a scoped view, filtered listings, and operator inventory. It is not a public library package. | [Example program](../../examples/go-library/prepare/main.go), [collection inputs](../../examples/go-library/prepare/collections/support/collection.yaml) |
| `examples/go-library/run` | Demonstrates an offline downstream consumer using a prompt file, programmatic profile, inline input, an injected deterministic model client, and `Run`. It is not a public library package. | [Example program](../../examples/go-library/run/main.go) | | `examples/go-library/run` | Demonstrates an offline downstream consumer using a prompt file, programmatic profile, inline input, an injected deterministic model client, and `Run`. It is not a public library package. | [Example program](../../examples/go-library/run/main.go) |
| `examples/go-library/observe` | Demonstrates offline built-in HTTP retry and synchronous metadata observation across structured proposal and assessment runs using embedded sources, two profiles on one finite backend, and an application decision. It is not a public library package. | [Example program](../../examples/go-library/observe/main.go) | | `examples/go-library/observe` | Demonstrates offline retry, observation, prepared notification, and separately budgeted coordinated proposal and assessment runs using two profiles on one finite backend. It is not a public library package. | [Example program](../../examples/go-library/observe/main.go) |
| `internal/backend` | Holds a defensive, immutable lookup of final backend definitions, including diagnostic metadata, plus capacity policies validated by the configuration adapter. | [Backend registry](../../internal/backend/registry.go) | | `internal/backend` | Holds a defensive, immutable lookup of final backend definitions, including diagnostic metadata, plus capacity policies validated by the configuration adapter. | [Backend registry](../../internal/backend/registry.go) |
| `internal/configuration` | Loads root and namespaced collection resources through ConfigKit; records effective field presence, validates final values including exact references and optional diagnostic metadata, captures indexed prompt definitions and source-relative message content, compiled templates, input metadata snapshots, sorted backend/profile/prompt identities, resource namespace sets, effective collection declarations, backend and presence-aware profile snapshots, and the unchanged read-only inspection view. | [Typed loader](../../internal/configuration/load.go), [identity listing](../../internal/configuration/listing.go), [internal sources](sources.md#typed-configuration-loading) | | `internal/configuration` | Supplies component rules and captures final shared settings and native inspection evidence into owned indexed resources, compiled templates, requirement metadata, identities, and execution profiles. | [Configuration capture](../../internal/configuration/load.go), [identity listing](../../internal/configuration/listing.go), [internal sources](sources.md#typed-configuration-loading) |
| `internal/settings` | Declares the shared public configuration data model exposed through root aliases under ADR 0002; contains no loading or execution logic. | [Settings declarations](../../internal/settings/settings.go), [ADR 0002](../adr/0002-shared-configuration-data-model.md) |
| `internal/capacity` | Owns engine-local bounded execution admission and FIFO model-generation permits for limited backend IDs, including cancellation-safe waiter removal and client wrapping. | [Internal capacity management](capacity.md) | | `internal/capacity` | Owns engine-local bounded execution admission and FIFO model-generation permits for limited backend IDs, including cancellation-safe waiter removal and client wrapping. | [Internal capacity management](capacity.md) |
| `internal/domain` | Defines internal framework values for requests, artifacts, prompt definitions, requirement inspection, profiles, execution targets, rendering, generation, validation, and source-neutral smoke diagnostics, and owns source-neutral invariants for shared execution settings, OpenAI-compatible base endpoints, session identifiers, output contracts, and resource-identity syntax and qualification. Source parsing, required fields, other source-specific normalization, defaulting, and boundary-specific error classification remain with their callers. | [Domain declarations](../../internal/domain/domain.go), [requirement values](../../internal/domain/requirements.go), [endpoint invariant](../../internal/domain/endpoint.go) | | `internal/domain` | Defines internal framework values for requests, immutable conversation exchanges, artifacts, prompt definitions, requirement inspection, profiles, execution targets, rendering, generation, validation, and source-neutral smoke diagnostics, and owns source-neutral invariants for shared execution settings, OpenAI-compatible base endpoints, session identifiers, output contracts, and resource-identity syntax and qualification. Source parsing, required fields, other source-specific normalization, defaulting, and boundary-specific error classification remain with their callers. | [Domain declarations](../../internal/domain/domain.go), [exchange storage](../../internal/domain/exchange.go), [requirement values](../../internal/domain/requirements.go), [endpoint invariant](../../internal/domain/endpoint.go) |
| `internal/engine` | Assembles caller-selected configuration into one snapshot, then constructs shared capacity, rendering, validation, generation, repair, the ordinary runner, and the smoke tester for each engine. Immutable namespace views share those collaborators and qualify caller-supplied execution, inspection, and diagnostic selections without rewriting captured references. Built-in engines provide the diagnostic HTTP collaborator; injected-client engines do not. It delegates identity listing directly to the captured configuration inventory rather than the runner. | [Engine assembly](../../internal/engine/engine.go), [internal sources](sources.md#typed-configuration-loading), [internal runner](runner.md) | | `internal/engine` | Assembles an already captured configuration into shared capacity, rendering, validation, generation, repair, runner, and smoke-test collaborators. Immutable namespace views share those collaborators and qualify selections without rewriting captured references. Built-in engines provide the diagnostic HTTP collaborator; injected-client engines do not. It delegates resource inspection and listing to the capture. | [Engine assembly](../../internal/engine/engine.go), [internal sources](sources.md#typed-configuration-loading), [internal runner](runner.md) |
| `internal/generation` | Coordinates frozen built-in HTTP attempts and one-shot injected calls through active backend permits, including bounded status retries, cancellable backoff, ordinary attempt observations, and invocation-local evidence for diagnostics. | [Internal generation coordination](generation.md) | | `internal/generation` | Coordinates frozen built-in HTTP attempts and one-shot injected calls through active backend permits, including bounded status retries, cancellable backoff, ordinary attempt observations, and invocation-local evidence for diagnostics. | [Internal generation coordination](generation.md) |
| `internal/defaults` | Defines application-neutral framework constants and constructs the default execution target. It contains no CLI, server, or inbound HTTP limits. | [Framework defaults](../../internal/defaults/defaults.go) | | `internal/defaults` | Defines application-neutral framework constants and constructs the default execution target. It contains no CLI, server, or inbound HTTP limits. | [Framework defaults](../../internal/defaults/defaults.go) |
| `internal/filecatalog` | Provides deterministic YAML discovery and path helpers for operating-system filesystems and `fs.FS` sources. | [File catalog](../../internal/filecatalog/catalog.go) | | `internal/filecatalog` | Provides deterministic YAML discovery and path helpers for operating-system filesystems and `fs.FS` sources. | [File catalog](../../internal/filecatalog/catalog.go) |
| `internal/jsonvalue` | Validates and deeply copies bounded JSON-compatible extra-parameter and prepared-schema trees while preserving supported concrete value types and rejecting cycles or excessive depth and work. | [JSON values](../../internal/jsonvalue/jsonvalue.go) | | `internal/jsonvalue` | Validates and deeply copies bounded JSON-compatible provider values, and detaches previously accepted values without repeating validation. | [JSON values](../../internal/jsonvalue/jsonvalue.go) |
| `internal/promptdef` | Provides a narrow captured-prompt lookup contract, not-found identity, and execution-ready definition hash. | [Prompt lookup](../../internal/promptdef/repository.go), [framework formats](../formats.md) | | `internal/promptdef` | Provides a narrow captured-prompt lookup contract, not-found identity, and execution-ready definition hash. | [Prompt lookup](../../internal/promptdef/repository.go), [framework formats](../formats.md) |
| `internal/profile` | Provides a narrow captured-profile lookup contract and not-found identity. | [Profile lookup](../../internal/profile/repository.go), [framework formats](../formats.md) | | `internal/profile` | Provides a narrow captured-profile lookup contract and not-found identity. | [Profile lookup](../../internal/profile/repository.go), [framework formats](../formats.md) |
| `internal/prompt` | Renders prompt messages from Go templates with artifact, variable, session, and cache-control data, and conservatively enumerates dependencies from captured template syntax. | [Go-template renderer](../../internal/prompt/go_renderer.go), [requirement discovery](../../internal/prompt/requirements.go) | | `internal/prompt` | Renders prompt messages from Go templates with artifact, variable, session, and cache-control data, and conservatively enumerates dependencies from captured template syntax. | [Go-template renderer](../../internal/prompt/go_renderer.go), [requirement discovery](../../internal/prompt/requirements.go) |
| `internal/artifact` | Resolves ordinary inline and unrestricted caller-selected file references into copied artifacts with metadata and hashes. | [Internal sources and validation](sources.md) | | `internal/artifact` | Resolves ordinary inline and unrestricted caller-selected file references into copied artifacts with metadata and hashes. | [Internal sources and validation](sources.md) |
| `internal/validate` | Validates basic, JSON, and JSON Schema output using operating-system filesystem or `fs.FS` schema sources and creates operation-local validation plans with canonical contained schema resources. | [Framework formats](../formats.md#schemas), [internal sources and validation](sources.md) | | `internal/validate` | Creates operation-local validation plans from inline schema documents or bound filesystem resources and validates basic, JSON, and JSON Schema output. | [Framework formats](../formats.md#schemas), [internal sources and validation](sources.md) |
| `internal/llm` | Defines the internal generation boundary and implements outbound OpenAI-compatible chat requests from resolved execution targets, including bounded response decoding, authentication, deadline handling, per-invocation HTTP evidence and built-in diagnostic probes, and ownership of the OpenAI-compatible reserved request-field policy. | [Internal model client](llm.md) | | `internal/llm` | Defines the internal generation boundary and implements outbound OpenAI-compatible chat requests from resolved execution targets, including bounded response decoding, authentication, deadline handling, per-invocation HTTP evidence and built-in diagnostic probes, and ownership of the OpenAI-compatible reserved request-field policy. | [Internal model client](llm.md) |
| `internal/observation` | Owns invocation-local run IDs, presence-aware usage and phase accounting, and synchronous metadata event delivery for ordinary and claimed prepared runs and generation attempts; smoke diagnostics keep their private evidence outside ordinary observer events. | [Internal execution observation](observation.md) | | `internal/observation` | Owns invocation-local run IDs, presence-aware usage and phase accounting, synchronous metadata event delivery and composition, and explicit terminal-based group aggregation; smoke diagnostics keep their private evidence outside ordinary observer events. | [Internal execution observation](observation.md) |
| `internal/usecase` | Selects captured prompts and profiles, enumerates requirements from captured prompt syntax, applies request overrides and completion checks, coordinates ordinary and single-claim prepared execution across rendering, artifact loading, operation-local validation plans, generation, capacity, and bounded repair, and runs explicit smoke diagnostics with private cases, shared generation/capacity, and optional built-in HTTP probes. | [Internal runner](runner.md), [requirement inspection](../../internal/usecase/requirements_inspection.go), [prepared-execution implementation](../../internal/usecase/prepared_execution.go) | | `internal/usecase` | Selects captured prompts and profiles, enumerates requirements from captured prompt syntax, applies request overrides and completion checks, coordinates ordinary and single-claim prepared execution across rendering, artifact loading, operation-local validation plans, generation, capacity, and bounded repair, advances a single conversation branch, and runs explicit smoke diagnostics with private cases, shared generation/capacity, and optional built-in HTTP probes. | [Internal runner](runner.md), [requirement inspection](../../internal/usecase/requirements_inspection.go), [prepared-execution implementation](../../internal/usecase/prepared_execution.go) |
The internal engine assembles these internal components without exposing their The internal engine assembles these internal components without exposing their
representations. Consumers depend only on the root facade. representations. Consumers depend only on the root facade.

View File

@@ -30,6 +30,29 @@ internal callers and tests.
Each invocation carries its state in request, prepared-run, and result values. Each invocation carries its state in request, prepared-run, and result values.
The runner has no durable run or session store. The runner has no durable run or session store.
Request output-contract replacements carry an optional repair budget. The
runner resolves that budget from the replacement's validation mode before
building the concrete validation plan; prepared execution retains the resolved
integer. Configured budgets are already resolved by the configuration adapter.
Captured exchanges are immutable parent-linked domain values. Each holds only
its new rendered suffix, final assistant message, and run identities; flattening
iteratively fills a single detached message buffer for preparation or projects
directly into public inspection values. Preparation
prepends this history to newly rendered definition messages and appended caller
messages before computing the rendered hash. Follow-up selection and credentials
remain those of the new request. Prepared execution retains capture and parent
references in its single-claim payload, not its durable details; discard clears
the payload. Completed runs capture the exact final output, including exhausted
structural repair, while operational errors produce no exchange.
The sequential coordinator uses ordinary facade runs and retains the current
immutable exchange independently of callback-owned values. Each run releases its
logical admission lease and finishes observation before application callback
delivery. Context checks preserve a completed but undelivered result only when
cancellation is observed before invoking the callback. Coordinator budgets count
started runs, not transport attempts or structural repairs.
## Shared Prompt Selection ## Shared Prompt Selection
The internal engine applies each namespace view to slashless caller selections The internal engine applies each namespace view to slashless caller selections
@@ -42,14 +65,13 @@ Ordinary preparation selects the full prompt definition and construction-time
hash. It retains its early request-ID check before direct-session hash. It retains its early request-ID check before direct-session
normalization. Referenced message content and static template syntax were normalization. Referenced message content and static template syntax were
captured when the engine was built. captured when the engine was built.
Rendering uses those captured parses, cloning them per call to bind request
input helpers; it does not parse templates during execution.
`InspectPrompt` selects captured metadata through an optional repository Resource inspection bypasses the runner: the internal engine selects a captured
capability, falling back to the full definition and existing hash resolution configuration resource, and the facade detaches its values for the caller.
for internal repositories without it. It returns owned metadata without Prompt resource inspection includes captured message text without rendering or
rendering templates or touching profile, artifact, schema, renderer, validator, touching profile, artifact, schema, admission, or model collaborators.
admission, or model collaborators. The root
[`Engine.InspectPrompt`](../../engine.go) GoDoc owns the public operation's
exact contract.
`InspectRequirements` uses the same metadata selection to resolve the `InspectRequirements` uses the same metadata selection to resolve the
request-profile/default identity, then obtains the captured requirement request-profile/default identity, then obtains the captured requirement
@@ -57,24 +79,18 @@ snapshot by the selected prompt ID. The analyzer receives only
those identity fields rather than an execution definition. It analyzes the those identity fields rather than an execution definition. It analyzes the
snapshot with the supplied caller variables and normalized direct-session snapshot with the supplied caller variables and normalized direct-session
bypass; it does not resolve target readiness or enter the preparation bypass; it does not resolve target readiness or enter the preparation
pipeline. Missing snapshot capability is a prompt-load failure rather than a pipeline. An unavailable captured snapshot is a prompt-load failure. The root
source fallback. The root
[`Engine.InspectRequirements`](../../requirements.go) GoDoc owns the public [`Engine.InspectRequirements`](../../requirements.go) GoDoc owns the public
operation's exact contract. operation's exact contract.
## Shared Profile Selection ## Shared Profile Selection
The runner uses one profile-selection and target-resolution boundary for The runner selects a request profile or a prompt default and uses its captured,
ordinary preparation and exact profile inspection. Preparation first selects a presence-aware target constructed from final profile and backend configuration.
request profile or a prompt default; inspection begins with its required Request overrides apply afterward; preparation checks endpoint and model
explicit profile ID. Both use the captured, presence-aware target constructed completeness only after those overrides. Resource profile inspection bypasses
from the final effective profile and backend configuration. Request overrides this path and reports the profile before backend completion. Credential
apply afterward; preparation checks endpoint and model completeness only after availability is checked when execution preparation requires it.
those overrides. Inspection can report a partial profile without requiring a
runnable target. Neither path checks credential availability or performs prompt,
artifact, schema, rendering, admission, or model-client work. The root
[`Engine.InspectProfile`](../../engine.go) GoDoc owns the public operation's
exact contract.
## Smoke-Test Diagnostics ## Smoke-Test Diagnostics
@@ -115,12 +131,12 @@ performs only the work needed to validate routing and admission:
1. validate the required prompt selection and normalize any direct session ID; 1. validate the required prompt selection and normalize any direct session ID;
2. select the captured prompt definition and hash; 2. select the captured prompt definition and hash;
3. select the request profile or the prompt's default profile; 3. resolve and validate the effective output contract and select its applicable
4. retain the profile's selected backend identity, when present; bound file without loading its schema;
5. apply explicit request overrides to its captured backend-completed target; 4. select the request profile or the prompt's default profile;
6. validate endpoint, model, numeric overrides, and credential requirements; 5. retain the profile's selected backend identity, when present;
7. resolve and validate the effective output contract without loading its 6. apply explicit request overrides to its captured backend-completed target;
schema; and 7. validate endpoint, model, numeric overrides, and credential requirements; and
8. retain the definition, source identities, effective settings, output 8. retain the definition, source identities, effective settings, output
contract, and preparation start time in invocation-local state. contract, and preparation start time in invocation-local state.
@@ -131,9 +147,9 @@ profile, or backend:
schema metadata from it when required; schema metadata from it when required;
2. load and hash input artifacts; 2. load and hash input artifacts;
3. render messages and the prompt-defined session; 3. render messages and the prompt-defined session;
4. apply any direct session ID, then append already-normalized request messages 4. apply any direct session ID, then compose captured history, rendered
after the rendered definition messages; definition messages, and already-normalized request messages in that order;
5. hash the effective rendered prompt; and 5. hash the effective rendered prompt and its ordered message content; and
6. construct the prepared value and preparation timing. 6. construct the prepared value and preparation timing.
`Prepare` runs both phases consecutively and never performs capacity admission. `Prepare` runs both phases consecutively and never performs capacity admission.
@@ -142,11 +158,10 @@ one execution-precedence and error-ordering implementation while allowing a
full backend pool to reject work before expensive schema, artifact, and full backend pool to reject work before expensive schema, artifact, and
rendering operations. `Prepare` discards the plan after returning its public rendering operations. `Prepare` discards the plan after returning its public
metadata. `Run` retains the plan through initial and repaired-output validation metadata. `Run` retains the plan through initial and repaired-output validation
and discards it when the operation ends. Prepared execution transfers the and discards it when the operation ends. Prepared execution retains one
completed preparation's owned message, input-hash, and copied-schema values immutable, redacted metadata snapshot shared by execution and internal details.
into its private payload, while deep-copying mutable provider extras and The facade detaches mutable public details. Claim completion or discard clears
creating an independent details snapshot. Claim completion or discard clears execution-only payload references without modifying the retained metadata.
the execution payload without removing the details snapshot.
The root facade's `Engine.ValidateRequest` uses the same complete preparation The root facade's `Engine.ValidateRequest` uses the same complete preparation
path as `Prepare` and discards the resulting internal prepared value without path as `Prepare` and discards the resulting internal prepared value without
@@ -175,8 +190,14 @@ sequence. The rendered-prompt hash uses a versioned, length-framed SHA-256
encoding of that session, every role and content value, and cache-control encoding of that session, every role and content value, and cache-control
presence and values; its hexadecimal value is opaque. presence and values; its hexadecimal value is opaque.
The separate message-content hash is computed once after that same composition.
It streams a versioned prefix, message count, and length-framed role and text
bytes into SHA-256, excluding session and cache metadata. Fixed reference
vectors protect the selected v1 scheme across refactors; the public
[`PreparedRun` GoDoc](../../types.go) owns its compatibility contract.
The configuration snapshot is read-only after engine construction. Concurrent The configuration snapshot is read-only after engine construction. Concurrent
`Prepare` and `Run` calls copy their resolved targets and keep all invocation `Prepare` and `Run` calls use detached captured targets and keep invocation
state local. state local.
## Run Flow ## Run Flow
@@ -199,6 +220,16 @@ mapping and after the runner's deferred admission release. Preparation and
prepared execution follows the same terminal mapping after a successful claim; prepared execution follows the same terminal mapping after a successful claim;
invalid or discarded handles do not acquire an observer. invalid or discarded handles do not acquire an observer.
The facade establishes an optional operation scope for ordinary runs before
observation and preparation. A claimed prepared execution establishes its own
scope from the retained duration; preparation and idle time do not consume it.
The shared scope finalizer marks only a context failure caused by that scope's
own deadline, preserving its underlying operation and cause. The scope is
released before terminal delivery. At the shared prepared execution boundary,
the runner conditionally invokes the adapter for the actual prepared snapshot
while admission is held, with context checks before and after notification.
The facade adapter projects a detached public value only when selected.
Validation preparation and execution honor cancellation at every Validation preparation and execution honor cancellation at every
Promptkit-controlled boundary and do not publish a partial plan or result. Promptkit-controlled boundary and do not publish a partial plan or result.
Schema reads are bounded and context-checked between chunks; JSON decoding, Schema reads are bounded and context-checked between chunks; JSON decoding,

View File

@@ -10,39 +10,24 @@ source semantics. Internal packages are not consumer APIs.
## Typed Configuration Loading ## Typed Configuration Loading
`internal/configuration` is the single adapter from ConfigKit's final The root facade composes Promptkit rules into a ConfigKit definition. An
effective configuration to execution-owned definitions. It performs one application-owned load or the standalone constructor produces final settings
layered load, captures effective presence (including explicit null), builds at `/promptkit`; `internal/configuration` captures execution-owned resources
detached normalized interpretations during validation, and validates all effective from those settings and a native inspection view without another load.
root and collection backend, profile, and prompt values. Replaced lower-layer `internal/settings` supplies the shared data declarations exposed through
domain values are not separately validated. It indexes root resources under root aliases. The adapter records normalization and validates effective values,
their unqualified IDs and collection resources under their full then uses one metadata snapshot for presence and source origins. Runtime/domain
`namespace/local-key` IDs while retaining ConfigKit's effective tree in its packages do not depend on ConfigKit.
authored shape. It builds an immutable backend index and presence-aware profile
targets, applying final backend defaults only after profile inheritance. Runtime
lookups use those final indexes and exact stored references; they do not rewrite
or resolve resources from ConfigKit source history. The runtime/domain packages
do not depend on ConfigKit.
ConfigKit owns source headers, imports, composition, and source evidence. Capture detaches known settings, builds indexed definitions and profile targets,
Promptkit does not preparse contributions: it validates the final effective and binds source-relative resources from the effective field's source context.
values once, captures the indexed identities and their original resource It retains a frozen native focused view and copied diagnostics, not the mutable
contexts, and retains ConfigKit's composed result for explicit operator inspection. application result. Configuration inspection delegates reporting, redaction,
Accepted non-main source headers are consumed during composition; the main history and comparison to ConfigKit; explicit queries and metadata retain
source declaration remains available in that inspection view. Collection whole-application paths and evidence. Runtime never derives values from history.
release labels are descriptive metadata, not selection inputs. The The [format reference](../formats.md#source-headers-collections-and-resource-ids)
[format reference](../formats.md#source-headers-collections-and-resource-ids) owns the configuration contract and the [consumer guide](../consumers/pkg-promptkit.md#inspect-the-captured-configuration)
owns the exact source-header and collection contracts. describes reporting scope.
The `ConfigurationInspection.Diff` facade method compares two retained results
through a private wrapper and a narrow `internal/configuration.Diff` bridge.
The wrapper keeps `Result[document]` inside the adapter; the bridge delegates
once to ConfigKit v1.0.0 and returns its public difference records. Comparison
uses only frozen effective values and does not reopen sources, resolve runtime
credentials, or enter the execution path. ConfigKit applies sensitivity from
both results; reports may still contain ordinary application data. See the
[consumer comparison guide](../consumers/pkg-promptkit.md#inspect-the-captured-configuration)
and ConfigKit's [v1.0.0 comparison contract](https://forge.maximumdirect.net/base-layer/configkit/src/tag/v1.0.0/docs/config.md#effective-configuration-diffs).
After resource materialization and effective-profile snapshot construction, the After resource materialization and effective-profile snapshot construction, the
adapter captures lexically ordered identity inventories for backends, profiles, adapter captures lexically ordered identity inventories for backends, profiles,
@@ -53,13 +38,15 @@ captured values without consulting ConfigKit or copying resource definitions.
The root `DirectoryInputs` and `FilesystemInputs` helpers discover sorted YAML The root `DirectoryInputs` and `FilesystemInputs` helpers discover sorted YAML
files under an explicit root, provide resource context, and leave imports files under an explicit root, provide resource context, and leave imports
disabled. Each returned slice belongs to one strict-composition group; file disabled. Each returned slice belongs to one strict-composition group; file
order does not confer priority. Caller underlays are composed in their given order does not confer priority. The standalone facade prepends a dedicated
embedded OpenRouter underlay unless disabled, without modifying caller-owned slices.
Caller underlays are composed in their given
order, followed by the optional main input. The order, followed by the optional main input. The
[configuration tests](../../internal/configuration/load_test.go) exercise the [configuration tests](../../internal/configuration/load_test.go) exercise the
real loader and final-value handoff. real loader and final-value handoff.
The public `CollectionInput` helper mounts one caller-selected source at The public `CollectionInput` helper mounts one caller-selected source at
`collections/<namespace>` and enables ConfigKit imports for that input. The `/promptkit/collections/<namespace>` and enables ConfigKit imports for that input. The
helper does not read the source; ConfigKit retains responsibility for loading, helper does not read the source; ConfigKit retains responsibility for loading,
composition, imports, and source-local resource context. Imported or inherited composition, imports, and source-local resource context. Imported or inherited
`content_file` values continue resolving relative to the source that contributed `content_file` values continue resolving relative to the source that contributed
@@ -82,23 +69,27 @@ declaration; absent and null fields remain distinguishable from explicit
strings, including empty strings. It closes content-file handles before strings, including empty strings. It closes content-file handles before
returning and retains no source capability for public inspection. returning and retains no source capability for public inspection.
Prompt materialization shares a lazy metadata snapshot across root and collection Prompt materialization shares one metadata snapshot across root and collection
resources during construction, avoiding a full metadata copy per collection. resources during construction, avoiding a full metadata copy per collection.
Input metadata, message content and templates, and output contracts are captured
separately before the definition, inspection, and hash are assembled. Schema
root and source-relative file bindings capture the effective field context
without opening schema files; capabilities are excluded from the hash. Inline
schema values and file-reference strings participate in the definition hash.
Normalization and defaults are applied to execution-owned interpretations, Normalization and defaults are recorded at individual resolved field paths,
including message roles and cache controls. ConfigKit retains the composed including message roles and cache controls. List indexes identify fields for
values and origins, rather than synthetic Promptkit normalization events. List presence, validation, sensitivity, and resource context; full IDs select
indexes identify fields for presence, validation, sensitivity, and resource captured runtime indexes. Unchanged content-file fields retain their origins.
context; full IDs select captured runtime indexes. Recording a sequence-wide Backend-to-profile and request-time completion remain separate runtime work.
normalization would replace origins of unchanged content-file fields.
The runner selects full definitions from this index for preparation and uses The runner selects full definitions from this index for preparation and retains
captured prompt metadata for `InspectPrompt` and `InspectRequirements`; the lightweight prompt metadata for `InspectRequirements`; the requirement snapshot
internal requirement snapshot lookup uses the selected prompt ID alone. lookup uses the selected prompt ID alone. Separate configuration-owned resource
These paths do not rescan configuration or reopen message files. `InspectPrompt` inspection projects the resolved record and captured messages without rendering
returns declared metadata without rendering or resolving profile, input, or or resolving profile, input, or schema resources. These paths do not rescan
schema resources, and its public shape and prompt hashes do not include the configuration or reopen message files. Presence metadata does not change the
captured presence state. Runtime rendering still checks request variables and opaque definition hash. Runtime rendering still checks request variables and
input helpers. The [prompt adapter tests](../../internal/configuration/prompts_test.go) input helpers. The [prompt adapter tests](../../internal/configuration/prompts_test.go)
cover resource origins, context containment, capture, and selection. cover resource origins, context containment, capture, and selection.
@@ -146,11 +137,15 @@ change rendering behavior.
ConfigKit resolves profile inheritance before the handoff. Within a collection, ConfigKit resolves profile inheritance before the handoff. Within a collection,
`extends: {id: parent}` names a sibling local ID; it is not a cross-collection reference. `extends: {id: parent}` names a sibling local ID; it is not a cross-collection reference.
The configuration adapter validates final effective resource references and The configuration adapter validates final effective resource references and
field values, then builds target field values, then captures backend definitions and profile target snapshots
snapshots with backend defaults and presence/suppression state. A profile can with backend defaults and outbound-control presence. Profiles without provider
remain partial until runtime overrides complete it. `InspectProfile` selects overrides share capture-owned immutable backend parameter maps; target lookups
that captured target without prompt, artifact, schema, credential lookup, or still detach mutable values before runtime use. Null profile endpoint,
provider work. Preparation applies runtime overrides and then checks target credential source, and whole provider map permit backend fallback after
composition, subject to the required-credential rule. A profile can remain
partial until runtime overrides complete it. Resource inspection selects indexed
resolved backend and profile records without execution completion or prompt,
artifact, schema, credential lookup, or provider work. Preparation applies runtime overrides and then checks target
completeness. Identity-only profile selection can identify a partial captured completeness. Identity-only profile selection can identify a partial captured
profile without copying its target or requiring readiness. The profile without copying its target or requiring readiness. The
[target tests](../../internal/configuration/targets_test.go) cover effective [target tests](../../internal/configuration/targets_test.go) cover effective
@@ -208,20 +203,24 @@ buffer-to-string copy. The
## Schemas And Output Validation ## Schemas And Output Validation
`internal/validate` provides validators backed by an operating-system `internal/validate` prepares a source-neutral input containing an output contract
filesystem or an `fs.FS`. Invalid generated content is returned as a validation and an optional bound `fs.FS` file. Invalid generated content is returned as a validation
result; inability to load, register, or compile a schema is an operational result; inability to load, register, or compile a schema is an operational
error. Basic validation checks body emptiness on the byte slice directly error. Basic validation checks body emptiness on the byte slice directly
instead of first making a response-sized string copy. instead of first making a response-sized string copy.
Every preparation operation creates one operation-local validation plan. None, Every preparation operation creates one operation-local validation plan. None,
basic, and JSON modes retain the effective output contract without source basic, and JSON modes retain the effective output contract without source
access. JSON Schema mode loads the root document once, resolves and compiles access. JSON Schema mode uses the captured inline document or loads the bound
each transitive reference, and retains the compiled validator. Schema compiler root document once, compiles its declared
resources use canonical escaped file or private-scheme URLs; loaders decode schema locations, and retains the compiled root validator. External resources
their paths once and enforce the configured source boundary. The are rejected without file or network loading; in-document references remain
supported. Bound files use canonical escaped private-scheme URLs; inline values
use a stable private retrieval URL. The
provider-facing structured-output metadata uses the root document captured by provider-facing structured-output metadata uses the root document captured by
the same plan. the same plan. Its detached schema document is not revalidated against the
runtime provider-value copier's traversal limits. Smoke inference uses a
separate JSON-only prepared plan before checking the prepared schema plan.
Schema preparation and execution remain synchronous. Promptkit checks Schema preparation and execution remain synchronous. Promptkit checks
cancellation before and after source resolution, JSON decoding, compilation, cancellation before and after source resolution, JSON decoding, compilation,

View File

@@ -26,67 +26,60 @@ License and notices: [CC BY-SA 4.0](licenses/CC-BY-SA-4.0.txt) |
## ConfigKit ## ConfigKit
Selected module: `go.maximumdirect.net/configkit` at `v1.0.0`. Selected module: `go.maximumdirect.net/configkit` at `v1.4.0`.
Copyright (C) 2026 Eric Rakestraw Copyright (C) 2026 Eric Rakestraw
SPDX-License-Identifier: LGPL-3.0-or-later SPDX-License-Identifier: LGPL-3.0-or-later
License and notices: [LGPL 3.0](../../LICENSE) | License and notices: [LGPL 3.0](../../LICENSE) |
[COPYRIGHT](https://forge.maximumdirect.net/base-layer/configkit/src/commit/c88aac990b23f48a31f9046b830d4d0954f57f53/COPYRIGHT) | [COPYRIGHT](https://forge.maximumdirect.net/base-layer/configkit/src/commit/4a7ae664a6287d71cba20f59c7c5b37b6f05b0a3/COPYRIGHT) |
[LICENSE](https://forge.maximumdirect.net/base-layer/configkit/src/commit/c88aac990b23f48a31f9046b830d4d0954f57f53/LICENSE). [LICENSE](https://forge.maximumdirect.net/base-layer/configkit/src/commit/4a7ae664a6287d71cba20f59c7c5b37b6f05b0a3/LICENSE).
[Source](https://forge.maximumdirect.net/base-layer/configkit/src/commit/c88aac990b23f48a31f9046b830d4d0954f57f53). [Source](https://forge.maximumdirect.net/base-layer/configkit/src/commit/4a7ae664a6287d71cba20f59c7c5b37b6f05b0a3).
## ReleaseKit
Selected module: `go.maximumdirect.net/releasekit` at `v0.10.0` (maintainer tool).
Copyright (C) 2026 Eric Rakestraw
SPDX-License-Identifier: LGPL-3.0-or-later
License and notices: [LGPL 3.0](../../LICENSE) |
[COPYRIGHT](https://forge.maximumdirect.net/base-layer/releasekit/src/commit/82bac1a3ab935beddd4b8789d1e07f52135d4243/COPYRIGHT) |
[LICENSE](https://forge.maximumdirect.net/base-layer/releasekit/src/commit/82bac1a3ab935beddd4b8789d1e07f52135d4243/LICENSE).
[Source](https://forge.maximumdirect.net/base-layer/releasekit/src/commit/82bac1a3ab935beddd4b8789d1e07f52135d4243).
## jsonschema ## jsonschema
Selected module: `github.com/santhosh-tekuri/jsonschema/v6` at `v6.0.2`. Selected module: `github.com/santhosh-tekuri/jsonschema/v6` at `v6.0.3`.
SPDX-License-Identifier: Apache-2.0 SPDX-License-Identifier: Apache-2.0
License and notices: [Apache 2.0](licenses/Apache-2.0.txt) | License and notices: [Apache 2.0](licenses/Apache-2.0.txt) |
[LICENSE](https://github.com/santhosh-tekuri/jsonschema/blob/29cbed948d24a04700eb94436416b18a07953b71/LICENSE). [LICENSE](https://github.com/santhosh-tekuri/jsonschema/blob/b0fc661f4939578bc429f408c18202573dbafcc4/LICENSE).
[Source](https://github.com/santhosh-tekuri/jsonschema/tree/29cbed948d24a04700eb94436416b18a07953b71). [Source](https://github.com/santhosh-tekuri/jsonschema/tree/b0fc661f4939578bc429f408c18202573dbafcc4).
## Go Text ## Go Text
Selected module: `golang.org/x/text` at `v0.14.0` (transitive dependency). Selected module: `golang.org/x/text` at `v0.44.0` (transitive dependency).
Copyright (c) 2009 The Go Authors. All rights reserved. Copyright 2009 The Go Authors.
SPDX-License-Identifier: BSD-3-Clause SPDX-License-Identifier: BSD-3-Clause
License and notices: [BSD 3-Clause](licenses/BSD-3-Clause.txt) | License and notices: [BSD 3-Clause](licenses/BSD-3-Clause.txt) |
[LICENSE](https://github.com/golang/text/blob/6c97a165dd661335ff7bce6104a008558123c353/LICENSE) | [LICENSE](https://github.com/golang/text/blob/6070ac32abcdd1670599eeb3ee4d513c8184a513/LICENSE) |
[PATENTS](https://github.com/golang/text/blob/6c97a165dd661335ff7bce6104a008558123c353/PATENTS). [PATENTS](https://github.com/golang/text/blob/6070ac32abcdd1670599eeb3ee4d513c8184a513/PATENTS).
[Source](https://github.com/golang/text/tree/6c97a165dd661335ff7bce6104a008558123c353). [Source](https://github.com/golang/text/tree/6070ac32abcdd1670599eeb3ee4d513c8184a513).
## Go YAML v3
Selected module: `gopkg.in/yaml.v3` at `v3.0.1` (transitive dependency).
Copyright (c) 2006-2010 Kirill Simonov
Copyright (c) 2006-2011 Kirill Simonov
Copyright (c) 2011-2019 Canonical Ltd
Copyright 2011-2016 Canonical Ltd.
SPDX-License-Identifier: MIT AND Apache-2.0
License and notices: [MIT](licenses/MIT.txt) |
[Apache 2.0](licenses/Apache-2.0.txt) |
[LICENSE](https://github.com/go-yaml/yaml/blob/f6f7691b1fdeb513f56608cd2c32c51f8194bf51/LICENSE) |
[NOTICE](https://github.com/go-yaml/yaml/blob/f6f7691b1fdeb513f56608cd2c32c51f8194bf51/NOTICE).
[Source](https://github.com/go-yaml/yaml/tree/f6f7691b1fdeb513f56608cd2c32c51f8194bf51).
## Go YAML v4 ## Go YAML v4
Selected module: `go.yaml.in/yaml/v4` at `v4.0.0-rc.6` (transitive dependency). Selected module: `go.yaml.in/yaml/v4` at `v4.0.0`.
Copyright 2006-2010 Kirill Simonov Copyright 2006-2010 Kirill Simonov
@@ -98,7 +91,34 @@ SPDX-License-Identifier: MIT AND Apache-2.0
License and notices: [MIT](licenses/MIT.txt) | License and notices: [MIT](licenses/MIT.txt) |
[Apache 2.0](licenses/Apache-2.0.txt) | [Apache 2.0](licenses/Apache-2.0.txt) |
[LICENSE](https://github.com/yaml/go-yaml/blob/1c17b9cee72bf81ae73a8be50fbd7dd3a9985125/LICENSE) | [LICENSE](https://github.com/yaml/go-yaml/blob/f2f9563503f11b7a60da0c532175dcc693ce8d5a/LICENSE) |
[NOTICE](https://github.com/yaml/go-yaml/blob/1c17b9cee72bf81ae73a8be50fbd7dd3a9985125/NOTICE). [NOTICE](https://github.com/yaml/go-yaml/blob/f2f9563503f11b7a60da0c532175dcc693ce8d5a/NOTICE).
[Source](https://github.com/yaml/go-yaml/tree/1c17b9cee72bf81ae73a8be50fbd7dd3a9985125). [Source](https://github.com/yaml/go-yaml/tree/f2f9563503f11b7a60da0c532175dcc693ce8d5a).
## Goldmark
Selected module: `github.com/yuin/goldmark/v2` at `v2.1.16` (maintainer-tool dependency).
Copyright (c) 2019 Yusuke Inuzuka
SPDX-License-Identifier: MIT
License and notices: [MIT](licenses/MIT.txt) |
[LICENSE](https://github.com/yuin/goldmark/blob/c3f939122ba4dd6f80e05043ad195b82e77fcb41/LICENSE).
[Source](https://github.com/yuin/goldmark/tree/c3f939122ba4dd6f80e05043ad195b82e77fcb41).
## Go Modules
Selected module: `golang.org/x/mod` at `v0.43.0` (maintainer-tool dependency).
Copyright 2009 The Go Authors.
SPDX-License-Identifier: BSD-3-Clause
License and notices: [BSD 3-Clause](licenses/BSD-3-Clause.txt) |
[LICENSE](https://go.googlesource.com/mod/+/d4f320b9266480df5ed3d791e0c68e64c58c5dde/LICENSE) |
[PATENTS](https://go.googlesource.com/mod/+/d4f320b9266480df5ed3d791e0c68e64c58c5dde/PATENTS).
[Source](https://go.googlesource.com/mod/+/d4f320b9266480df5ed3d791e0c68e64c58c5dde).

View File

@@ -2,7 +2,7 @@
**Base Policy Version:** 1.3 **Base Policy Version:** 1.3
**Last Updated**: 2026-09-27 **Last Updated**: 2026-11-21
**Upstream Source:** [architecture.md](https://gitea.maximumdirect.net/eric/policy/src/commit/d84a38ebab26f048cafbe59a2246700c35d29843/docs/policy/architecture.md) **Upstream Source:** [architecture.md](https://gitea.maximumdirect.net/eric/policy/src/commit/d84a38ebab26f048cafbe59a2246700c35d29843/docs/policy/architecture.md)
@@ -51,9 +51,10 @@ The [internal component overview](../internal/overview.md) owns the inventory
of implemented packages and their concrete responsibilities. This policy owns of implemented packages and their concrete responsibilities. This policy owns
the boundaries and dependency rules those components must preserve. the boundaries and dependency rules those components must preserve.
The internal engine composes caller-selected ConfigKit inputs once, then The facade either registers Promptkit rules in the application's ConfigKit load
assembles the captured definitions, capacity manager, renderer, validator, or performs one standalone load, then passes captured configuration to the
outbound client, and use-case runner. The configuration adapter interprets and internal engine. The internal engine assembles the captured definitions,
capacity manager, renderer, validator, outbound client, and use-case runner. The configuration adapter interprets and
validates only the final effective values, including presence, null, and validates only the final effective values, including presence, null, and
resource context. It retains source history for explicit operator inspection, resource context. It retains source history for explicit operator inspection,
not for validation or a second resolution pass. The root facade translates not for validation or a second resolution pass. The root facade translates
@@ -128,7 +129,8 @@ consumer contract requires a public package.
Each package must have one cohesive responsibility and a clear dependency Each package must have one cohesive responsibility and a clear dependency
direction. Internal packages must not expose their types merely to simplify direction. Internal packages must not expose their types merely to simplify
wiring, and the public facade must not leak internal representations through wiring, and the public facade must not leak internal representations through
exported signatures. Production packages under `internal/` must not import the exported signatures, except for the deliberately public configuration data model
specified below. Production packages under `internal/` must not import the
module root; the root facade may depend on internal implementation packages. module root; the root facade may depend on internal implementation packages.
New public packages require a durable consumer need that cannot be served New public packages require a durable consumer need that cannot be served
cleanly by the root facade. cleanly by the root facade.
@@ -136,12 +138,34 @@ cleanly by the root facade.
The [internal component overview](../internal/overview.md) must be updated as The [internal component overview](../internal/overview.md) must be updated as
packages are implemented or their responsibilities change. packages are implemented or their responsibilities change.
### Shared Configuration Data Exception
[ADR 0002](../adr/0002-shared-configuration-data-model.md) permits one canonical
set of public configuration settings and their nested data types in a small,
data-only internal package, exposed through aliases in the root `promptkit`
package. The facade and internal configuration adapter may use those same
declarations directly. Their fields, tags, and nested types are intentional
public contracts and receive the same API review as root-defined types.
The shared package owns data declarations, not loading, normalization,
validation, or execution behavior. This exception does not authorize aliases
for arbitrary internal implementation types or merging configuration settings
with source-construction helpers, execution requests, runtime objects, or
inspection reports that have different meanings. Internal packages still must
not import the root facade. Runtime capture still owns the values it retains;
shared declarations do not replace necessary ownership copies.
All other facade, dependency, and ownership rules remain in force. Concrete
package placement and implementation status belong in the active plan and,
once implemented, the internal component inventory.
## Exported API Discipline ## Exported API Discipline
Export the smallest contract required by real consumers. Exported declarations Export the smallest contract required by real consumers. Exported declarations
must have accurate GoDoc, stable semantics, and tests proportionate to their must have accurate GoDoc, stable semantics, and tests proportionate to their
compatibility risk. Avoid speculative extension points, aliases for internal compatibility risk. Avoid speculative extension points, aliases for internal
types, and public constructors that expose assembly details. types outside the shared configuration data exception, and public constructors
that expose assembly details.
Once an exported API exists, its Go declaration and GoDoc own its exact public Once an exported API exists, its Go declaration and GoDoc own its exact public
contract. Architecture documentation owns boundary rules, not a duplicate API contract. Architecture documentation owns boundary rules, not a duplicate API

View File

@@ -74,8 +74,8 @@ Successful publication needs no separate remote read-back or CI wait.
For optional local diagnosis, use `./scripts/releasekit.sh check VERSION`. For optional local diagnosis, use `./scripts/releasekit.sh check VERSION`.
It is not a required step before publication. The upstream It is not a required step before publication. The upstream
[candidate-check guide](https://forge.maximumdirect.net/base-layer/releasekit/src/tag/v0.8.0/docs/consumers/check.md) [candidate-check guide](https://forge.maximumdirect.net/base-layer/releasekit/src/tag/v0.10.0/docs/consumers/check.md)
and [CLI contract](https://forge.maximumdirect.net/base-layer/releasekit/src/tag/v0.8.0/docs/cli.md) and [CLI contract](https://forge.maximumdirect.net/base-layer/releasekit/src/tag/v0.10.0/docs/cli.md)
own shared checks and exit-status definitions. own shared checks and exit-status definitions.
## Deliver The Hosted Release ## Deliver The Hosted Release
@@ -97,14 +97,26 @@ For a later retry, use a clean checkout of the original tag with its recorded
module files and supply the token there. In a separate checkout, for example: module files and supply the token there. In a separate checkout, for example:
```sh ```sh
git switch --detach vMAJOR.MINOR.PATCH (
./scripts/releasekit.sh deliver vMAJOR.MINOR.PATCH set -e
release_version=vMAJOR.MINOR.PATCH
release_tag_ref="refs/tags/$release_version"
git fetch origin "$release_tag_ref:$release_tag_ref"
release_tag_commit=$(git rev-parse --verify "$release_tag_ref^{commit}")
git switch --detach "$release_tag_commit"
release_worktree_status=$(git status --porcelain --untracked-files=all)
if [ -n "$release_worktree_status" ]; then
printf '%s\n' 'Delivery retry requires a clean tagged checkout.' >&2
exit 1
fi
./scripts/releasekit.sh deliver "$release_version"
)
``` ```
Resolve uncertain hosted writes before retrying the same immutable version. Resolve uncertain hosted writes before retrying the same immutable version.
Matching content is accepted; conflicts are not overwritten. Do not rerun Matching content is accepted; conflicts are not overwritten. Do not rerun
`publish` or move tags to repair delivery. Follow the upstream `publish` or move tags to repair delivery. Follow the upstream
[delivery guide](https://forge.maximumdirect.net/base-layer/releasekit/src/tag/v0.8.0/docs/consumers/delivery.md) [delivery guide](https://forge.maximumdirect.net/base-layer/releasekit/src/tag/v0.10.0/docs/consumers/delivery.md)
for result interpretation and recovery. These commands apply to tags containing for result interpretation and recovery. These commands apply to tags containing
this delivery configuration; older releases retain their original workflows. this delivery configuration; older releases retain their original workflows.
@@ -131,13 +143,13 @@ and explicitly select the matching attempt:
Discovery is local; reconciliation observes the original destination without Discovery is local; reconciliation observes the original destination without
rerunning validation, changing tags, or delivering hosted content. Neither is rerunning validation, changing tags, or delivering hosted content. Neither is
a routine gate after successful publication. Follow the upstream a routine gate after successful publication. Follow the upstream
[recovery procedure](https://forge.maximumdirect.net/base-layer/releasekit/src/tag/v0.8.0/docs/release.md#resolve-publication-failures) [recovery procedure](https://forge.maximumdirect.net/base-layer/releasekit/src/tag/v0.10.0/docs/release.md#resolve-publication-failures)
for outcomes, incomplete evidence, and older attempts without records. for outcomes, incomplete evidence, and older attempts without records.
Records are stored under Git's common directory by default and can contain Records are stored under Git's common directory by default and can contain
credential-bearing endpoints. Keep them private, out of source control and credential-bearing endpoints. Keep them private, out of source control and
shared logs; retain them while needed for recovery and remove them when that shared logs; retain them while needed for recovery and remove them when that
value expires. See the [evidence guide](https://forge.maximumdirect.net/base-layer/releasekit/src/tag/v0.8.0/docs/consumers/publish.md#retain-and-reconcile-publication-evidence) value expires. See the [evidence guide](https://forge.maximumdirect.net/base-layer/releasekit/src/tag/v0.10.0/docs/consumers/publish.md#retain-and-reconcile-publication-evidence)
for storage and retention details. for storage and retention details.
Current absence does not prove a version was never published. Remove a retained Current absence does not prove a version was never published. Remove a retained

65
docs/releases/v0.18.0.md Normal file
View File

@@ -0,0 +1,65 @@
# Promptkit v0.18.0
Configuration inspection now exposes Promptkit's recorded field normalization
and derived queue-capacity defaults through ConfigKit v1.1.0. Engines also
include a default-on OpenRouter configuration. The minimum supported Go
version remains 1.27.0.
## Bundled OpenRouter Backend And Profiles
New engines include an embedded OpenRouter backend and seventeen root profiles:
ten latest-family aliases, the auto and free routers, and five auto-router
cost-tier variants. Select a profile explicitly in a prompt or request; the
bundle does not supply an implicit execution profile or contact providers
during construction. The backend uses `OPENROUTER_API_KEY` when available under
the existing optional-credential rules.
The bundle is the lowest-priority underlay. Applications can replace its records
or derive named variants through ordinary configuration, and external collections
remain supported. Cost-tier variants inherit the effective `auto` profile, so
replacing that parent changes their inherited target while retaining plugins.
The [bundle reference](../formats.md#bundled-openrouter-configuration) owns the
inventory and composition details.
Explicit smoke diagnostics now have OpenRouter authentication metadata and an
inference fallback profile. They remain opt-in and may incur provider costs.
Latest aliases and routers can change upstream; configured slugs do not pin
the concrete model, pricing, or availability. See the
[consumer guide](../consumers/pkg-promptkit.md#construct-an-engine) for selection
and opt-out examples.
## Recorded Configuration Normalization
Inspection shows normalized backend and profile settings, prompt metadata, and
configured message roles and cache controls, with field-level resolution
history. Existing source evidence remains available, and unchanged content-file
origins continue to select their resource context.
Configuration diffs compare normalized effective values, so spelling-only
changes that normalize identically may no longer appear. Lists remain atomic in
diffs. Inspection does not include backend-to-profile completion or request
overrides; see the [configuration inspection guide](../consumers/pkg-promptkit.md#inspect-the-captured-configuration).
## Compatibility And Upgrade
New engines install a bundled OpenRouter backend and root profiles by default.
Inventory listings and inspection sources therefore change, and references to
bundled IDs that previously failed may now resolve. Pass
`promptkit.WithBundledOpenRouter(false)` to retain application-only behavior.
Caller records can replace bundled definitions through ordinary composition;
same-ID replacements replace complete records rather than merging selected fields.
No configuration-format migration is required from v0.17.0.
Consumers that display or compare configuration inspection results should account
for normalized values and derived defaults replacing their earlier authored
representations. Request-time overrides and caller-appended messages retain
their existing processing.
```sh
go get go.maximumdirect.net/promptkit@v0.18.0
```
See the [framework format reference](../formats.md) for current configuration
semantics. Consumers upgrading from versions before v0.17.0 should also follow
the [configuration and API migration guide](../consumers/configuration-migration.md).
[COPYRIGHT](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.18.0/COPYRIGHT) | [LICENSE](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.18.0/LICENSE) | [ATTRIBUTION](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.18.0/docs/oss/attribution.md)

80
docs/releases/v0.19.0.md Normal file
View File

@@ -0,0 +1,80 @@
# Promptkit v0.19.0
This release completes two rounds of audit remediation, correcting error and
diagnostic reporting and simplifying configuration, validation, and value
ownership. It includes intentional compatibility changes for custom artifact
readers, programmatic configuration sources, and schemas. The minimum supported
Go version remains 1.27.0; ConfigKit remains at v1.1.0.
## Compatibility And Upgrade
Consumers upgrading from v0.18.0 should review these changes:
- A successful `ArtifactReader.Read` transfers ownership of the returned
artifact and its mutable body storage to Promptkit. Custom readers must stop
mutating returned values through retained aliases. Readers that reuse scratch
buffers must return independent storage. Promptkit retains its defensive
body copy, but that copy does not make concurrent buffer reuse safe. See the
[extension guidance](../consumers/pkg-promptkit.md#extension-interfaces).
- Provider maps passed through `ProfileSource` and `BackendSource` now use
ConfigKit's native Mapping input forms directly: built-in scalars, `[]any`,
and `map[string]any` recursively. Convert `json.Number` and typed containers
explicitly to supported values, choosing numeric precision appropriate to
the application. The former recursive compatibility conversion is removed.
Runtime execution overrides retain their separate accepted-value behavior.
- Schemas used for local validation and native structured output must be
self-contained. Move externally referenced definitions into the root document
and use internal references or explicit schema IDs. References requiring
another file or network resource, or relying on the source filename as a base,
fail during preparation. Promptkit does not bundle external schemas. See the
[schema reference](../formats.md#schemas).
Prepared schemas no longer inherit the depth and node limits used when copying
runtime provider values. Schemas previously rejected solely by those incidental
limits may now be accepted; decoder/compiler checks and the self-contained
schema requirement still apply.
No configuration-format migration is required from v0.18.0. After addressing
any affected integrations:
```sh
go get go.maximumdirect.net/promptkit@v0.19.0
```
## Errors And Diagnostics
Public errors and terminal observations now preserve the category of the
operation that failed even when an injected collaborator returns an error from
another Promptkit operation. Underlying causes remain available through
`errors.Is` and `errors.As`, without repeatedly duplicating nested cause trees.
Built-in generation preserves response-body read failures separately from
malformed response content, and failures during request preparation are
distinguished from dispatched attempts.
OpenRouter authentication diagnostics bound response recognition to 1 MiB.
An oversized HTTP 200 response reports
`SmokeTestReasonAuthenticationResponseLimitExceeded`, leaving credential
acceptance unknown. Cancellation during the read remains an interruption.
Consumers that enumerate diagnostic reasons should handle the new reason;
see the [diagnostic integration contract](../integrations/openai-compatible-chat.md#explicit-smoke-diagnostics).
## Internal Simplification And Ownership
Prompt lookup and rendering now consistently use captured configuration, and
smoke syntax and schema checks share the prepared output-validation lifecycle.
Unused routing defaults, alternate private workflows, and duplicate wire
representations have been removed.
Provider-setting resolution, inspection, prepared details, and smoke reports
avoid redundant validation and copying of already accepted values. Prepared
execution shares one private redacted metadata snapshot while preserving
independent caller-visible values. Provider values are detached before custom
JSON encoders run, protecting retained metadata from encoder-side mutation.
Requirement analysis, namespace-filtered listings, and repair-feedback
construction also avoid unnecessary traversal and temporary allocation. These
changes preserve existing request isolation and execution behavior without
introducing new performance guarantees.
[COPYRIGHT](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.19.0/COPYRIGHT) | [LICENSE](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.19.0/LICENSE) | [ATTRIBUTION](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.19.0/docs/oss/attribution.md)

55
docs/releases/v0.20.0.md Normal file
View File

@@ -0,0 +1,55 @@
# Promptkit v0.20.0
This minor release adds generically renderable resource inspection and typed
prompt source construction. The Go consumer minimum remains 1.27.0 and
ConfigKit remains at v1.1.0.
## Compatibility And Migration
`InspectProfile` now returns the resolved profile resource instead of the
backend-completed `EffectiveModelParams`. Replace accesses to that field with
the optional profile fields; use `Prepare` or `PreparedExecution.Details` for
completed request settings. `InspectPrompt` now includes captured messages and
uses optional pointers for presence-sensitive fields, including default profile,
inputs, input metadata, and output settings. Check pointers before dereferencing.
All three resource reports support ordinary JSON and YAML rendering without
per-field adapters. See the [consumer guidance](../consumers/pkg-promptkit.md#inspect-a-profile-before-prompt-work).
After configuration composition and inheritance, profile null endpoint,
API-key source, and whole provider map now permit backend defaults during
execution as well as appearing absent in resource inspection. Use explicit
empty strings to clear endpoint or API-key source. There is no replacement for
profile-level suppression of an entire backend provider map; edit backend
defaults through composition instead. The required-key rule still applies.
See [defaults and overrides](../formats.md#defaults-and-overrides) for the
migration example and retained ConfigKit null provenance.
## Resource Inspection And Sources
`InspectBackend` reports the resolved backend and configured diagnostics without
contacting a provider or deriving diagnostic destinations. Prompt and profile
inspection likewise report selected resources independently of request readiness;
configuration provenance, requirement analysis, and preparation remain separate.
`PromptSource` adds typed programmatic prompt definitions, and `BackendConfig`
supports sibling inheritance. ConfigKit continues to own source composition,
snapshotting, and relative message-file contexts.
The source helpers' [GoDoc](../../prompts.go) and
[backend configuration API](../../backends.go) describe programmatic construction.
## Refinements
Package GoDoc and migration guidance now consistently describe the resource
inspection contract, and the full package overview is visible in generated
documentation. Prompt inspection avoids repeating the same metadata lookup for
each declared input.
## Upgrade
After adapting affected inspection consumers and reviewing profile null usage:
```sh
go get go.maximumdirect.net/promptkit@v0.20.0
```
[COPYRIGHT](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.20.0/COPYRIGHT) | [LICENSE](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.20.0/LICENSE) | [ATTRIBUTION](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.20.0/docs/oss/attribution.md)

34
docs/releases/v0.21.0.md Normal file
View File

@@ -0,0 +1,34 @@
# Promptkit v0.21.0
This minor release updates ConfigKit from v1.1.0 to v1.2.0, making main-relative
companion imports available through Promptkit's existing configuration-source
API. The Go consumer minimum remains 1.27.0.
## Main-Relative Companion Imports
Applications can bind a non-file underlay's declared imports to the selected
main configuration's filesystem context. For example, an application can load
prompt definitions from an adjacent `prompts/` directory without calculating
that directory before loading the main configuration.
Imports remain explicitly selected by the caller; the upgrade does not enable
automatic discovery. Imported prompt message files retain their source-local
context. The [consumer example](../consumers/pkg-promptkit.md#load-companion-prompts-relative-to-the-main-configuration)
shows collection mounting and YAML filename filtering. ConfigKit's
[layered-loading contract](https://forge.maximumdirect.net/base-layer/configkit/src/tag/v1.2.0/docs/config.md#layered-loading)
defines context selection and fallback behavior.
## Compatibility And Upgrade
Promptkit's exported methods and request shapes are unchanged. No configuration
format migration is required from v0.20.0. Applications constructing ConfigKit
`Input` or `LayeredOptions` with positional struct literals must switch to
field-named literals because those upstream types gained fields. See the
[ConfigKit release notes](https://forge.maximumdirect.net/base-layer/configkit/src/tag/v1.2.0/docs/releases/v1.2.0.md)
for upstream changes.
```sh
go get go.maximumdirect.net/promptkit@v0.21.0
```
[COPYRIGHT](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.21.0/COPYRIGHT) | [LICENSE](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.21.0/LICENSE) | [ATTRIBUTION](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.21.0/docs/oss/attribution.md)

67
docs/releases/v0.22.0.md Normal file
View File

@@ -0,0 +1,67 @@
# Promptkit v0.22.0
This minor release adds captured conversation exchanges, optional sequential
coordination, and explicit execution-group summaries. Applications can continue
or branch a conversation without reconstructing rendered history, while keeping
response interpretation and workflow decisions in application code.
## Capture And Continue Conversations
Set `RunRequest.CaptureExchange` to capture a completed run's effective messages
and final assistant content. Supply its `RunResult.Exchange` as `ContinueFrom`
on a new request to continue it. Exchanges are reusable, support independent
branches and different engines, and expose detached history through `Details()`.
Prepared execution also supports capture of the exact snapshot it executes.
Internal structural-repair instructions and intermediate candidates are not
conversation history. A completed result with failed structural validation can
still be captured; an operational error returning no result creates no exchange.
Follow-ups select their own profiles, output contracts, credentials, and sessions
under ordinary request rules. See [capture and continuation guidance](../consumers/pkg-promptkit.md#capture-and-continue-a-conversation)
and the [public exchange contract](../../exchange.go).
## Coordinate Application-Directed Turns
`Engine.Coordinate` advances one conversation branch using an application callback
and an additional-turn budget. The application selects the next request or stops;
Promptkit handles turn progression and cancellation. Each completed run releases
backend admission before the callback, allowing assessment calls to use the same
backend. Stop and budget exhaustion are control-flow outcomes, not judgments of
response acceptance. Transport retries and structural repairs remain separate
from the turn budget.
See the [coordination walkthrough](../consumers/pkg-promptkit.md#coordinate-application-directed-turns)
and [coordinator GoDoc](../../coordinate.go) for callback and cancellation behavior.
## Collect Selected Execution Facts
`ExecutionGroup` collects explicitly selected execution observations, while
`ComposeObservers` combines groups with existing diagnostic observers. Detached
group snapshots report member identities, terminal coverage, attempt/retry/repair
counts, and available token totals without counting a run more than once.
Terminal coverage remains distinct from token-usage completeness; membership is
not inferred from ancestry or nested calls.
See [execution-group guidance](../consumers/pkg-promptkit.md#collect-selected-execution-facts),
the [group GoDoc](../../execution_group.go), and the updated
[offline observation example](../../examples/go-library/observe/main.go).
## Efficiency And Compatibility
History assembly and inspection avoid intermediate full-history copies. Observer
composition retains storage only for selected callbacks and avoids fan-out
closures for zero or one selected callback.
Capture and coordination are opt-in; existing execution workflows and
configuration formats need no migration from v0.21.0. The Go consumer minimum
remains 1.27.0, and dependency versions are unchanged. Exported request, prepared,
result, and observation structs gained fields; consumers using positional struct
literals must switch to keyed literals. Continuation metadata adds `parent_run_id`
to prepared/result JSON when applicable; the opaque exchange itself is excluded.
Exchanges are in-process capabilities, not serialized restartable sessions.
```sh
go get go.maximumdirect.net/promptkit@v0.22.0
```
[COPYRIGHT](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.22.0/COPYRIGHT) | [LICENSE](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.22.0/LICENSE) | [ATTRIBUTION](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.22.0/docs/oss/attribution.md)

62
docs/releases/v0.23.0.md Normal file
View File

@@ -0,0 +1,62 @@
# Promptkit v0.23.0
This minor release adds optional whole-operation timeouts and prepared-request
notifications to ordinary and coordinated execution. Applications can bound
each continuation independently and retain the request selected for execution
without replacing Promptkit's turn loop or preparing the same request twice.
## Bound Individual Operations
`RunRequest.OperationTimeout` selects a fresh cooperative budget covering
preparation and execution, including capacity waiting, retries, validation,
structural repair, and prepared-request notification. It is independent of the
provider generation timeout. Zero or negative values select no additional
timeout; existing parent cancellation and earlier deadlines remain effective.
For split execution, `PrepareExecution` retains the duration without starting
a timer. `RunPrepared` starts that budget after claiming the handle, so prior
preparation and idle time do not consume it. Callers can still supply one shared
deadline context to bound both calls together.
Local budget failures match both `ErrOperationTimeout` and
`context.DeadlineExceeded`. Each coordinated turn uses its selected budget;
expiry stops that coordination through the existing operational-error path
without canceling its parent or sibling work. Completed outcomes are fixed
before terminal observation and are not changed by a slow terminal observer.
Cancellation does not forcibly interrupt callbacks or collaborators that ignore
their context.
See the [execution-controls guide](../consumers/pkg-promptkit.md#bound-execution-and-retain-prepared-details)
and the [public request contract](../../types.go) and
[timeout error contract](../../engine.go) for exact behavior.
## Retain Prepared Details Before Generation
`RunRequest.OnPrepared` optionally receives a detached, credential-redacted
`PreparedRun` and the operation context before first generation. It uses the
actual prepared snapshot, including continuation history, without rereading
inputs or rerendering. Applications may retain these details even if generation
later fails without returning a result.
Notification is synchronous, at most once per logical execution, and runs while
logical admission is held. It is not repeated for retries or structural repair.
Preparation or admission failure can prevent delivery. The application owns
diagnostic storage and disclosure decisions; ordinary observers remain
metadata-only. See the [OnPrepared GoDoc](../../types.go) and updated
[offline observation example](../../examples/go-library/observe/main.go).
## Compatibility And Upgrade
Both controls are opt-in. Existing keyed requests and configuration files need
no migration from v0.22.0. `Prepare` and `ValidateRequest` continue using only
their supplied contexts and do not invoke `OnPrepared`. `RunRequest` gained
fields; consumers using positional struct literals must switch to keyed
literals. Prepared/result JSON representations are unchanged.
The Go consumer minimum remains 1.27.0, and dependency versions are unchanged.
```sh
go get go.maximumdirect.net/promptkit@v0.23.0
```
[COPYRIGHT](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.23.0/COPYRIGHT) | [LICENSE](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.23.0/LICENSE) | [ATTRIBUTION](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.23.0/docs/oss/attribution.md)

33
docs/releases/v0.23.1.md Normal file
View File

@@ -0,0 +1,33 @@
# Promptkit v0.23.1
This patch release fixes construction-error wrapping so downstream consumers
can classify failures from their underlying causes without treating the
`ErrInvalidConfig` category as an independent failure.
## Fixed
`NewEngine` now uses single-cause wrapping for both configuration-loading and
option-application failures. `errors.Is` still recognizes `ErrInvalidConfig`,
and `errors.As` still exposes ConfigKit loading errors with their diagnostics
and source context. Genuine joined causes, including cancellation, remain
inspectable. Consumers retain responsibility for mapping those failures to
application outcomes or exit statuses.
See the [constructor GoDoc](../../engine.go) and
[error-handling guide](../consumers/pkg-promptkit.md#handle-errors) for the
public error contracts and usage.
## Compatibility And Upgrade
No configuration or request changes are required when upgrading from v0.23.0.
Consumers that inspect constructor error trees directly should account for
the category no longer appearing as a separate sibling branch; use
`errors.Is` and `errors.As` for category and typed-cause inspection.
The Go consumer minimum remains 1.27.0, and dependency versions are unchanged.
```sh
go get go.maximumdirect.net/promptkit@v0.23.1
```
[COPYRIGHT](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.23.1/COPYRIGHT) | [LICENSE](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.23.1/LICENSE) | [ATTRIBUTION](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.23.1/docs/oss/attribution.md)

35
docs/releases/v0.23.2.md Normal file
View File

@@ -0,0 +1,35 @@
# Promptkit v0.23.2
This patch release refreshes configuration, validation, and maintainer-tool
dependencies. No Promptkit configuration or request API migration is required.
## Dependency Updates
- ConfigKit v1.2.0 → v1.2.1 and go-yaml v4.0.0-rc.6 → v4.0.0 move
configuration loading to stable go-yaml, with ConfigKit's corresponding
loader-plugin adaptation.
- jsonschema v6.0.2 → v6.0.3 includes fixes to validation-location reporting,
numeric edge cases, and email validation.
- Go Text (`golang.org/x/text`) v0.14.0 → v0.44.0 refreshes the transitive
dependency used by jsonschema.
- ReleaseKit v0.9.0 → v0.10.0 updates maintainer tooling, selecting Goldmark
v2.1.16 and Go Modules (`golang.org/x/mod`) v0.43.0. ReleaseKit's Go 1.27.0
minimum is already satisfied by Promptkit's existing minimum.
Dependency notices and revision links are maintained in the
[attribution document](../oss/attribution.md). Maintainers should use the
updated [release runbook](../release.md); publication and hosted delivery
remain separate operations.
## Compatibility And Upgrade
The Go consumer minimum remains 1.27.0. Existing configuration, requests,
and release commands require no changes from v0.23.1. Upstream validator
fixes may affect validation results or diagnostic locations for edge cases.
See the [consumer guide](../consumers/pkg-promptkit.md) for current usage.
```sh
go get go.maximumdirect.net/promptkit@v0.23.2
```
[COPYRIGHT](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.23.2/COPYRIGHT) | [LICENSE](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.23.2/LICENSE) | [ATTRIBUTION](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.23.2/docs/oss/attribution.md)

36
docs/releases/v0.23.3.md Normal file
View File

@@ -0,0 +1,36 @@
# Promptkit v0.23.3
This patch release corrects timeout attribution and preserves error identities
in the built-in OpenAI-compatible HTTP client.
## Fixed
- Operation budgets expiring during an in-flight HTTP generation request now
match both `ErrOperationTimeout` and `context.DeadlineExceeded`. Terminal
observations report a deadline failure instead of a transport failure.
- HTTP failures carrying custom caller cancellation or deadline causes retain
both the custom cause and the standard context identity. Unrelated transport
failures are not relabeled solely because a context has been canceled.
- Cancellation while reading a successful response body preserves the original
reader error alongside the context identity, without exposing its diagnostic
text through the redacted error message.
Regression coverage exercises `Run` and `RunPrepared`, public error identities,
and attempt and terminal classifications. See the [request and timeout
contracts](../../types.go), [public error GoDoc](../../engine.go), and
[error-handling guide](../consumers/pkg-promptkit.md#handle-errors).
## Compatibility And Upgrade
No configuration or request API migration is required from v0.23.2. Consumers
using `errors.Is` for cancellation or deadline handling now receive the
correct identities on the affected paths; applications continue to own their
status mapping and diagnostic presentation.
The Go consumer minimum remains 1.27.0, and dependency versions are unchanged.
```sh
go get go.maximumdirect.net/promptkit@v0.23.3
```
[COPYRIGHT](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.23.3/COPYRIGHT) | [LICENSE](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.23.3/LICENSE) | [ATTRIBUTION](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.23.3/docs/oss/attribution.md)

83
docs/releases/v0.24.0.md Normal file
View File

@@ -0,0 +1,83 @@
# Promptkit v0.24.0
This minor release adds application-owned configuration loading, prompt-owned
schemas, and a prepared message-content hash. It also enables one repair attempt
by default for enabled output validation. This is a breaking pre-v1 release;
consumers upgrading from v0.23.x should review the migration steps below.
## Added And Changed
- Applications can embed `promptkit.Settings`, compose Promptkit's rules with
`Configure`, load configuration once through ConfigKit, and pass the entire
result to `NewFromConfig`. `NewEngine` remains the standalone convenience
constructor. See the [shared-loading guide](../consumers/pkg-promptkit.md#reuse-an-application-owned-load)
and [offline preparation example](../../examples/go-library/prepare/main.go).
- Prompts can select inline, source-relative, or explicit-root JSON Schemas.
Collections can supply their own schema root. Schema resources are loaded
during preparation and retained by prepared execution. See the
[schema reference](../formats.md#schemas).
- Prepared results include `MessageContentHash` (`message_content_hash` in
JSON), an opaque equality value for ordered message roles and exact text,
independent of session identity, cache hints, and other metadata. The existing
rendered-prompt hash algorithm is unchanged. See the
[prepared-result GoDoc](../../types.go) and
[preparation guide](../consumers/pkg-promptkit.md#prepare-now-and-execute-the-same-snapshot-later).
- Configuration inspection uses a native ConfigKit view focused on `/promptkit`
and adds `HistoricalPaths`. It retains whole-load evidence and supports
explicit queries outside that focus; comparison can include ancestor changes.
See the [inspection guide](../consumers/pkg-promptkit.md#inspect-the-captured-configuration).
## Fixes And Efficiency
- Prompt-input inspection and requirements now retain effective metadata supplied
by application defaults, including inherited input requirements.
- Configuration capture avoids expanding inheritance provenance just to determine
input-field presence, and shares immutable backend defaults across profiles
without overrides while preserving defensive copies at public lookups.
- Schema handling avoids copying inactive runtime schemas and unused slice
capacity. Inline schemas are separate from provider extra-parameter copy limits.
See the [schema reference](../formats.md#schemas) and
[request GoDoc](../../types.go).
## Compatibility And Upgrade
The [configuration migration guide](../consumers/configuration-migration.md#move-settings-under-promptkit)
owns the full migration walkthrough. In particular:
- Move complete-document resources and `schema_root` under `promptkit`.
Previous root-level resources are now application data, not Promptkit
resources; IDs and references keep their existing names. `SettingsInput`
mounts wrapperless fragments at `/promptkit`; `CollectionInput` mounts them
at `/promptkit/collections/<namespace>`. Discovery helpers still produce
root-mounted inputs, so adjust fragment destinations explicitly.
- For an application-owned load, select `BundledOpenRouterUnderlay` before
loading if you want the bundled catalog. Bundling remains default-on for
`NewEngine`; it is not implicitly added by `NewFromConfig`.
- Replace `RunRequest.Validation` values of type `*OutputContract` with
`*OutputContractInput`; its `RepairAttempts` field is a pointer. Omitted or
null budgets now resolve to one for `basic`, `json`, and `json_schema`, or
zero for `none`. Set an explicit zero to retain single-pass behavior.
Runtime replacements still replace the whole contract and resolve their
defaults independently. A failed check may now incur another generation call,
latency, and usage, without guaranteeing a valid result. See the
[output contract](../formats.md#output-contract).
- `PromptOutputInspection.RepairAttempts` is now a resolved integer rather than
an optional pointer, and its JSON includes zero. `OutputContract` remains the
resolved contract type used by prepared details. See the
[inspection GoDoc](../../resource_inspection.go) and [contract GoDoc](../../types.go).
- Schema paths require an explicitly configured applicable root; there is no
implicit working-directory root. Review file references and make schemas
self-contained as described in the [schema reference](../formats.md#schemas).
- Promptkit no longer validates resource-version fields. Applications own version
policy and sensitivity rules for their extension fields; see the
[format reference](../formats.md).
Older prepared-result JSON without `message_content_hash` decodes with an empty
value. Callers using positional `PreparedRun` literals may need to adjust them.
ConfigKit is updated to v1.4.0; the Go consumer minimum remains 1.27.0.
```sh
go get go.maximumdirect.net/promptkit@v0.24.0
```
[COPYRIGHT](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.24.0/COPYRIGHT) | [LICENSE](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.24.0/LICENSE) | [ATTRIBUTION](https://forge.maximumdirect.net/base-layer/promptkit/src/tag/v0.24.0/docs/oss/attribution.md)

View File

@@ -38,30 +38,17 @@ consumers.
## Ideas ## Ideas
### Field-Level Normalization With Preserved List Provenance ### Schema Bundling
Reevaluate normalizing resource fields directly in ConfigKit's effective Allow consumers to author reusable schemas across multiple files and package
configuration after upstream ConfigKit supports recording changes within a them into a single self-contained schema for local validation and provider
list without replacing the source attribution of unchanged fields. submission. This could restore multi-file authoring convenience while retaining
the [self-contained schema requirement](../formats.md#schemas).
This could let configuration inspection show PromptKit-normalized values and Bundling would need to preserve reference semantics, identifiers, anchors, and
field-level change history while preserving the origins needed to resolve recursion. A complete bundle would not guarantee that a provider supports every
relative resources such as `content_file`. Assess whether that inspection schema feature. This is a possible future enhancement; implementation is not
benefit justifies changing the execution-owned normalization approach. currently approved.
Include normalization of roles and cache-control values in configured messages
in this reevaluation. This could make inspection reflect the values used for
execution and reduce separate normalization paths. Upstream support would need
to preserve untouched field origins within nested message lists as well as outer
resource lists. PromptKit would continue owning the normalization rules;
ConfigKit would record their effects and provenance. Caller-appended messages
would still need request-time normalization because they do not pass through
ConfigKit.
Consider moving these transformations only if it improves inspection consistency
and simplifies implementation without changing execution behavior.
Upstream support is the reconsideration trigger, not a dependency of the
current ConfigKit migration or a commitment to adopt the enhancement.
## Entry Format ## Entry Format

273
engine.go
View File

@@ -4,11 +4,13 @@ import (
"context" "context"
"errors" "errors"
"fmt" "fmt"
"io/fs"
"net/http" "net/http"
"reflect"
"time" "time"
"go.maximumdirect.net/configkit" "go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit/internal/backend"
"go.maximumdirect.net/promptkit/internal/configuration"
"go.maximumdirect.net/promptkit/internal/domain" "go.maximumdirect.net/promptkit/internal/domain"
internalengine "go.maximumdirect.net/promptkit/internal/engine" internalengine "go.maximumdirect.net/promptkit/internal/engine"
"go.maximumdirect.net/promptkit/internal/observation" "go.maximumdirect.net/promptkit/internal/observation"
@@ -30,6 +32,8 @@ var (
// ErrProfileNotFound identifies a selected profile ID that is absent from // ErrProfileNotFound identifies a selected profile ID that is absent from
// every configured profile source. It does not also match ErrProfileLoad. // every configured profile source. It does not also match ErrProfileLoad.
ErrProfileNotFound = errors.New("profile not found") ErrProfileNotFound = errors.New("profile not found")
// ErrBackendNotFound identifies a backend absent from resource inspection.
ErrBackendNotFound = errors.New("backend not found")
// ErrProfileRequired identifies a request for which neither RunRequest.ProfileID // ErrProfileRequired identifies a request for which neither RunRequest.ProfileID
// nor the selected prompt's default profile is present. Such an error also // nor the selected prompt's default profile is present. Such an error also
// matches ErrInvalidRequest. // matches ErrInvalidRequest.
@@ -57,6 +61,13 @@ var (
// A [CapacityError] reports the selected backend ID. It is not an invalid // A [CapacityError] reports the selected backend ID. It is not an invalid
// request, an LLM or provider rate-limit response, or ErrLLMGenerate. // request, an LLM or provider rate-limit response, or ErrLLMGenerate.
ErrCapacityExceeded = errors.New("backend capacity exceeded") ErrCapacityExceeded = errors.New("backend capacity exceeded")
// ErrOperationTimeout identifies expiration of the timeout selected by
// RunRequest.OperationTimeout. Such errors also match context.DeadlineExceeded
// and retain their underlying operation and collaborator identities.
// Earlier parent cancellation or deadlines and provider-only deadlines do
// not acquire this identity. A completed result remains completed even if
// a terminal observer outlasts the selected budget.
ErrOperationTimeout = errors.New("operation timeout exceeded")
// ErrLLMGenerate identifies a model-client failure or a nil successful // ErrLLMGenerate identifies a model-client failure or a nil successful
// response. A built-in OpenAI-compatible non-2xx response is available as a // response. A built-in OpenAI-compatible non-2xx response is available as a
// [GenerationError]. Errors returned by an injected LLMClient remain // [GenerationError]. Errors returned by an injected LLMClient remain
@@ -81,12 +92,12 @@ var (
ErrValidation = errors.New("failed to validate output") ErrValidation = errors.New("failed to validate output")
) )
// Engine inspects prompts and profiles, enumerates prompt requirements, and // Engine inspects resources, enumerates prompt requirements, and
// prepares and runs Promptkit prompt requests. // prepares and runs Promptkit prompt requests.
// //
// An Engine is safe for concurrent calls to [Engine.WithNamespaces], // An Engine is safe for concurrent calls to [Engine.WithNamespaces],
// [Engine.InspectConfiguration], // [Engine.InspectConfiguration],
// [Engine.InspectPrompt], // [Engine.InspectPrompt], [Engine.InspectBackend],
// [Engine.InspectRequirements], [Engine.InspectProfile], [Engine.Prepare], [Engine.PrepareExecution], // [Engine.InspectRequirements], [Engine.InspectProfile], [Engine.Prepare], [Engine.PrepareExecution],
// [Engine.Run], [Engine.RunPrepared], [Engine.SmokeTest], [Engine.ListProfiles], // [Engine.Run], [Engine.RunPrepared], [Engine.SmokeTest], [Engine.ListProfiles],
// [Engine.ListPrompts], [Engine.ListBackends], [Engine.ListNamespaces], and // [Engine.ListPrompts], [Engine.ListBackends], [Engine.ListNamespaces], and
@@ -101,8 +112,9 @@ type Engine struct {
} }
// ConfigurationSources selects ordered ConfigKit underlays and an optional // ConfigurationSources selects ordered ConfigKit underlays and an optional
// highest-priority main input. Each underlay's inputs strictly compose within // highest-priority main input. By default a bundled OpenRouter underlay precedes
// their group; later underlays overlay earlier ones. A nil Main uses // the caller underlays; [WithBundledOpenRouter] can disable it. Inputs within
// each underlay strictly compose; later underlays overlay earlier ones. A nil Main uses
// ConfigKit's empty source. A non-nil Main is passed through unchanged and // ConfigKit's empty source. A non-nil Main is passed through unchanged and
// may fail construction if invalid. // may fail construction if invalid.
type ConfigurationSources struct { type ConfigurationSources struct {
@@ -115,9 +127,6 @@ type ConfigurationSources struct {
type Config struct { type Config struct {
// Sources combines maintained definitions with caller-selected inputs. // Sources combines maintained definitions with caller-selected inputs.
Sources ConfigurationSources Sources ConfigurationSources
// SchemaDir is the root for JSON Schema files. An empty value uses the
// current directory. WithSchemaFS or WithSchemaFile replaces this source.
SchemaDir string
// Timeout is the transport-wide safety cap for the built-in LLM client // Timeout is the transport-wide safety cap for the built-in LLM client
// when HTTPClient is absent or has a non-positive timeout. A zero or negative // when HTTPClient is absent or has a non-positive timeout. A zero or negative
// value selects the 10-minute default. // value selects the 10-minute default.
@@ -137,7 +146,7 @@ type Config struct {
// Option customizes engine construction. // Option customizes engine construction.
// //
// NewEngine applies options in argument order and ignores nil options. Within // NewEngine applies options in argument order and ignores nil options. Within
// each schema-source, model-client, and artifact-reader category, the last // each model-client and artifact-reader category, the last
// non-nil valid option replaces earlier options. An invalid option fails // non-nil valid option replaces earlier options. An invalid option fails
// construction even if a later option would replace it. // construction even if a later option would replace it.
type Option interface { type Option interface {
@@ -151,7 +160,18 @@ func (f optionFunc) apply(options *engineOptions) error {
} }
type engineOptions struct { type engineOptions struct {
internal internalengine.Options internal internalengine.Options
disableBundledOpenRouter bool
}
// WithBundledOpenRouter enables or disables the built-in OpenRouter underlay.
// It is enabled by default for NewEngine; when repeated, the last option wins.
// It has no effect on an already resolved NewFromConfig handoff.
func WithBundledOpenRouter(enabled bool) Option {
return optionFunc(func(options *engineOptions) error {
options.disableBundledOpenRouter = !enabled
return nil
})
} }
// WithLLMClient replaces the built-in model client used by [Engine.Run] and // WithLLMClient replaces the built-in model client used by [Engine.Run] and
@@ -187,64 +207,145 @@ func WithArtifactReader(reader ArtifactReader) Option {
}) })
} }
// WithSchemaFS loads JSON Schema documents from fsys under root.
//
// Prompt schema_path values resolve within this source when schema validation
// or structured output is requested. fsys must be non-nil and root must be
// non-empty; otherwise NewEngine fails with ErrInvalidConfig. This option
// replaces Config.SchemaDir and earlier schema-source options.
func WithSchemaFS(fsys fs.FS, root string) Option {
return optionFunc(func(options *engineOptions) error {
return options.internal.SetSchemaFS(fsys, root)
})
}
// WithSchemaFile loads JSON Schema documents from the single schema file at path.
//
// Prompt schema_path values refer to the file's base name. path must name an
// existing non-directory file when NewEngine applies the option. This option
// replaces Config.SchemaDir and earlier schema-source options.
func WithSchemaFile(path string) Option {
return optionFunc(func(options *engineOptions) error {
return options.internal.SetSchemaFile(path)
})
}
// NewEngine constructs an Engine from configuration and options. // NewEngine constructs an Engine from configuration and options.
// //
// ConfigKit composes caller-selected underlays in their given order, followed // ConfigKit composes caller-selected underlays in their given order, followed
// by the optional main input. Construction validates the final effective // by the optional main input, after the default-on bundled OpenRouter underlay.
// The bundle adds a root backend and profiles to inventories, but no prompts
// or implicit profile selection. Construction validates the final effective
// backend, profile, and prompt definitions and captures referenced prompt // backend, profile, and prompt definitions and captures referenced prompt
// content. A missing prompt collection is allowed. Schema documents and // content. A missing prompt collection is allowed. Schema documents and
// request artifacts remain runtime resources. // request artifacts remain runtime resources.
// //
// NewEngine returns an error matching ErrInvalidConfig for invalid // NewEngine returns an error matching ErrInvalidConfig through errors.Is for
// configuration or options, preserving a ConfigKit LoadError when loading // invalid configuration or options. The error unwraps to the original
// fails. Each constructed // construction cause, preserving a ConfigKit LoadError when loading fails.
// Engine has independent backend-capacity pools. Construction does not perform // Each constructed Engine has independent backend-capacity pools. Construction
// model requests or require credentials. A negative Config.TransientRetries // does not perform model requests or require credentials. A negative
// is invalid. // Config.TransientRetries is invalid.
func NewEngine(cfg Config, opts ...Option) (*Engine, error) { func NewEngine(cfg Config, opts ...Option) (*Engine, error) {
options, err := applyEngineOptions(opts)
if err != nil {
return nil, err
}
runtime := RuntimeConfig{Timeout: cfg.Timeout, HTTPClient: cfg.HTTPClient, TransientRetries: cfg.TransientRetries}
if _, err := domain.ResolveTransientRetries(2, runtime.TransientRetries); err != nil {
return nil, newPublicOperationError(ErrInvalidConfig, err)
}
underlays := append([]configkit.Underlay(nil), cfg.Sources.Underlays...)
if !options.disableBundledOpenRouter {
underlays = append([]configkit.Underlay{BundledOpenRouterUnderlay()}, underlays...)
}
main := configkit.Input{Source: configkit.NoSource()}
if cfg.Sources.Main != nil {
main = *cfg.Sources.Main
}
definition := Configure(configkit.Definition[standaloneConfiguration]{}, func(value *standaloneConfiguration) *Settings { return &value.PromptKit })
result, err := configkit.LoadLayered(main, configkit.LayeredOptions[standaloneConfiguration]{Underlays: underlays, Definition: definition})
if err != nil {
return nil, newPublicOperationError(ErrInvalidConfig, err)
}
return newFromConfig(result, runtime, options)
}
type standaloneConfiguration struct {
PromptKit Settings `yaml:"promptkit"`
Application map[string]any `yaml:",inline"`
}
// Configuration is the completed application-owned configuration evidence
// needed to capture PromptKit resources without loading configuration again.
type Configuration interface {
Value(configkit.Path) (any, bool)
InspectSubtree(configkit.Path) (configkit.Inspection, bool)
Diagnostics() []configkit.Diagnostic
}
// RuntimeConfig selects runtime-only client and retry settings for NewFromConfig.
// Its fields have the same semantics as the corresponding Config fields.
type RuntimeConfig struct {
// Timeout is the built-in client's transport timeout fallback.
Timeout time.Duration
// HTTPClient is cloned for the built-in client and ignored by injected clients.
HTTPClient *http.Client
// TransientRetries selects additional built-in attempts; nil defaults to two.
TransientRetries *int
}
// NewFromConfig captures resources from an application-owned completed load.
// The application must register PromptKit rules during that load, using Configure
// or equivalent explicit composition. Missing or nil selected settings contribute
// no resources; Settings and *Settings are accepted, while other nonnil types,
// nil results, and unavailable inspection views fail with ErrInvalidConfig.
// Construction consumes resolved settings while inspection retains the load's
// frozen evidence. Callers should keep those consistent at handoff. After
// construction, later settings mutations do not alter the engine. This method
// does not load sources, rerun callbacks, or select bundled underlays.
func NewFromConfig(result Configuration, cfg RuntimeConfig, opts ...Option) (*Engine, error) {
options, err := applyEngineOptions(opts)
if err != nil {
return nil, err
}
return newFromConfig(result, cfg, options)
}
func applyEngineOptions(opts []Option) (engineOptions, error) {
var options engineOptions var options engineOptions
for _, opt := range opts { for _, opt := range opts {
if opt == nil { if opt == nil {
continue continue
} }
if err := opt.apply(&options); err != nil { if err := opt.apply(&options); err != nil {
return nil, fmt.Errorf("%w: %w", ErrInvalidConfig, err) return engineOptions{}, newPublicOperationError(ErrInvalidConfig, err)
} }
} }
return options, nil
}
assembled, err := internalengine.New( func newFromConfig(result Configuration, cfg RuntimeConfig, options engineOptions) (*Engine, error) {
internalengine.Config{Sources: internalengine.ConfigurationSources{Underlays: cfg.Sources.Underlays, Main: cfg.Sources.Main}, SchemaDir: cfg.SchemaDir, Timeout: cfg.Timeout, HTTPClient: cfg.HTTPClient, TransientRetries: copyIntPtr(cfg.TransientRetries)}, if result == nil || isNilConfiguration(result) {
options.internal, return nil, newPublicOperationError(ErrInvalidConfig, errors.New("configuration result is nil"))
) }
if _, err := domain.ResolveTransientRetries(2, cfg.TransientRetries); err != nil {
return nil, newPublicOperationError(ErrInvalidConfig, err)
}
value, available := result.Value(configkit.Keys("promptkit"))
var selected *Settings
if available && value != nil {
switch section := value.(type) {
case Settings:
selected = &section
case *Settings:
selected = section
default:
return nil, newPublicOperationError(ErrInvalidConfig, fmt.Errorf("promptkit section has incompatible type %T", value))
}
}
view, ok := result.InspectSubtree(configkit.Keys("promptkit"))
if !ok {
return nil, newPublicOperationError(ErrInvalidConfig, errors.New("configuration inspection is unavailable"))
}
snapshot, err := configuration.Capture(selected, view, result.Diagnostics())
if err != nil { if err != nil {
return nil, fmt.Errorf("%w: %w", ErrInvalidConfig, err) return nil, newPublicOperationError(ErrInvalidConfig, err)
}
assembled, err := internalengine.New(snapshot, internalengine.Config{Timeout: cfg.Timeout, HTTPClient: cfg.HTTPClient, TransientRetries: copyIntPtr(cfg.TransientRetries)}, options.internal)
if err != nil {
return nil, newPublicOperationError(ErrInvalidConfig, err)
} }
return &Engine{internal: assembled}, nil return &Engine{internal: assembled}, nil
} }
func isNilConfiguration(result Configuration) bool {
value := reflect.ValueOf(result)
switch value.Kind() {
case reflect.Chan, reflect.Func, reflect.Interface, reflect.Map, reflect.Ptr, reflect.Slice:
return value.IsNil()
default:
return false
}
}
// InspectPrompt resolves one prompt ID without selecting a profile or starting // InspectPrompt resolves one prompt ID without selecting a profile or starting
// execution work. // execution work.
// //
@@ -254,16 +355,17 @@ func NewEngine(cfg Config, opts ...Option) (*Engine, error) {
// qualified IDs pass through. // qualified IDs pass through.
// //
// A successful result proves that the selected definition and any referenced // A successful result proves that the selected definition and any referenced
// message content files were structurally loaded. Inputs are returned in // message content files were structurally loaded. Inputs and captured,
// definition order. DefaultProfileID is declared metadata only and is not // unrendered messages are returned in definition order. DefaultProfileID is
// declared metadata only and is not
// resolved. OutputContract is the normalized declared contract, with a JSON // resolved. OutputContract is the normalized declared contract, with a JSON
// Schema path when declared but without loading or compiling that schema. // Schema selection when declared but without loading or compiling schema files.
// PromptHash is the same opaque equality value as PreparedRun.PromptHash for // PromptHash is the same opaque equality value as PreparedRun.PromptHash for
// the selected captured definition; its spelling, length, // the selected captured definition; its spelling, length,
// encoding, algorithm, and security properties are not contracts. // encoding, algorithm, and security properties are not contracts.
// //
// This method does not return prompt bodies, templates, source paths, schemas, // This method does not return parsed templates, filesystem capabilities,
// rendered messages, or execution settings. It does not resolve a profile or // schema file contents, rendered messages, or execution settings. It does not resolve a profile or
// credential, read artifacts or schemas, render, validate, admit capacity, // credential, read artifacts or schemas, render, validate, admit capacity,
// contact a provider, or generate model output. The returned PromptInspection // contact a provider, or generate model output. The returned PromptInspection
// and its input slice are caller-owned. Inspection does not reload configuration // and its input slice are caller-owned. Inspection does not reload configuration
@@ -286,21 +388,19 @@ func (e *Engine) InspectPrompt(
if err != nil { if err != nil {
return nil, mapPublicError(err) return nil, mapPublicError(err)
} }
return fromDomainPromptInspection(inspection), nil return fromPromptResource(inspection), nil
} }
// InspectProfile resolves one explicit profile without selecting a prompt or // InspectProfile resolves one explicit profile without selecting a prompt or
// starting execution work. // starting execution work.
// //
// InspectProfile looks up a nonblank profileID exactly and case-sensitively // InspectProfile looks up a nonblank profileID exactly and case-sensitively
// in the captured effective configuration. It applies the framework timeout // in the captured effective configuration. It reports the resolved profile
// baseline, selected // resource without backend completion, execution defaults, or request overrides.
// backend, and then selected profile to EffectiveModelParams without a request
// override. BackendID is empty for an endpoint-only profile.
// //
// APIKeyEnv in the returned target is an environment-variable name, never its // APIKeyEnv in the returned resource is an environment-variable name, never its
// value. APIKeyRequired instead reports a direct credential requirement and is // value. APIKeyRequired, when present and true, reports that a credential is
// mutually exclusive with a nonblank APIKeyEnv. InspectProfile neither derives // required; an explicit profile environment source can also be present. InspectProfile neither derives
// an ID from a prompt default_profile nor checks credential availability, so an // an ID from a prompt default_profile nor checks credential availability, so an
// absent or blank named environment variable is not an error. // absent or blank named environment variable is not an error.
// //
@@ -324,7 +424,32 @@ func (e *Engine) InspectProfile(ctx context.Context, profileID string) (*Profile
if err != nil { if err != nil {
return nil, mapPublicError(err) return nil, mapPublicError(err)
} }
return fromDomainProfileInspection(inspection), nil return fromProfileResource(inspection), nil
}
// InspectBackend selects a resolved backend by exact ID from the captured
// configuration, qualifying a slashless ID with the backend namespace. It
// reports configured resource fields and diagnostics, not credential values,
// derived diagnostic destinations, or runtime capacity occupancy. References
// are not resolved and no network request is made. The result and its mutable
// values are caller-owned. Source changes require a new engine.
//
// An invalid engine matches ErrInvalidConfig, an invalid ID matches
// ErrInvalidRequest, and an absent backend matches ErrBackendNotFound (not
// ErrInvalidRequest). Cancellation remains detectable via errors.Is. No
// partial result is returned on error.
func (e *Engine) InspectBackend(ctx context.Context, backendID string) (*BackendInspection, error) {
if e == nil || e.internal == nil {
return nil, fmt.Errorf("%w: engine is nil", ErrInvalidConfig)
}
inspection, err := e.internal.InspectBackend(ctx, backendID)
if err != nil {
if errors.Is(err, backend.ErrBackendNotFound) {
return nil, fmt.Errorf("%w: %w", ErrBackendNotFound, err)
}
return nil, mapPublicError(err)
}
return fromBackendResource(inspection), nil
} }
// Prepare resolves and renders a prompt request without calling an LLM. // Prepare resolves and renders a prompt request without calling an LLM.
@@ -346,6 +471,7 @@ func (e *Engine) InspectProfile(ctx context.Context, profileID string) (*Profile
// reported in the applicable operation category; no general errors.Is // reported in the applicable operation category; no general errors.Is
// relationship to ctx.Err is promised. Prepare returns no partial result on // relationship to ctx.Err is promised. Prepare returns no partial result on
// error. It validates RunRequest.TransientRetries but makes no model attempts. // error. It validates RunRequest.TransientRetries but makes no model attempts.
// RunRequest.OperationTimeout and RunRequest.OnPrepared have no effect here.
func (e *Engine) Prepare(ctx context.Context, req RunRequest) (*PreparedRun, error) { func (e *Engine) Prepare(ctx context.Context, req RunRequest) (*PreparedRun, error) {
prepared, err := e.prepareRequest(ctx, req) prepared, err := e.prepareRequest(ctx, req)
if err != nil { if err != nil {
@@ -367,6 +493,7 @@ func (e *Engine) Prepare(ctx context.Context, req RunRequest) (*PreparedRun, err
// matching ErrInvalidConfig when the Engine is nil. Success does not guarantee // matching ErrInvalidConfig when the Engine is nil. Success does not guarantee
// provider acceptance or later execution success. Use [Engine.PrepareExecution] // provider acceptance or later execution success. Use [Engine.PrepareExecution]
// when execution must use the snapshot produced by preparation. // when execution must use the snapshot produced by preparation.
// Execution-only controls do not start a timer or invoke a notification here.
func (e *Engine) ValidateRequest(ctx context.Context, req RunRequest) error { func (e *Engine) ValidateRequest(ctx context.Context, req RunRequest) error {
_, err := e.prepareRequest(ctx, req) _, err := e.prepareRequest(ctx, req)
return err return err
@@ -409,6 +536,8 @@ func (e *Engine) prepareRequest(ctx context.Context, req RunRequest) (*domain.Pr
// ErrInvalidConfig. The effective transient retry budget is frozen privately // ErrInvalidConfig. The effective transient retry budget is frozen privately
// for RunPrepared and is absent from Details. A supplied Observer is retained // for RunPrepared and is absent from Details. A supplied Observer is retained
// privately for RunPrepared; preparation itself emits no events. // privately for RunPrepared; preparation itself emits no events.
// RunRequest.OperationTimeout and RunRequest.OnPrepared are retained without
// starting a timer or invoking the notification during preparation.
func (e *Engine) PrepareExecution(ctx context.Context, req RunRequest) (*PreparedExecution, error) { func (e *Engine) PrepareExecution(ctx context.Context, req RunRequest) (*PreparedExecution, error) {
ctx = observation.WithoutExecutionState(ctx) ctx = observation.WithoutExecutionState(ctx)
if e == nil || e.internal == nil { if e == nil || e.internal == nil {
@@ -439,6 +568,9 @@ func (e *Engine) PrepareExecution(ctx context.Context, req RunRequest) (*Prepare
// returns the final failed validation result with cumulative usage and actual // returns the final failed validation result with cumulative usage and actual
// repair attempts. An inability to generate or validate returns an error and // repair attempts. An inability to generate or validate returns an error and
// no partial result. // no partial result.
// RunRequest.OperationTimeout selects a cooperative whole-operation deadline;
// RunRequest.OnPrepared selects conditional notification before generation.
// The outcome and timeout attribution are fixed before terminal observation.
// //
// The built-in client retries eligible provider statuses using the effective // The built-in client retries eligible provider statuses using the effective
// Config or request budget. With R additional transient attempts and P repair // Config or request budget. With R additional transient attempts and P repair
@@ -470,18 +602,30 @@ func (e *Engine) PrepareExecution(ctx context.Context, req RunRequest) (*Prepare
// result on error. // result on error.
func (e *Engine) Run(ctx context.Context, req RunRequest) (*RunResult, error) { func (e *Engine) Run(ctx context.Context, req RunRequest) (*RunResult, error) {
ctx = observation.WithoutExecutionState(ctx) ctx = observation.WithoutExecutionState(ctx)
scope := usecase.NewOperationScope(ctx, req.OperationTimeout)
defer scope.Close()
ctx = scope.Context
if req.Observer == nil { if req.Observer == nil {
return e.run(ctx, req) result, err := e.run(ctx, req)
err = mapPublicError(scope.Finalize(err))
scope.Close()
return result, err
} }
recorder, idErr := observation.New(func(event domain.ExecutionEvent) { req.Observer(fromDomainExecutionEvent(event)) }) recorder, idErr := observation.New(func(event domain.ExecutionEvent) { req.Observer(fromDomainExecutionEvent(event)) })
if req.ContinueFrom != nil && req.ContinueFrom.internal != nil {
recorder.SetParentRunID(req.ContinueFrom.internal.RunID())
}
recorder.Start() recorder.Start()
if idErr != nil { if idErr != nil {
err := fmt.Errorf("failed to create run id: %w", idErr) err := mapPublicError(scope.Finalize(fmt.Errorf("failed to create run id: %w", idErr)))
scope.Close()
terminal := observedFinish(nil, err) terminal := observedFinish(nil, err)
recorder.Finish(observation.FinishData{Outcome: terminal.Outcome, Failure: terminal.Failure}) recorder.Finish(observation.FinishData{Outcome: terminal.Outcome, Failure: terminal.Failure})
return nil, err return nil, err
} }
result, err := e.run(observation.WithRecorder(ctx, recorder), req) result, err := e.run(observation.WithRecorder(ctx, recorder), req)
err = mapPublicError(scope.Finalize(err))
scope.Close()
terminal := observedFinish(result, err) terminal := observedFinish(result, err)
recorder.Finish(observation.FinishData{Outcome: terminal.Outcome, Failure: terminal.Failure, recorder.Finish(observation.FinishData{Outcome: terminal.Outcome, Failure: terminal.Failure,
StatusCode: terminal.StatusCode, RetriesExhausted: terminal.RetriesExhausted}) StatusCode: terminal.StatusCode, RetriesExhausted: terminal.RetriesExhausted})
@@ -500,7 +644,7 @@ func (e *Engine) run(ctx context.Context, req RunRequest) (*RunResult, error) {
result, err := e.internal.Run(ctx, domainReq) result, err := e.internal.Run(ctx, domainReq)
if err != nil { if err != nil {
return nil, mapPublicError(err) return nil, err
} }
return fromDomainRunResult(result), nil return fromDomainRunResult(result), nil
} }
@@ -544,6 +688,9 @@ func (e *Engine) run(ctx context.Context, req RunRequest) (*RunResult, error) {
// mapping. Invalid, foreign, discarded, or reused handles emit no events. // mapping. Invalid, foreign, discarded, or reused handles emit no events.
// Callbacks run outside the handle and capacity locks; panics propagate and // Callbacks run outside the handle and capacity locks; panics propagate and
// do not promise a terminal event. // do not promise a terminal event.
// The retained RunRequest.OperationTimeout starts after claim; the retained
// RunRequest.OnPrepared is notified only if execution reaches its prepared
// point. Final timeout attribution precedes terminal observation.
func (e *Engine) RunPrepared(ctx context.Context, prepared *PreparedExecution) (*RunResult, error) { func (e *Engine) RunPrepared(ctx context.Context, prepared *PreparedExecution) (*RunResult, error) {
ctx = observation.WithoutExecutionState(ctx) ctx = observation.WithoutExecutionState(ctx)
if e == nil || e.internal == nil { if e == nil || e.internal == nil {

View File

@@ -13,10 +13,48 @@ import (
"go.maximumdirect.net/promptkit/internal/usecase" "go.maximumdirect.net/promptkit/internal/usecase"
) )
// publicOperationError keeps classification separate from inspectable causes.
// Cache wording so formatting nested operations does not repeatedly rebuild it.
type publicOperationError struct {
owner error
cause error
message string
}
func newPublicOperationError(owner, cause error) *publicOperationError {
return &publicOperationError{owner: owner, cause: cause, message: fmt.Sprintf("%v: %v", owner, cause)}
}
func (e *publicOperationError) Error() string { return e.message }
func (e *publicOperationError) Unwrap() error { return e.cause }
func (e *publicOperationError) Is(target error) bool { return errors.Is(e.owner, target) }
func mapPublicError(err error) error { func mapPublicError(err error) error {
if err == nil { if err == nil {
return nil return nil
} }
var operation *usecase.OperationError
if errors.As(err, &operation) && operation != nil {
owner := publicErrorFor(operation.Category)
if owner != nil {
// Project the underlying specialized error separately, retaining both
// the outer operation and the original cause for Is/As callers.
mapped := mapPublicError(operation.Cause)
if errors.Is(operation.Cause, llm.ErrTransport) {
mapped = errors.Join(ErrTransport, mapped)
}
if errors.Is(operation.Cause, llm.ErrMalformedResponse) {
mapped = errors.Join(ErrMalformedResponse, mapped)
}
// The projection already retains ordinary causes; adding the original
// again would duplicate its entire subtree at each enclosing operation.
return newPublicOperationError(owner, mapped)
}
}
return mapUnownedPublicError(err)
}
func mapUnownedPublicError(err error) error {
var internalCapacityError *usecase.CapacityError var internalCapacityError *usecase.CapacityError
if errors.As(err, &internalCapacityError) && internalCapacityError != nil && if errors.As(err, &internalCapacityError) && internalCapacityError != nil &&
strings.TrimSpace(internalCapacityError.BackendID) != "" { strings.TrimSpace(internalCapacityError.BackendID) != "" {
@@ -62,6 +100,8 @@ func mapPublicError(err error) error {
func publicErrorFor(err error) error { func publicErrorFor(err error) error {
switch { switch {
case errors.Is(err, usecase.ErrOperationTimeout):
return ErrOperationTimeout
case errors.Is(err, promptdef.ErrPromptDefinitionNotFound): case errors.Is(err, promptdef.ErrPromptDefinitionNotFound):
return ErrPromptNotFound return ErrPromptNotFound
case errors.Is(err, profile.ErrProfileNotFound): case errors.Is(err, profile.ErrProfileNotFound):

View File

@@ -10,6 +10,25 @@ import (
"go.maximumdirect.net/promptkit/internal/usecase" "go.maximumdirect.net/promptkit/internal/usecase"
) )
func TestConstructionErrorPreservesJoinedCause(t *testing.T) {
operational := errors.New("independent operational failure")
cause := errors.Join(operational, context.Canceled)
engine, err := NewEngine(Config{}, optionFunc(func(*engineOptions) error {
return cause
}))
if engine != nil || err == nil {
t.Fatalf("construction = (%#v, %v), want failure", engine, err)
}
if unwrapped := errors.Unwrap(err); unwrapped != cause {
t.Fatalf("construction cause = %v, want the unchanged joined cause", unwrapped)
}
for _, target := range []error{ErrInvalidConfig, operational, context.Canceled} {
if !errors.Is(err, target) {
t.Errorf("construction error = %v, want identity %v", err, target)
}
}
}
func TestMapPublicErrorPreservesGenerationCancellation(t *testing.T) { func TestMapPublicErrorPreservesGenerationCancellation(t *testing.T) {
internalErr := fmt.Errorf("%w: %w", usecase.ErrLLMGenerate, context.Canceled) internalErr := fmt.Errorf("%w: %w", usecase.ErrLLMGenerate, context.Canceled)
@@ -80,3 +99,12 @@ func TestMapPublicErrorDoesNotTreatRequestEncodingAsTransport(t *testing.T) {
t.Fatalf("request encoding category = %v", err) t.Fatalf("request encoding category = %v", err)
} }
} }
func TestMapPublicErrorRetainsOwnerWithSpecializedCapacity(t *testing.T) {
internal := &usecase.CapacityError{BackendID: "limited"}
err := mapPublicError(&usecase.OperationError{Category: usecase.ErrLLMGenerate, Cause: internal})
var capacity *CapacityError
if !errors.Is(err, ErrLLMGenerate) || !errors.As(err, &capacity) || capacity.BackendID != "limited" || !errors.Is(err, ErrCapacityExceeded) {
t.Fatalf("mapped owned capacity = %v", err)
}
}

View File

@@ -9,6 +9,7 @@ import (
"net/http" "net/http"
"os" "os"
"strings" "strings"
"time"
"go.maximumdirect.net/configkit" "go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit" "go.maximumdirect.net/promptkit"
@@ -69,12 +70,18 @@ func (l *eventLog) terminal() (promptkit.ExecutionEvent, error) {
} }
type summary struct { type summary struct {
ProposalAttempts uint64 `json:"proposal_attempts"` ProposalAttempts uint64 `json:"proposal_attempts"`
ProposalTransientRetries uint64 `json:"proposal_transient_retries"` ProposalTransientRetries uint64 `json:"proposal_transient_retries"`
ProposalUsageComplete bool `json:"proposal_usage_complete"` ProposalUsageComplete bool `json:"proposal_usage_complete"`
AssessmentAttempts uint64 `json:"assessment_attempts"` AssessmentAttempts uint64 `json:"assessment_attempts"`
AssessmentTransientRetries uint64 `json:"assessment_transient_retries"` AssessmentTransientRetries uint64 `json:"assessment_transient_retries"`
AssessmentOutcome string `json:"assessment_outcome"` AssessmentOutcome string `json:"assessment_outcome"`
CoordinationOutcome promptkit.CoordinationOutcome `json:"coordination_outcome"`
ExhaustedOutcome promptkit.CoordinationOutcome `json:"exhausted_outcome"`
GroupMembers int `json:"group_members"`
GroupAttempts uint64 `json:"group_attempts"`
PreparedMessages int `json:"prepared_messages"`
PreparedParentMatches bool `json:"prepared_parent_matches"`
} }
func run() (summary, error) { func run() (summary, error) {
@@ -83,29 +90,35 @@ func run() (summary, error) {
if err != nil { if err != nil {
return summary{}, err return summary{}, err
} }
for i := range inputs {
inputs[i].Destination = configkit.Keys("promptkit")
}
backendID, proposalModel, assessmentModel := "offline-backend", "proposal-model", "assessment-model" backendID, proposalModel, assessmentModel := "offline-backend", "proposal-model", "assessment-model"
filesystem := configkit.NewFilesystem("example schemas", assets)
backend := promptkit.LocalBackend("https://example.invalid/v1", 1) backend := promptkit.LocalBackend("https://example.invalid/v1", 1)
backend.ID = backendID backend.ID = backendID
inputs = append(inputs, inputs = append(inputs,
configkit.Input{Source: promptkit.BackendSource("example backend", []promptkit.BackendConfig{ configkit.Input{Source: promptkit.SchemaRootSource("example schemas", "schemas"), Destination: configkit.Keys("promptkit"), Filesystem: &configkit.FilesystemContext{FS: &filesystem, Root: ".", BaseDir: "."}},
promptkit.SettingsInput(promptkit.BackendSource("example backend", []promptkit.BackendConfig{
backend, backend,
})}, })),
configkit.Input{Source: promptkit.ProfileSource("example profiles", []promptkit.ProfileConfig{ promptkit.SettingsInput(promptkit.ProfileSource("example profiles", []promptkit.ProfileConfig{
{ID: "proposal-profile", Backend: &backendID, Model: &proposalModel}, {ID: "proposal-profile", Backend: &backendID, Model: &proposalModel},
{ID: "assessment-profile", Backend: &backendID, Model: &assessmentModel}, {ID: "assessment-profile", Backend: &backendID, Model: &assessmentModel},
})}, })),
) )
engine, err := promptkit.NewEngine(promptkit.Config{ engine, err := promptkit.NewEngine(promptkit.Config{
Sources: promptkit.ConfigurationSources{Underlays: []configkit.Underlay{{Inputs: inputs}}}, Sources: promptkit.ConfigurationSources{Underlays: []configkit.Underlay{{Inputs: inputs}}},
HTTPClient: &http.Client{Transport: transport}, HTTPClient: &http.Client{Transport: transport},
}, promptkit.WithSchemaFS(assets, "schemas")) }, promptkit.WithBundledOpenRouter(false)) // Keep the example's configuration self-contained.
if err != nil { if err != nil {
return summary{}, err return summary{}, err
} }
var proposalLog eventLog var proposalLog eventLog
var executions promptkit.ExecutionGroup
proposal, err := engine.Run(context.Background(), promptkit.RunRequest{ proposal, err := engine.Run(context.Background(), promptkit.RunRequest{
PromptID: "example.proposal", Inputs: map[string]promptkit.ArtifactRef{"idea": promptkit.Inline("Reduce queue cleanup toil.")}, Observer: proposalLog.record, PromptID: "example.proposal", Inputs: map[string]promptkit.ArtifactRef{"idea": promptkit.Inline("Reduce queue cleanup toil.")}, CaptureExchange: true, Observer: promptkit.ComposeObservers(proposalLog.record, executions.Observe),
}) })
if err != nil { if err != nil {
return summary{}, err return summary{}, err
@@ -118,35 +131,56 @@ func run() (summary, error) {
return summary{}, err return summary{}, err
} }
var assessmentLog eventLog var assessmentLog eventLog
assessment, err := engine.Run(context.Background(), promptkit.RunRequest{ var preparedDetails promptkit.PreparedRun
PromptID: "example.assessment", Inputs: map[string]promptkit.ArtifactRef{"proposal": promptkit.Inline(proposal.RawOutput)}, Observer: assessmentLog.record, decision := "rejected"
coordination, err := engine.Coordinate(context.Background(), promptkit.CoordinateRequest{
Start: proposal.Exchange, AdditionalTurns: 1, Observer: executions.Observe,
Next: func(ctx context.Context, turn promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
if turn.Result == nil {
return &promptkit.RunRequest{PromptID: "example.proposal", Inputs: map[string]promptkit.ArtifactRef{"idea": promptkit.Inline("Refine the proposal.")}, OperationTimeout: 10 * time.Second, OnPrepared: func(_ context.Context, details promptkit.PreparedRun) { preparedDetails = details }}, nil
}
if !turn.Result.Validation.IsValid {
return nil, fmt.Errorf("follow-up failed structural validation")
}
assessment, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "example.assessment", Inputs: map[string]promptkit.ArtifactRef{"proposal": promptkit.Inline(turn.Result.RawOutput)}, OperationTimeout: 10 * time.Second, Observer: promptkit.ComposeObservers(assessmentLog.record, executions.Observe)})
if err != nil {
return nil, err
}
if !assessment.Validation.IsValid {
return nil, fmt.Errorf("assessment failed structural validation")
}
var proposalValue struct {
Risk string `json:"risk"`
}
var assessmentValue struct {
Recommendation string `json:"recommendation"`
}
if err := json.Unmarshal([]byte(turn.Result.RawOutput), &proposalValue); err != nil {
return nil, err
}
if err := json.Unmarshal([]byte(assessment.RawOutput), &assessmentValue); err != nil {
return nil, err
}
if proposalValue.Risk == "low" && assessmentValue.Recommendation == "accept" {
decision = "accepted"
}
return nil, nil
},
}) })
if err != nil { if err != nil {
return summary{}, err return summary{}, err
} }
if !assessment.Validation.IsValid { exhausted, err := engine.Coordinate(context.Background(), promptkit.CoordinateRequest{Start: proposal.Exchange, Next: func(context.Context, promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
return summary{}, fmt.Errorf("assessment failed structural validation") return &promptkit.RunRequest{PromptID: "example.proposal"}, nil
}})
if err != nil {
return summary{}, err
} }
assessmentTerminal, err := assessmentLog.terminal() assessmentTerminal, err := assessmentLog.terminal()
if err != nil { if err != nil {
return summary{}, err return summary{}, err
} }
var proposalValue struct { group := executions.Snapshot()
Risk string `json:"risk"`
}
var assessmentValue struct {
Recommendation string `json:"recommendation"`
}
if err := json.Unmarshal([]byte(proposal.RawOutput), &proposalValue); err != nil {
return summary{}, err
}
if err := json.Unmarshal([]byte(assessment.RawOutput), &assessmentValue); err != nil {
return summary{}, err
}
decision := "rejected"
if proposalValue.Risk == "low" && assessmentValue.Recommendation == "accept" {
decision = "accepted"
}
return summary{ return summary{
ProposalAttempts: proposalTerminal.Attempts, ProposalAttempts: proposalTerminal.Attempts,
ProposalTransientRetries: proposalTerminal.TransientRetries, ProposalTransientRetries: proposalTerminal.TransientRetries,
@@ -154,6 +188,12 @@ func run() (summary, error) {
AssessmentAttempts: assessmentTerminal.Attempts, AssessmentAttempts: assessmentTerminal.Attempts,
AssessmentTransientRetries: assessmentTerminal.TransientRetries, AssessmentTransientRetries: assessmentTerminal.TransientRetries,
AssessmentOutcome: decision, AssessmentOutcome: decision,
CoordinationOutcome: coordination.Outcome,
ExhaustedOutcome: exhausted.Outcome,
GroupMembers: len(group.Members),
GroupAttempts: group.Attempts,
PreparedMessages: len(preparedDetails.Messages),
PreparedParentMatches: preparedDetails.ParentRunID == proposal.RunID,
}, nil }, nil
} }

View File

@@ -12,19 +12,25 @@ func TestObserveExample(t *testing.T) {
t.Fatal(err) t.Fatal(err)
} }
var got struct { var got struct {
ProposalAttempts *int `json:"proposal_attempts"` ProposalAttempts *int `json:"proposal_attempts"`
ProposalRetries *int `json:"proposal_transient_retries"` ProposalRetries *int `json:"proposal_transient_retries"`
UsageComplete *bool `json:"proposal_usage_complete"` UsageComplete *bool `json:"proposal_usage_complete"`
AssessmentAttempts *int `json:"assessment_attempts"` AssessmentAttempts *int `json:"assessment_attempts"`
AssessmentRetries *int `json:"assessment_transient_retries"` AssessmentRetries *int `json:"assessment_transient_retries"`
Outcome string `json:"assessment_outcome"` Outcome string `json:"assessment_outcome"`
CoordinationOutcome string `json:"coordination_outcome"`
ExhaustedOutcome string `json:"exhausted_outcome"`
GroupMembers int `json:"group_members"`
GroupAttempts int `json:"group_attempts"`
PreparedMessages int `json:"prepared_messages"`
PreparedParentMatches bool `json:"prepared_parent_matches"`
} }
if err := json.Unmarshal(output.Bytes(), &got); err != nil { if err := json.Unmarshal(output.Bytes(), &got); err != nil {
t.Fatal(err) t.Fatal(err)
} }
if got.ProposalAttempts == nil || *got.ProposalAttempts != 2 || got.ProposalRetries == nil || *got.ProposalRetries != 1 || if got.ProposalAttempts == nil || *got.ProposalAttempts != 2 || got.ProposalRetries == nil || *got.ProposalRetries != 1 ||
got.UsageComplete == nil || *got.UsageComplete || got.AssessmentAttempts == nil || *got.AssessmentAttempts != 1 || got.UsageComplete == nil || *got.UsageComplete || got.AssessmentAttempts == nil || *got.AssessmentAttempts != 1 ||
got.AssessmentRetries == nil || *got.AssessmentRetries != 0 || got.Outcome != "accepted" { got.AssessmentRetries == nil || *got.AssessmentRetries != 0 || got.Outcome != "accepted" || got.CoordinationOutcome != "stopped" || got.ExhaustedOutcome != "exhausted" || got.GroupMembers != 3 || got.GroupAttempts != 4 || got.PreparedMessages < 3 || !got.PreparedParentMatches {
t.Fatalf("unexpected observation summary: %+v", got) t.Fatalf("unexpected observation summary: %+v", got)
} }
} }

View File

@@ -14,4 +14,12 @@ prompts:
output: output:
format: json format: json
validation_mode: json_schema validation_mode: json_schema
schema_path: assessment.schema.json repair_attempts: 0
schema:
$schema: https://json-schema.org/draft/2020-12/schema
type: object
required: [recommendation, confidence]
properties:
recommendation: {type: string}
confidence: {type: number}
additionalProperties: false

View File

@@ -14,4 +14,5 @@ prompts:
output: output:
format: json format: json
validation_mode: json_schema validation_mode: json_schema
repair_attempts: 0
schema_path: proposal.schema.json schema_path: proposal.schema.json

View File

@@ -1,10 +0,0 @@
{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"type": "object",
"required": ["recommendation", "confidence"],
"properties": {
"recommendation": {"type": "string"},
"confidence": {"type": "number"}
},
"additionalProperties": false
}

View File

@@ -1,4 +1,3 @@
version: "2"
config: config:
imports: imports:
- path: fragments/resources.yaml - path: fragments/resources.yaml

View File

@@ -13,15 +13,16 @@ import (
) )
type summary struct { type summary struct {
PromptID string `json:"prompt_id"` PromptID string `json:"prompt_id"`
SelectedProfile string `json:"selected_profile"` SelectedProfile string `json:"selected_profile"`
Model string `json:"model"` Model string `json:"model"`
MessageCount int `json:"message_count"` MessageCount int `json:"message_count"`
ProfileTimeout int `json:"profile_timeout_seconds"` MessageContentHash string `json:"message_content_hash"`
Namespaces []string `json:"namespaces"` ProfileTimeout int `json:"profile_timeout_seconds"`
Collections []collectionSummary `json:"collections"` Namespaces []string `json:"namespaces"`
ScopedPrompts []string `json:"scoped_prompts"` Collections []collectionSummary `json:"collections"`
ReviewPrompts []string `json:"review_prompts"` ScopedPrompts []string `json:"scoped_prompts"`
ReviewPrompts []string `json:"review_prompts"`
} }
type collectionSummary struct { type collectionSummary struct {
@@ -29,6 +30,10 @@ type collectionSummary struct {
ReleaseVersion *string `json:"release_version"` ReleaseVersion *string `json:"release_version"`
} }
type applicationSettings struct {
PromptKit promptkit.Settings `yaml:"promptkit"`
}
func main() { func main() {
if err := run(context.Background(), "examples/go-library/prepare", os.Stdout); err != nil { if err := run(context.Background(), "examples/go-library/prepare", os.Stdout); err != nil {
fmt.Fprintln(os.Stderr, err) fmt.Fprintln(os.Stderr, err)
@@ -58,12 +63,18 @@ func run(ctx context.Context, exampleDir string, out io.Writer) error {
if err != nil { if err != nil {
return err return err
} }
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{ definition := promptkit.Configure(configkit.Definition[applicationSettings]{}, func(value *applicationSettings) *promptkit.Settings { return &value.PromptKit })
result, err := configkit.LoadLayered(configkit.Input{Source: configkit.NoSource()}, configkit.LayeredOptions[applicationSettings]{
Definition: definition,
Underlays: []configkit.Underlay{ Underlays: []configkit.Underlay{
{Inputs: []configkit.Input{support, review}}, {Inputs: []configkit.Input{support, review}},
{Inputs: []configkit.Input{override}}, {Inputs: []configkit.Input{override}},
}, },
}}) })
if err != nil {
return err
}
engine, err := promptkit.NewFromConfig(result, promptkit.RuntimeConfig{})
if err != nil { if err != nil {
return err return err
} }
@@ -77,7 +88,7 @@ func run(ctx context.Context, exampleDir string, out io.Writer) error {
if err != nil { if err != nil {
return err return err
} }
if _, found := view.Explain(configkit.Keys("collections", "support", "profiles").Index(1).Key("timeout_seconds")); !found { if _, found := view.Explain(configkit.Keys("promptkit", "collections", "support", "profiles").Index(1).Key("timeout_seconds")); !found {
return fmt.Errorf("collection profile override was not found") return fmt.Errorf("collection profile override was not found")
} }
requirements, err := scoped.InspectRequirements(ctx, promptkit.RequirementInspectionRequest{ requirements, err := scoped.InspectRequirements(ctx, promptkit.RequirementInspectionRequest{
@@ -110,10 +121,10 @@ func run(ctx context.Context, exampleDir string, out io.Writer) error {
if err != nil { if err != nil {
return err return err
} }
if profile.EffectiveModelParams.TimeoutSeconds != 30 { if profile.TimeoutSeconds == nil || *profile.TimeoutSeconds != 30 {
return fmt.Errorf( return fmt.Errorf(
"later profile override was not applied: timeout is %d seconds", "later profile override was not applied: timeout is %v",
profile.EffectiveModelParams.TimeoutSeconds, profile.TimeoutSeconds,
) )
} }
namespaces, err := scoped.ListNamespaces(ctx) namespaces, err := scoped.ListNamespaces(ctx)
@@ -152,14 +163,15 @@ func run(ctx context.Context, exampleDir string, out io.Writer) error {
encoder := json.NewEncoder(out) encoder := json.NewEncoder(out)
encoder.SetIndent("", " ") encoder.SetIndent("", " ")
return encoder.Encode(summary{ return encoder.Encode(summary{
PromptID: prepared.PromptID, PromptID: prepared.PromptID,
SelectedProfile: prepared.SelectedProfileID, SelectedProfile: prepared.SelectedProfileID,
Model: prepared.EffectiveModelParams.Model, Model: prepared.EffectiveModelParams.Model,
MessageCount: len(prepared.Messages), MessageCount: len(prepared.Messages),
ProfileTimeout: profile.EffectiveModelParams.TimeoutSeconds, MessageContentHash: prepared.MessageContentHash,
Namespaces: namespaces, ProfileTimeout: *profile.TimeoutSeconds,
Collections: collectionSummaries, Namespaces: namespaces,
ScopedPrompts: scopedPromptIDs, Collections: collectionSummaries,
ReviewPrompts: reviewPromptIDs, ScopedPrompts: scopedPromptIDs,
ReviewPrompts: reviewPromptIDs,
}) })
} }

View File

@@ -13,21 +13,22 @@ func TestPrepareExample(t *testing.T) {
t.Fatal(err) t.Fatal(err)
} }
var got struct { var got struct {
PromptID string `json:"prompt_id"` PromptID string `json:"prompt_id"`
Profile string `json:"selected_profile"` Profile string `json:"selected_profile"`
Model string `json:"model"` Model string `json:"model"`
MessageCount int `json:"message_count"` MessageCount int `json:"message_count"`
Timeout int `json:"profile_timeout_seconds"` MessageContentHash string `json:"message_content_hash"`
Namespaces []string `json:"namespaces"` Timeout int `json:"profile_timeout_seconds"`
Collections []map[string]json.RawMessage `json:"collections"` Namespaces []string `json:"namespaces"`
ScopedPrompts []string `json:"scoped_prompts"` Collections []map[string]json.RawMessage `json:"collections"`
ReviewPrompts []string `json:"review_prompts"` ScopedPrompts []string `json:"scoped_prompts"`
ReviewPrompts []string `json:"review_prompts"`
} }
if err := json.Unmarshal(output.Bytes(), &got); err != nil { if err := json.Unmarshal(output.Bytes(), &got); err != nil {
t.Fatal(err) t.Fatal(err)
} }
if got.PromptID != "support/prepare" || got.Profile != "support/fast" || got.Model != "offline-model" || if got.PromptID != "support/prepare" || got.Profile != "support/fast" || got.Model != "offline-model" ||
got.MessageCount <= 0 || got.Timeout != 30 || !reflect.DeepEqual(got.Namespaces, []string{"review", "support"}) || got.MessageCount <= 0 || got.MessageContentHash == "" || got.Timeout != 30 || !reflect.DeepEqual(got.Namespaces, []string{"review", "support"}) ||
!reflect.DeepEqual(got.ScopedPrompts, []string{"support/prepare"}) || !reflect.DeepEqual(got.ReviewPrompts, []string{"review/check"}) || len(got.Collections) != 2 { !reflect.DeepEqual(got.ScopedPrompts, []string{"support/prepare"}) || !reflect.DeepEqual(got.ReviewPrompts, []string{"review/check"}) || len(got.Collections) != 2 {
t.Fatalf("unexpected preparation summary: %+v", got) t.Fatalf("unexpected preparation summary: %+v", got)
} }

View File

@@ -50,11 +50,11 @@ func main() {
func run(ctx context.Context, exampleDir string, out io.Writer) error { func run(ctx context.Context, exampleDir string, out io.Writer) error {
endpoint, model := "https://example.invalid/v1", "offline-model" endpoint, model := "https://example.invalid/v1", "offline-model"
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
Underlays: []configkit.Underlay{{Inputs: []configkit.Input{{Source: promptkit.ProfileSource("example profile", []promptkit.ProfileConfig{ Underlays: []configkit.Underlay{{Inputs: []configkit.Input{promptkit.SettingsInput(promptkit.ProfileSource("example profile", []promptkit.ProfileConfig{
{ID: "offline-example", Endpoint: &endpoint, Model: &model}, {ID: "offline-example", Endpoint: &endpoint, Model: &model},
})}}}}, }))}}},
Main: &configkit.Input{Source: configkit.File(filepath.Join(exampleDir, "prompt.yaml"))}, Main: &configkit.Input{Source: configkit.File(filepath.Join(exampleDir, "prompt.yaml")), Destination: configkit.Keys("promptkit")},
}}, promptkit.WithLLMClient(deterministicClient{})) }}, promptkit.WithBundledOpenRouter(false), promptkit.WithLLMClient(deterministicClient{})) // Use only this example's offline configuration.
if err != nil { if err != nil {
return err return err
} }

47
exchange.go Normal file
View File

@@ -0,0 +1,47 @@
package promptkit
import "go.maximumdirect.net/promptkit/internal/domain"
const exchangeString = "promptkit.Exchange{opaque}"
// Exchange is an immutable, reusable in-process snapshot of a completed
// conversation. It is not a persistence or restart format. Different engines
// can continue it, and independent branches do not alter one another.
//
// History preserves the effective request's message content, roles, and cache
// metadata, followed by the exact final [RunResult.RawOutput] as an assistant
// message. Intermediate structural-repair candidates and corrective instructions
// are excluded. A continuation extends that logical history according to
// [RunRequest.ContinueFrom], without changing its source exchange.
type Exchange struct{ internal *domain.Exchange }
// ExchangeDetails contains a detached copy of an exchange's logical history.
type ExchangeDetails struct {
// RunID identifies the execution that produced this exchange.
RunID string `json:"run_id"`
// ParentRunID identifies the immediate source execution, or is empty.
ParentRunID string `json:"parent_run_id,omitempty"`
Messages []RenderedMessage `json:"messages"`
}
// Details returns detached history, or zero details for a nil or empty handle.
// Messages include caller and model content and require appropriate handling.
// Inspection uses only retained state, without rendering, filesystem reads, or
// provider calls. Mutating returned messages or cache metadata cannot affect
// this exchange, another branch, or subsequent Details calls.
func (e *Exchange) Details() ExchangeDetails {
if e == nil || e.internal == nil {
return ExchangeDetails{}
}
messages := make([]RenderedMessage, e.internal.MessageCount())
e.internal.VisitMessages(func(index int, message domain.RenderedMessage) {
messages[index] = RenderedMessage{Role: message.Role, Content: message.Content, CacheControl: fromDomainCacheControl(message.CacheControl)}
})
return ExchangeDetails{RunID: e.internal.RunID(), ParentRunID: e.internal.ParentRunID(), Messages: messages}
}
// String returns an opaque representation without conversation content.
func (e Exchange) String() string { return exchangeString }
// GoString returns an opaque representation without conversation content.
func (e Exchange) GoString() string { return exchangeString }

View File

@@ -0,0 +1,113 @@
package promptkit_test
import (
"context"
"encoding/json"
"errors"
"reflect"
"strings"
"testing"
"go.maximumdirect.net/promptkit"
)
func TestExchangeCaptureAndPreparedContinuation(t *testing.T) {
client := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: "answer"}}
engine := newSimpleEngine(t, "question", promptkit.WithLLMClient(client))
ctx := context.Background()
plain, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt"})
if err != nil || plain.Exchange != nil {
t.Fatalf("plain run = %#v, %v", plain, err)
}
initial, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true})
if err != nil || initial.Exchange == nil {
t.Fatalf("captured run = %#v, %v", initial, err)
}
want := []promptkit.RenderedMessage{{Role: promptkit.RoleUser, Content: "question"}, {Role: promptkit.RoleAssistant, Content: "answer"}}
if details := initial.Exchange.Details(); details.RunID != initial.RunID || !reflect.DeepEqual(details.Messages, want) {
t.Fatalf("details = %#v", details)
}
encoded, err := json.Marshal(initial)
if err != nil || strings.Contains(string(encoded), "exchange") || strings.Contains(string(encoded), "question") {
t.Fatalf("result JSON = %s, %v", encoded, err)
}
handle, err := engine.PrepareExecution(ctx, promptkit.RunRequest{PromptID: "prompt", ContinueFrom: initial.Exchange})
if err != nil {
t.Fatal(err)
}
if handle.Details().ParentRunID != initial.RunID {
t.Fatal("prepared ancestry missing")
}
continued, err := engine.RunPrepared(ctx, handle)
if err != nil || continued.Exchange == nil || continued.ParentRunID != initial.RunID {
t.Fatalf("continued = %#v, %v", continued, err)
}
want = append(want, promptkit.RenderedMessage{Role: promptkit.RoleUser, Content: "question"}, promptkit.RenderedMessage{Role: promptkit.RoleAssistant, Content: "answer"})
if details := continued.Exchange.Details(); details.ParentRunID != initial.RunID || !reflect.DeepEqual(details.Messages, want) {
t.Fatalf("continued details = %#v", details)
}
}
func TestExchangeCapturesInvalidFinalOutput(t *testing.T) {
client := &fakeLLMClient{responses: []*promptkit.GenerateResponse{{Content: ""}, {Content: " "}}}
engine := newSimpleEngine(t, "question", promptkit.WithLLMClient(client))
result, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true,
Validation: &promptkit.OutputContractInput{Format: promptkit.FormatText, ValidationMode: promptkit.ValidationBasic, RepairAttempts: new(1)}})
if err != nil || result == nil || result.Validation.Status != promptkit.ValidationFailed || result.Exchange == nil {
t.Fatalf("invalid result = %#v, %v", result, err)
}
if messages := result.Exchange.Details().Messages; len(messages) != 2 || messages[1].Content != " " {
t.Fatalf("history = %#v", messages)
}
}
func TestExchangeFailuresLeaveSourceReusable(t *testing.T) {
client := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: "answer"}}
engine := newSimpleEngine(t, "question", promptkit.WithLLMClient(client))
ctx := context.Background()
initial, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true})
if err != nil {
t.Fatal(err)
}
original := initial.Exchange.Details()
request := promptkit.RunRequest{PromptID: "prompt", ContinueFrom: initial.Exchange}
handle, err := engine.PrepareExecution(ctx, request)
if err != nil {
t.Fatal(err)
}
handle.Discard()
if result, err := engine.RunPrepared(ctx, handle); result != nil || !errors.Is(err, promptkit.ErrInvalidRequest) {
t.Fatalf("discarded = %#v, %v", result, err)
}
client.err = errors.New("provider failure")
if result, err := engine.Run(ctx, request); result != nil || !errors.Is(err, promptkit.ErrLLMGenerate) {
t.Fatalf("failure = %#v, %v", result, err)
}
client.err = nil
result, err := engine.Run(ctx, request)
if err != nil || result.Exchange == nil || !reflect.DeepEqual(initial.Exchange.Details(), original) {
t.Fatalf("reused source = %#v, %v", result, err)
}
}
func TestEmptyExchangeRejectedBeforePreparation(t *testing.T) {
client := &fakeLLMClient{response: &promptkit.GenerateResponse{Content: "answer"}}
engine := newSimpleEngine(t, "question", promptkit.WithLLMClient(client))
request := promptkit.RunRequest{PromptID: "missing", ContinueFrom: &promptkit.Exchange{}}
ctx := context.Background()
if _, err := engine.Prepare(ctx, request); !errors.Is(err, promptkit.ErrInvalidRequest) {
t.Fatalf("prepare error = %v", err)
}
if err := engine.ValidateRequest(ctx, request); !errors.Is(err, promptkit.ErrInvalidRequest) {
t.Fatalf("validate error = %v", err)
}
if _, err := engine.PrepareExecution(ctx, request); !errors.Is(err, promptkit.ErrInvalidRequest) {
t.Fatalf("prepare execution error = %v", err)
}
if _, err := engine.Run(ctx, request); !errors.Is(err, promptkit.ErrInvalidRequest) {
t.Fatalf("run error = %v", err)
}
if len(client.requests) != 0 {
t.Fatal("invalid exchange reached the model")
}
}

38
exchange_test.go Normal file
View File

@@ -0,0 +1,38 @@
package promptkit
import (
"encoding/json"
"fmt"
"strings"
"testing"
"go.maximumdirect.net/promptkit/internal/domain"
)
func TestExchangeDetailsAndRequestConversion(t *testing.T) {
var nilExchange *Exchange
if got := nilExchange.Details(); got.RunID != "" || got.Messages != nil {
t.Fatalf("nil details = %#v", got)
}
if _, err := toDomainRunRequest(RunRequest{ContinueFrom: &Exchange{}}); err == nil {
t.Fatal("empty exchange accepted")
}
history := domain.CaptureExchange(nil, "source", []domain.RenderedMessage{{Role: domain.RoleUser, Content: "private", CacheControl: &domain.CacheControl{Type: domain.CacheControlEphemeral, TTL: "1h"}}}, "answer")
handle := &Exchange{internal: history}
request, err := toDomainRunRequest(RunRequest{ContinueFrom: handle})
if err != nil || request.ContinueFrom != history || !request.CaptureExchange {
t.Fatalf("request = %#v, error = %v", request, err)
}
details := handle.Details()
details.Messages[0].CacheControl.TTL = "changed"
if handle.Details().Messages[0].CacheControl.TTL != "1h" {
t.Fatal("details changed exchange")
}
if strings.Contains(fmt.Sprintf("%v %#v", handle, handle), "private") {
t.Fatal("format exposed history")
}
encoded, err := json.Marshal(PreparedRun{ParentRunID: "source"})
if err != nil || !strings.Contains(string(encoded), `"parent_run_id":"source"`) {
t.Fatalf("prepared JSON = %s, %v", encoded, err)
}
}

View File

@@ -0,0 +1,192 @@
package promptkit_test
import (
"context"
"errors"
"net/http"
"testing"
"testing/synctest"
"time"
"go.maximumdirect.net/promptkit"
)
func TestOperationBudgetCoversPreparationAndReleasesAdmission(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
release := make(chan struct{})
reader := &capacityArtifactReader{entered: make(chan struct{}, 2), release: release}
client := newCapacityGateClient(closedCapacityChannel(), 2)
engine := newBackendCapacityEngine(t, client, 1, capacityInt(0), reader)
parent := context.Background()
req := capacityInputRequest("http://example.invalid/v1")
req.OperationTimeout = time.Second
_, err := engine.Run(parent, req)
if !errors.Is(err, promptkit.ErrOperationTimeout) || !errors.Is(err, context.DeadlineExceeded) || !errors.Is(err, promptkit.ErrArtifactLoad) || parent.Err() != nil {
t.Fatalf("preparation error=%v parent=%v", err, parent.Err())
}
close(release)
if result, err := engine.Run(parent, req); err != nil || result == nil {
t.Fatalf("capacity not released: %v", err)
}
})
}
func TestOperationBudgetCoversGenerationCapacityWait(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
release := make(chan struct{})
client := newCapacityGateClient(release, 2)
engine := newBackendCapacityEngine(t, client, 1, capacityInt(1), nil)
results := make(chan capacityRunResult, 1)
go runCapacityRequest(engine, context.Background(), promptkit.RunRequest{PromptID: "prompt"}, results)
<-client.started
_, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt", OperationTimeout: time.Second})
if !errors.Is(err, promptkit.ErrOperationTimeout) || !errors.Is(err, promptkit.ErrLLMGenerate) || !errors.Is(err, context.DeadlineExceeded) {
t.Fatalf("capacity wait error=%v", err)
}
close(release)
if result := <-results; result.err != nil {
t.Fatal(result.err)
}
if _, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt"}); err != nil {
t.Fatal(err)
}
})
}
func TestOperationBudgetInterruptsRetryAndRepair(t *testing.T) {
for _, repair := range []bool{false, true} {
t.Run(map[bool]string{false: "retry", true: "repair"}[repair], func(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
calls := 0
config := simpleConfig("message")
one := 1
config.TransientRetries = &one
config.HTTPClient = &http.Client{Transport: roundTripFunc(func(r *http.Request) (*http.Response, error) {
calls++
if repair && calls == 1 {
return retryHTTPResponse(200, "not json"), nil
}
if repair {
<-r.Context().Done()
return nil, r.Context().Err()
}
response := retryHTTPResponse(503, "")
response.Header = make(http.Header)
response.Header.Set("Retry-After", "5")
return response, nil
})}
engine, err := promptkit.NewEngine(config)
if err != nil {
t.Fatal(err)
}
req := simpleGenerationRequest()
req.OperationTimeout = time.Second
if repair {
req.Validation = &promptkit.OutputContractInput{Format: promptkit.FormatJSON, ValidationMode: promptkit.ValidationJSON, RepairAttempts: new(1)}
}
_, err = engine.Run(context.Background(), req)
if !errors.Is(err, promptkit.ErrOperationTimeout) || !errors.Is(err, promptkit.ErrLLMGenerate) || !errors.Is(err, context.DeadlineExceeded) {
t.Fatalf("error=%v calls=%d", err, calls)
}
})
})
}
}
func TestOperationOutcomeFrozenBeforeTerminalObserver(t *testing.T) {
for _, split := range []bool{false, true} {
t.Run(map[bool]string{false: "run", true: "prepared"}[split], func(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
engine := observationEngine(t, observationClientFunc(func(context.Context, promptkit.GenerateRequest) (*promptkit.GenerateResponse, error) {
return &promptkit.GenerateResponse{Content: "ok"}, nil
}))
var terminal promptkit.ExecutionEvent
req := promptkit.RunRequest{PromptID: "prompt", OperationTimeout: time.Second, Observer: func(event promptkit.ExecutionEvent) {
if event.Kind == promptkit.ExecutionFinished {
time.Sleep(2 * time.Second)
terminal = event
}
}}
var result *promptkit.RunResult
var err error
if split {
var handle *promptkit.PreparedExecution
handle, err = engine.PrepareExecution(context.Background(), req)
if err == nil {
result, err = engine.RunPrepared(context.Background(), handle)
}
} else {
result, err = engine.Run(context.Background(), req)
}
if err != nil || result == nil || terminal.Outcome != promptkit.ExecutionOutcomeSuccess {
t.Fatalf("result=%v error=%v terminal=%+v", result, err, terminal)
}
})
})
}
}
func TestNotificationConsumesOperationBudget(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
engine := observationEngine(t, observationClientFunc(func(context.Context, promptkit.GenerateRequest) (*promptkit.GenerateResponse, error) {
t.Fatal("generation ran")
return nil, nil
}))
_, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt", OperationTimeout: time.Second, OnPrepared: func(ctx context.Context, _ promptkit.PreparedRun) { <-ctx.Done() }})
if !errors.Is(err, promptkit.ErrOperationTimeout) || !errors.Is(err, promptkit.ErrLLMGenerate) || !errors.Is(err, context.DeadlineExceeded) {
t.Fatal(err)
}
})
}
func TestOperationTimeoutDoesNotRelabelOtherFailures(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
failure := errors.New("independent failure")
engine := observationEngine(t, observationClientFunc(func(ctx context.Context, _ promptkit.GenerateRequest) (*promptkit.GenerateResponse, error) {
<-ctx.Done()
return nil, failure
}))
var terminal promptkit.ExecutionEvent
_, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt", OperationTimeout: time.Second, Observer: func(event promptkit.ExecutionEvent) {
if event.Kind == promptkit.ExecutionFinished {
terminal = event
}
}})
if !errors.Is(err, failure) || errors.Is(err, promptkit.ErrOperationTimeout) || terminal.Failure != promptkit.ExecutionFailureGeneration {
t.Fatalf("error=%v terminal=%+v", err, terminal)
}
})
}
func TestOperationTimeoutTerminalObserverSeesFinalFailure(t *testing.T) {
for _, split := range []bool{false, true} {
t.Run(map[bool]string{false: "run", true: "prepared"}[split], func(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
engine := observationEngine(t, observationClientFunc(func(ctx context.Context, _ promptkit.GenerateRequest) (*promptkit.GenerateResponse, error) {
<-ctx.Done()
return nil, ctx.Err()
}))
var terminal promptkit.ExecutionEvent
req := promptkit.RunRequest{PromptID: "prompt", OperationTimeout: time.Second, Observer: func(event promptkit.ExecutionEvent) {
if event.Kind == promptkit.ExecutionFinished {
time.Sleep(time.Second)
terminal = event
}
}}
var err error
if split {
var handle *promptkit.PreparedExecution
handle, err = engine.PrepareExecution(context.Background(), req)
if err == nil {
_, err = engine.RunPrepared(context.Background(), handle)
}
} else {
_, err = engine.Run(context.Background(), req)
}
if !errors.Is(err, promptkit.ErrOperationTimeout) || !errors.Is(err, context.DeadlineExceeded) || terminal.Failure != promptkit.ExecutionFailureDeadline {
t.Fatalf("error=%v terminal=%+v", err, terminal)
}
})
})
}
}

View File

@@ -0,0 +1,38 @@
package promptkit
import (
"context"
"testing"
"time"
"go.maximumdirect.net/promptkit/internal/domain"
)
func TestExecutionControlsConversion(t *testing.T) {
ctx := context.WithValue(context.Background(), struct{}{}, "selected")
called := false
req := RunRequest{OperationTimeout: 3 * time.Second, OnPrepared: func(got context.Context, details PreparedRun) {
called = true
if got != ctx || details.PromptID != "prompt" {
t.Fatalf("unexpected notification: %v, %+v", got, details)
}
details.InputHashes["input"] = "changed"
}}
converted, err := toDomainRunRequest(req)
if err != nil {
t.Fatal(err)
}
if converted.OperationTimeout != req.OperationTimeout || converted.OnPrepared == nil || called {
t.Fatal("execution controls were not retained without invocation")
}
req.OnPrepared = func(context.Context, PreparedRun) { t.Fatal("callback selection changed") }
prepared := &domain.PreparedRun{PromptID: "prompt", InputHashes: map[string]string{"input": "original"}}
converted.OnPrepared(ctx, prepared)
if !called || prepared.InputHashes["input"] != "original" {
t.Fatal("notification did not deliver detached details")
}
withoutCallback, err := toDomainRunRequest(RunRequest{})
if err != nil || withoutCallback.OnPrepared != nil {
t.Fatalf("nil callback was adapted: %v", err)
}
}

56
execution_group.go Normal file
View File

@@ -0,0 +1,56 @@
package promptkit
import (
"go.maximumdirect.net/promptkit/internal/domain"
"go.maximumdirect.net/promptkit/internal/observation"
)
// ExecutionGroup collects facts from explicitly selected execution events.
// Its zero value is ready for concurrent use. Do not copy it after first use.
type ExecutionGroup struct{ group observation.Group }
// ExecutionGroupMember identifies an observed run and its terminal coverage.
type ExecutionGroupMember struct {
RunID string `json:"run_id"`
TerminalObserved bool `json:"terminal_observed"`
}
// ExecutionGroupSummary is a detached snapshot of terminal-only totals.
// Usage sums known terminal values, with Known as their logical OR and
// Complete as their logical AND. Coverage describes whether every observed
// member has a terminal, independently of usage completeness. An empty group
// has complete coverage and known complete zero usage; a group with members
// but no terminals has unknown, incomplete usage. Overlapping groups are not
// generally additive.
type ExecutionGroupSummary struct {
Members []ExecutionGroupMember `json:"members"`
TerminalCoverageComplete bool `json:"terminal_coverage_complete"`
Attempts uint64 `json:"attempts"`
TransientRetries uint64 `json:"transient_retries"`
RepairCalls uint64 `json:"repair_calls"`
Usage ObservedTokenUsage `json:"usage"`
}
// Observe enrolls an event's run, counting its first terminal event only.
func (g *ExecutionGroup) Observe(event ExecutionEvent) {
count := func(c ObservedTokenCount) domain.ObservedTokenCount {
return domain.ObservedTokenCount{Value: c.Value, Known: c.Known, Complete: c.Complete}
}
g.group.Observe(domain.ExecutionEvent{Kind: domain.ExecutionEventKind(event.Kind), RunID: event.RunID, Attempts: event.Attempts, TransientRetries: event.TransientRetries, RepairCalls: event.RepairCalls,
Usage: domain.ObservedTokenUsage{PromptTokens: count(event.Usage.PromptTokens), CompletionTokens: count(event.Usage.CompletionTokens), TotalTokens: count(event.Usage.TotalTokens), CachedTokens: count(event.Usage.CachedTokens), CacheWriteTokens: count(event.Usage.CacheWriteTokens)}})
}
// Snapshot returns an internally consistent, caller-owned summary. Coverage
// describes observed members only, not future members or application progress.
func (g *ExecutionGroup) Snapshot() ExecutionGroupSummary {
value := g.group.Snapshot()
count := func(c domain.ObservedTokenCount) ObservedTokenCount {
return ObservedTokenCount{Value: c.Value, Known: c.Known, Complete: c.Complete}
}
members := make([]ExecutionGroupMember, len(value.Members))
for i, member := range value.Members {
members[i] = ExecutionGroupMember{RunID: member.RunID, TerminalObserved: member.TerminalObserved}
}
return ExecutionGroupSummary{Members: members, TerminalCoverageComplete: value.TerminalCoverageComplete, Attempts: value.Attempts, TransientRetries: value.TransientRetries, RepairCalls: value.RepairCalls,
Usage: ObservedTokenUsage{PromptTokens: count(value.Usage.PromptTokens), CompletionTokens: count(value.Usage.CompletionTokens), TotalTokens: count(value.Usage.TotalTokens), CachedTokens: count(value.Usage.CachedTokens), CacheWriteTokens: count(value.Usage.CacheWriteTokens)}}
}

View File

@@ -0,0 +1,151 @@
package promptkit_test
import (
"context"
"net/http"
"testing"
"go.maximumdirect.net/promptkit"
)
func TestExecutionGroupExplicitMembershipAndTerminalTotals(t *testing.T) {
responses := []*http.Response{
observedHTTPResponse(200, `{"choices":[{"message":{"content":"initial"}}],"usage":{"prompt_tokens":2}}`),
observedHTTPResponse(503, `{}`),
observedHTTPResponse(200, `{"choices":[{"message":{"content":"invalid"}}],"usage":{"prompt_tokens":3}}`),
observedHTTPResponse(200, `{"choices":[{"message":{"content":"{\"ok\":true}"}}],"usage":{"prompt_tokens":4}}`),
observedHTTPResponse(200, `{"choices":[{"message":{"content":"invalid"}}],"usage":{"prompt_tokens":5}}`),
observedHTTPResponse(422, `{"error":{}}`),
observedHTTPResponse(200, `{"choices":[{"message":{"content":"validator"}}],"usage":{"prompt_tokens":6}}`),
}
calls := 0
one := 1
config := simpleConfig("message")
config.TransientRetries = &one
config.HTTPClient = &http.Client{Transport: roundTripFunc(func(*http.Request) (*http.Response, error) {
response := responses[calls]
calls++
return response, nil
})}
engine, err := promptkit.NewEngine(config)
if err != nil {
t.Fatal(err)
}
ctx := context.Background()
var all, selected promptkit.ExecutionGroup
initial, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true, Observer: all.Observe})
if err != nil {
t.Fatal(err)
}
validation := &promptkit.OutputContractInput{Format: promptkit.FormatJSON, ValidationMode: promptkit.ValidationJSON, RepairAttempts: new(1)}
callbackCount := 0
var branchID, validatorID string
result, err := engine.Coordinate(ctx, promptkit.CoordinateRequest{Start: initial.Exchange, AdditionalTurns: 1, Observer: promptkit.ComposeObservers(all.Observe, selected.Observe), Next: func(ctx context.Context, turn promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
callbackCount++
if turn.Result == nil {
return &promptkit.RunRequest{PromptID: "prompt", Validation: validation}, nil
}
branchID = turn.Result.RunID
return nil, nil
}})
if err != nil || result.Outcome != promptkit.CoordinationStopped || callbackCount != 2 {
t.Fatalf("coordinate=%#v, %v", result, err)
}
if failed, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", ContinueFrom: initial.Exchange, Validation: validation, Observer: all.Observe}); err == nil || failed != nil {
t.Fatalf("failed branch=%#v, %v", failed, err)
}
validator, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", Observer: selected.Observe})
if err != nil {
t.Fatal(err)
}
validatorID = validator.RunID
a, b := all.Snapshot(), selected.Snapshot()
if calls != 7 || len(a.Members) != 3 || a.Attempts != 6 || a.TransientRetries != 1 || a.RepairCalls != 2 || a.Usage.PromptTokens.Value != 14 || !a.Usage.PromptTokens.Known || !a.TerminalCoverageComplete {
t.Fatalf("all summary=%#v calls=%d", a, calls)
}
if len(b.Members) != 2 || b.Attempts != 4 || b.Usage.PromptTokens.Value != 13 || !b.TerminalCoverageComplete {
t.Fatalf("selected summary=%#v", b)
}
ids := map[string]bool{}
for _, member := range b.Members {
ids[member.RunID] = true
}
if !ids[branchID] || !ids[validatorID] || ids[initial.RunID] {
t.Fatalf("selected members=%#v", b.Members)
}
}
func TestExecutionGroupPreparedSilenceAndPanicCoverage(t *testing.T) {
engine := newSimpleEngine(t, "message", promptkit.WithLLMClient(&fakeLLMClient{response: &promptkit.GenerateResponse{Content: "ok"}}))
ctx := context.Background()
var group promptkit.ExecutionGroup
prepared, err := engine.PrepareExecution(ctx, promptkit.RunRequest{PromptID: "prompt", Observer: group.Observe})
if err != nil {
t.Fatal(err)
}
prepared.Discard()
_, _ = engine.RunPrepared(ctx, prepared)
if len(group.Snapshot().Members) != 0 {
t.Fatal("preparation or failed claim enrolled a run")
}
func() {
defer func() {
if recover() == nil {
t.Error("expected panic")
}
}()
_, _ = engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", Observer: promptkit.ComposeObservers(group.Observe, func(event promptkit.ExecutionEvent) {
_ = group.Snapshot()
if event.Kind == promptkit.ExecutionStarted {
panic("stop")
}
})})
}()
result := group.Snapshot()
if len(result.Members) != 1 || result.TerminalCoverageComplete || result.Members[0].TerminalObserved || result.Attempts != 0 {
t.Fatalf("panic summary=%#v", result)
}
}
func TestExecutionGroupPreparedAndCallbackMembership(t *testing.T) {
engine := newSimpleEngine(t, "message", promptkit.WithLLMClient(&fakeLLMClient{response: &promptkit.GenerateResponse{Content: "ok"}}))
ctx := context.Background()
var continuations, nested promptkit.ExecutionGroup
initial, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", CaptureExchange: true})
if err != nil {
t.Fatal(err)
}
prepared, err := engine.PrepareExecution(ctx, promptkit.RunRequest{PromptID: "prompt", Observer: nested.Observe})
if err != nil {
t.Fatal(err)
}
preparedResult, err := engine.RunPrepared(ctx, prepared)
if err != nil {
t.Fatal(err)
}
var callbackID string
_, err = engine.Coordinate(ctx, promptkit.CoordinateRequest{Start: initial.Exchange, AdditionalTurns: 1, Observer: continuations.Observe, Next: func(ctx context.Context, turn promptkit.ConversationTurn) (*promptkit.RunRequest, error) {
if turn.Result == nil {
return &promptkit.RunRequest{PromptID: "prompt"}, nil
}
callbackResult, err := engine.Run(ctx, promptkit.RunRequest{PromptID: "prompt", Observer: nested.Observe})
if err == nil {
callbackID = callbackResult.RunID
}
return nil, err
}})
if err != nil {
t.Fatal(err)
}
a, b := continuations.Snapshot(), nested.Snapshot()
if len(a.Members) != 1 || len(b.Members) != 2 || !a.TerminalCoverageComplete || !b.TerminalCoverageComplete {
t.Fatalf("summaries = %#v %#v", a, b)
}
ids := map[string]bool{}
for _, member := range b.Members {
ids[member.RunID] = true
}
if !ids[preparedResult.RunID] || !ids[callbackID] || ids[initial.RunID] {
t.Fatalf("members=%#v", b.Members)
}
}

View File

@@ -0,0 +1,167 @@
package promptkit_test
import (
"context"
"errors"
"net/http"
"net/http/httptest"
"net/url"
"testing"
"testing/synctest"
"time"
"go.maximumdirect.net/promptkit"
)
func TestBuiltInGenerationOperationTimeout(t *testing.T) {
for _, split := range []bool{false, true} {
t.Run(map[bool]string{false: "run", true: "prepared"}[split], func(t *testing.T) {
started := make(chan struct{}, 1)
release := make(chan struct{})
provider := httptest.NewServer(http.HandlerFunc(func(http.ResponseWriter, *http.Request) {
select {
case started <- struct{}{}:
default:
}
<-release
}))
defer provider.Close()
defer close(release)
config := simpleConfig("message")
zero := 0
config.TransientRetries = &zero
config.HTTPClient = provider.Client()
engine, err := promptkit.NewEngine(config)
if err != nil {
t.Fatal(err)
}
var terminal promptkit.ExecutionEvent
req := promptkit.RunRequest{PromptID: "prompt", OperationTimeout: time.Second,
Execution: &promptkit.ExecutionTargetOverride{Endpoint: provider.URL + "/v1", TimeoutSeconds: &zero},
Observer: func(event promptkit.ExecutionEvent) {
if event.Kind == promptkit.ExecutionFinished {
terminal = event
}
},
}
parent := context.Background()
var result *promptkit.RunResult
if split {
handle, prepareErr := engine.PrepareExecution(parent, req)
if prepareErr != nil {
t.Fatal(prepareErr)
}
result, err = engine.RunPrepared(parent, handle)
} else {
result, err = engine.Run(parent, req)
}
select {
case <-started:
default:
t.Fatal("operation expired before provider received the request")
}
for _, target := range []error{promptkit.ErrOperationTimeout, context.DeadlineExceeded, promptkit.ErrLLMGenerate, promptkit.ErrTransport} {
if !errors.Is(err, target) {
t.Errorf("error=%v, want identity %v", err, target)
}
}
var transportErr *url.Error
if !errors.As(err, &transportErr) {
t.Errorf("error=%v, want retained HTTP transport cause", err)
}
if result != nil || parent.Err() != nil || terminal.Failure != promptkit.ExecutionFailureDeadline {
t.Fatalf("result=%v parent=%v terminal=%+v", result, parent.Err(), terminal)
}
})
}
}
func TestRunOperationTimeoutAttribution(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
engine := observationEngine(t, observationClientFunc(func(ctx context.Context, _ promptkit.GenerateRequest) (*promptkit.GenerateResponse, error) {
<-ctx.Done()
return nil, ctx.Err()
}))
var terminal promptkit.ExecutionEvent
_, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt", OperationTimeout: time.Second,
Observer: func(event promptkit.ExecutionEvent) {
if event.Kind == promptkit.ExecutionFinished {
terminal = event
}
}})
if !errors.Is(err, promptkit.ErrOperationTimeout) || !errors.Is(err, context.DeadlineExceeded) || !errors.Is(err, promptkit.ErrLLMGenerate) || terminal.Failure != promptkit.ExecutionFailureDeadline {
t.Fatalf("error=%v terminal=%+v", err, terminal)
}
})
}
func TestRunProviderDeadlineIsNotOperationTimeout(t *testing.T) {
engine := observationEngine(t, observationClientFunc(func(context.Context, promptkit.GenerateRequest) (*promptkit.GenerateResponse, error) {
return nil, context.DeadlineExceeded
}))
_, err := engine.Run(context.Background(), promptkit.RunRequest{PromptID: "prompt", OperationTimeout: time.Minute})
if !errors.Is(err, context.DeadlineExceeded) || errors.Is(err, promptkit.ErrOperationTimeout) {
t.Fatalf("provider deadline attribution: %v", err)
}
}
func TestPreparedExecutionRetainsTimeoutWithoutIdleTimer(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
engine := observationEngine(t, observationClientFunc(func(ctx context.Context, _ promptkit.GenerateRequest) (*promptkit.GenerateResponse, error) {
<-ctx.Done()
return nil, ctx.Err()
}))
req := promptkit.RunRequest{PromptID: "prompt", OperationTimeout: time.Second}
handle, err := engine.PrepareExecution(context.Background(), req)
if err != nil {
t.Fatal(err)
}
req.OperationTimeout = 0
time.Sleep(2 * time.Second)
start := time.Now()
_, err = engine.RunPrepared(context.Background(), handle)
if time.Since(start) != time.Second || !errors.Is(err, promptkit.ErrOperationTimeout) || !errors.Is(err, context.DeadlineExceeded) || !errors.Is(err, promptkit.ErrLLMGenerate) {
t.Fatalf("elapsed=%v error=%v", time.Since(start), err)
}
})
}
func TestPreparedExecutionRespectsCallerDeadline(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
engine := observationEngine(t, observationClientFunc(func(ctx context.Context, _ promptkit.GenerateRequest) (*promptkit.GenerateResponse, error) {
<-ctx.Done()
return nil, ctx.Err()
}))
ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
defer cancel()
handle, err := engine.PrepareExecution(ctx, promptkit.RunRequest{PromptID: "prompt", OperationTimeout: 5 * time.Second})
if err != nil {
t.Fatal(err)
}
time.Sleep(time.Second)
_, err = engine.RunPrepared(ctx, handle)
if !errors.Is(err, context.DeadlineExceeded) || errors.Is(err, promptkit.ErrOperationTimeout) {
t.Fatalf("caller deadline attribution: %v", err)
}
})
}
func TestPreparedExecutionDoesNotTimePreparation(t *testing.T) {
synctest.Test(t, func(t *testing.T) {
release := make(chan struct{})
reader := &capacityArtifactReader{entered: make(chan struct{}, 1), release: release}
engine := newBackendCapacityEngine(t, newCapacityGateClient(closedCapacityChannel(), 1), 1, capacityInt(0), reader)
req := capacityInputRequest("http://example.invalid/v1")
req.OperationTimeout = time.Second
go func() { time.Sleep(2 * time.Second); close(release) }()
start := time.Now()
handle, err := engine.PrepareExecution(context.Background(), req)
if err != nil || time.Since(start) != 2*time.Second {
t.Fatalf("prepare elapsed=%v error=%v", time.Since(start), err)
}
if _, err = engine.RunPrepared(context.Background(), handle); err != nil {
t.Fatalf("execution after preparation: %v", err)
}
})
}

View File

@@ -0,0 +1,204 @@
package promptkit_test
import (
"bytes"
"errors"
"strings"
"sync/atomic"
"testing"
"go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit"
)
type focusedInspectionRoot struct {
Sibling string `yaml:"sibling"`
PromptKit *promptkit.Settings `yaml:"promptkit"`
}
func TestFocusedInspectionReturnsDetachedDiagnostics(t *testing.T) {
var resolves, validates atomic.Int64
definition := integrationDefinition(&resolves, &validates)
result, err := configkit.Load(configkit.Bytes("diagnostics", []byte(integrationDocument)), configkit.Options[integrationApplication]{Definition: definition})
if err != nil {
t.Fatal(err)
}
engine, err := promptkit.NewFromConfig(result, promptkit.RuntimeConfig{})
if err != nil {
t.Fatal(err)
}
view, err := engine.InspectConfiguration()
if err != nil {
t.Fatal(err)
}
first := view.Diagnostics()
if len(first) != 1 || first[0].Primary == nil {
t.Fatalf("diagnostics = %+v", first)
}
original := first[0].Primary.SourceID
first[0].Primary.SourceID = "changed"
again := view.Diagnostics()
if again[0].Primary.SourceID != original {
t.Fatalf("diagnostics alias caller mutation: %+v", again)
}
}
func focusedInspection(t *testing.T, source string, sensitive func(configkit.Path) bool) (promptkit.ConfigurationInspection, *configkit.Result[focusedInspectionRoot]) {
t.Helper()
definition := promptkit.Configure(configkit.Definition[focusedInspectionRoot]{Sensitive: sensitive}, func(value *focusedInspectionRoot) *promptkit.Settings { return value.PromptKit })
result, err := configkit.Load(configkit.Bytes("focused", []byte(source)), configkit.Options[focusedInspectionRoot]{Definition: definition})
if err != nil {
t.Fatal(err)
}
engine, err := promptkit.NewFromConfig(result, promptkit.RuntimeConfig{})
if err != nil {
t.Fatal(err)
}
view, err := engine.InspectConfiguration()
if err != nil {
t.Fatal(err)
}
return view, result
}
func TestFocusedInspectionRetainsNativeEvidence(t *testing.T) {
view, result := focusedInspection(t, "sibling: private\npromptkit:\n app_secret: hidden\n app_value: original\n", func(path configkit.Path) bool {
return path.String() == "/sibling" || path.String() == "/promptkit/app_secret"
})
result.Config.PromptKit.Application["app_value"] = "mutated"
var rendered bytes.Buffer
if err := view.WriteEffective(&rendered); err != nil || !strings.Contains(rendered.String(), "original") || strings.Contains(rendered.String(), "hidden") || strings.Contains(rendered.String(), "mutated") || strings.Contains(rendered.String(), "sibling") {
t.Fatalf("focused effective = %q, %v", rendered.String(), err)
}
if _, ok := view.Metadata().Lookup(configkit.Keys("sibling")); !ok {
t.Fatal("unfiltered metadata omitted sibling")
}
sibling, ok := view.Explain(configkit.Keys("sibling"))
if !ok || !sibling.Effective.Redacted {
t.Fatalf("outside-focus explanation = %+v, %t", sibling, ok)
}
if _, ok := view.Explain(configkit.Keys("promptkit")); !ok {
t.Fatal("focus explanation missing")
}
if _, ok := view.Explain(configkit.Path{}); !ok {
t.Fatal("whole-application explanation missing")
}
if len(view.Sources()) == 0 {
t.Fatal("native source catalog missing")
}
for _, path := range view.HistoricalPaths() {
if path.String() != "/promptkit" && !strings.HasPrefix(path.String(), "/promptkit/") {
t.Fatalf("out of scope history: %v", path)
}
}
if len(view.HistoricalPaths()) == 0 {
t.Fatal("focused history empty")
}
}
func TestFocusedInspectionHonorsAncestorSensitivityAndHistory(t *testing.T) {
redacted, _ := focusedInspection(t, "promptkit: {app_value: secret-value}\n", func(path configkit.Path) bool { return path.String() == "/promptkit" })
var rendered bytes.Buffer
if err := redacted.WriteEffective(&rendered); err != nil || strings.Contains(rendered.String(), "secret-value") || !strings.Contains(rendered.String(), "REDACTED") {
t.Fatalf("ancestor redaction = %q, %v", rendered.String(), err)
}
definition := promptkit.Configure(configkit.Definition[focusedInspectionRoot]{}, func(value *focusedInspectionRoot) *promptkit.Settings { return value.PromptKit })
lower := configkit.Input{Source: configkit.Bytes("lower", []byte("promptkit: {historical: old}\n"))}
main := configkit.Input{Source: configkit.Bytes("main", []byte("promptkit: null\n"))}
result, err := configkit.LoadLayered(main, configkit.LayeredOptions[focusedInspectionRoot]{Definition: definition, Underlays: []configkit.Underlay{{Inputs: []configkit.Input{lower}}}})
if err != nil {
t.Fatal(err)
}
engine, err := promptkit.NewFromConfig(result, promptkit.RuntimeConfig{})
if err != nil {
t.Fatal(err)
}
view, err := engine.InspectConfiguration()
if err != nil {
t.Fatal(err)
}
path := configkit.Keys("promptkit", "historical")
if len(view.Metadata().History(path)) == 0 {
t.Fatal("historical-only field has no native history")
}
found := false
for _, item := range view.HistoricalPaths() {
if item.String() == path.String() {
found = true
}
}
if !found {
t.Fatalf("historical path missing: %+v", view.HistoricalPaths())
}
}
func TestFocusedInspectionAbsenceNullAndCrossTypeDiff(t *testing.T) {
type noSection struct {
Sibling string `yaml:"sibling"`
}
missing, err := configkit.Load[noSection](configkit.Bytes("absent", []byte("sibling: before\n")))
if err != nil {
t.Fatal(err)
}
missingEngine, err := promptkit.NewFromConfig(missing, promptkit.RuntimeConfig{})
if err != nil {
t.Fatal(err)
}
absent, err := missingEngine.InspectConfiguration()
if err != nil {
t.Fatal(err)
}
var output bytes.Buffer
if err := absent.WriteEffective(&output); err == nil || output.Len() != 0 {
t.Fatalf("absent output = %q, %v", output.String(), err)
}
nullView, _ := focusedInspection(t, "sibling: before\npromptkit: null\n", nil)
if err := nullView.WriteEffective(&output); err != nil || strings.TrimSpace(output.String()) != "null" {
t.Fatalf("null output = %q, %v", output.String(), err)
}
differences, err := absent.Diff(nullView)
if err != nil || len(differences) == 0 {
t.Fatalf("absent/null diff = %+v, %v", differences, err)
}
changed, _ := focusedInspection(t, "sibling: before\npromptkit: {label: one}\n", nil)
if differences, err := absent.Diff(changed); err != nil || len(differences) == 0 {
t.Fatalf("added scope diff = %+v, %v", differences, err)
}
if differences, err := changed.Diff(absent); err != nil || len(differences) == 0 {
t.Fatalf("removed scope diff = %+v, %v", differences, err)
}
otherSibling, _ := focusedInspection(t, "sibling: after\npromptkit: {label: one}\n", nil)
differences, err = changed.Diff(otherSibling)
if err != nil || len(differences) != 0 {
t.Fatalf("sibling-only diff = %+v, %v", differences, err)
}
type differentRoot struct {
PromptKit promptkit.Settings `yaml:"promptkit"`
Extra bool `yaml:"extra"`
}
definition := promptkit.Configure(configkit.Definition[differentRoot]{}, func(value *differentRoot) *promptkit.Settings { return &value.PromptKit })
result, err := configkit.Load(configkit.Bytes("other", []byte("promptkit: {label: two}\n")), configkit.Options[differentRoot]{Definition: definition})
if err != nil {
t.Fatal(err)
}
engine, err := promptkit.NewFromConfig(result, promptkit.RuntimeConfig{})
if err != nil {
t.Fatal(err)
}
other, err := engine.InspectConfiguration()
if err != nil {
t.Fatal(err)
}
differences, err = changed.Diff(other)
if err != nil || len(differences) == 0 {
t.Fatalf("cross-root diff = %+v, %v", differences, err)
}
var zero promptkit.ConfigurationInspection
if _, err := zero.Diff(changed); !errors.Is(err, promptkit.ErrInvalidConfig) || zero.HistoricalPaths() != nil {
t.Fatalf("zero view: %v", err)
}
writeErr := errors.New("writer failure")
if err := changed.WriteEffective(failingInspectionWriter{err: writeErr}); !errors.Is(err, writeErr) {
t.Fatalf("writer error = %v", err)
}
}

View File

@@ -38,8 +38,8 @@ func capturedPromptEngine(t *testing.T, source fs.FS) *promptkit.Engine {
endpoint, model := "https://example.invalid/v1", "offline-model" endpoint, model := "https://example.invalid/v1", "offline-model"
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
Underlays: []configkit.Underlay{ Underlays: []configkit.Underlay{
{Inputs: inputs}, {Inputs: mountSettingsInputs(inputs)},
{Inputs: []configkit.Input{{Source: promptkit.ProfileSource("offline profile", []promptkit.ProfileConfig{ {Inputs: []configkit.Input{{Destination: configkit.Keys("promptkit"), Source: promptkit.ProfileSource("offline profile", []promptkit.ProfileConfig{
{ID: "offline", Endpoint: &endpoint, Model: &model}, {ID: "offline", Endpoint: &endpoint, Model: &model},
})}}}, })}}},
}, },
@@ -144,10 +144,10 @@ func TestPreparationStillReadsRequestArtifactsAndSchemasAfterConstruction(t *tes
endpoint, model := "https://example.invalid/v1", "offline-model" endpoint, model := "https://example.invalid/v1", "offline-model"
schemas := fstest.MapFS{} schemas := fstest.MapFS{}
engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{ engine, err := promptkit.NewEngine(promptkit.Config{Sources: promptkit.ConfigurationSources{
Underlays: []configkit.Underlay{{Inputs: []configkit.Input{{Source: promptkit.ProfileSource("profile", []promptkit.ProfileConfig{ Underlays: []configkit.Underlay{{Inputs: []configkit.Input{schemaRootInput(schemas, ".")}}, {Inputs: []configkit.Input{{Destination: configkit.Keys("promptkit"), Source: promptkit.ProfileSource("profile", []promptkit.ProfileConfig{
{ID: "offline", Endpoint: &endpoint, Model: &model}, {ID: "offline", Endpoint: &endpoint, Model: &model},
})}}}}, })}}}},
Main: &configkit.Input{Source: configkit.Mapping("structured prompt", map[string]any{ Main: &configkit.Input{Destination: configkit.Keys("promptkit"), Source: configkit.Mapping("structured prompt", map[string]any{
"prompts": []any{map[string]any{"id": "structured", "prompts": []any{map[string]any{"id": "structured",
"default_profile": "offline", "default_profile": "offline",
"inputs": []any{map[string]any{"name": "note", "required": true}}, "inputs": []any{map[string]any{"name": "note", "required": true}},
@@ -155,7 +155,7 @@ func TestPreparationStillReadsRequestArtifactsAndSchemasAfterConstruction(t *tes
"output": map[string]any{"format": "json", "validation_mode": "json_schema", "schema_path": "schema.json"}, "output": map[string]any{"format": "json", "validation_mode": "json_schema", "schema_path": "schema.json"},
}}, }},
})}, })},
}}, promptkit.WithSchemaFS(schemas, ".")) }})
if err != nil { if err != nil {
t.Fatalf("construct engine before request resources exist: %v", err) t.Fatalf("construct engine before request resources exist: %v", err)
} }

View File

@@ -78,10 +78,11 @@ func TestBuiltInRepairGenerationError(t *testing.T) {
t.Fatalf("NewEngine: %v", err) t.Fatalf("NewEngine: %v", err)
} }
req := simpleGenerationRequest() req := simpleGenerationRequest()
req.Validation = &promptkit.OutputContract{ req.CaptureExchange = true
req.Validation = &promptkit.OutputContractInput{
Format: promptkit.FormatJSON, Format: promptkit.FormatJSON,
ValidationMode: promptkit.ValidationJSON, ValidationMode: promptkit.ValidationJSON,
RepairAttempts: 1, RepairAttempts: new(1),
} }
result, err := engine.Run(context.Background(), req) result, err := engine.Run(context.Background(), req)

14
go.mod
View File

@@ -3,16 +3,16 @@ module go.maximumdirect.net/promptkit
go 1.27.0 go 1.27.0
require ( require (
github.com/santhosh-tekuri/jsonschema/v6 v6.0.2 github.com/santhosh-tekuri/jsonschema/v6 v6.0.3
go.maximumdirect.net/configkit v1.0.0 go.maximumdirect.net/configkit v1.4.0
go.yaml.in/yaml/v4 v4.0.0
) )
require ( require (
github.com/yuin/goldmark/v2 v2.1.6 // indirect github.com/yuin/goldmark/v2 v2.1.16 // indirect
go.maximumdirect.net/releasekit v0.8.0 // indirect go.maximumdirect.net/releasekit v0.10.0 // indirect
go.yaml.in/yaml/v4 v4.0.0-rc.6 // indirect golang.org/x/mod v0.43.0 // indirect
golang.org/x/mod v0.40.0 // indirect golang.org/x/text v0.44.0 // indirect
golang.org/x/text v0.14.0 // indirect
) )
tool go.maximumdirect.net/releasekit/cmd/releasekit tool go.maximumdirect.net/releasekit/cmd/releasekit

30
go.sum
View File

@@ -1,16 +1,18 @@
github.com/dlclark/regexp2 v1.11.0 h1:G/nrcoOa7ZXlpoa/91N3X7mM3r8eIlMBBJZvsz/mxKI= github.com/dlclark/regexp2 v1.11.0 h1:G/nrcoOa7ZXlpoa/91N3X7mM3r8eIlMBBJZvsz/mxKI=
github.com/dlclark/regexp2 v1.11.0/go.mod h1:DHkYz0B9wPfa6wondMfaivmHpzrQ3v9q8cnmRbL6yW8= github.com/dlclark/regexp2 v1.11.0/go.mod h1:DHkYz0B9wPfa6wondMfaivmHpzrQ3v9q8cnmRbL6yW8=
github.com/santhosh-tekuri/jsonschema/v6 v6.0.2 h1:KRzFb2m7YtdldCEkzs6KqmJw4nqEVZGK7IN2kJkjTuQ= github.com/santhosh-tekuri/jsonschema/v6 v6.0.3 h1:1EYB5IzjZawrrnELUi78f9fPu57HuXjmddZPjrls/28=
github.com/santhosh-tekuri/jsonschema/v6 v6.0.2/go.mod h1:JXeL+ps8p7/KNMjDQk3TCwPpBy0wYklyWTfbkIzdIFU= github.com/santhosh-tekuri/jsonschema/v6 v6.0.3/go.mod h1:JXeL+ps8p7/KNMjDQk3TCwPpBy0wYklyWTfbkIzdIFU=
github.com/yuin/goldmark/v2 v2.1.6 h1:igy0q9JcqovoFXFx/XjtU8sHVhcUHavPj2sA3EoKJUE= github.com/yuin/goldmark/v2 v2.1.16 h1:MXvCVo+Cb2SVJ5cwJUzXetDjI7HECRJCkcqylEQtcJo=
github.com/yuin/goldmark/v2 v2.1.6/go.mod h1:G6M4/qOFtfn01/o14BU1UR2Lo5N3S9Qo7xCuz/sHjGQ= github.com/yuin/goldmark/v2 v2.1.16/go.mod h1:G6M4/qOFtfn01/o14BU1UR2Lo5N3S9Qo7xCuz/sHjGQ=
go.maximumdirect.net/configkit v1.0.0 h1:cguQnxkaRk3ClQOZNzzhTdc4bfzbraGKv8CkoCp+HLc= go.maximumdirect.net/configkit v1.3.0 h1:X3LZnNJL6T2C6DyUugcvIvHMXX6qhtAF+fQEUBeAwuw=
go.maximumdirect.net/configkit v1.0.0/go.mod h1:B6fJ5xBLBF1nuywqfLeEI4sLm29UkIcDLu8i6I6gGXE= go.maximumdirect.net/configkit v1.3.0/go.mod h1:ewu6blSokUDCMovy8JBXAdXRikGmlHXAQpFghnPWRfs=
go.maximumdirect.net/releasekit v0.8.0 h1:hSaHPbUI7tAhnolfbu7UMj2fvmm9FGBY+jhkygdbxR0= go.maximumdirect.net/configkit v1.4.0 h1:TfOupZbQ23/ZMs+xt5lmyukCF11ekqKep8KbmFQD63o=
go.maximumdirect.net/releasekit v0.8.0/go.mod h1:QE4BLrAhVJTzi+/YTcVl21K/qZp9ojRLg5kuoy3Zb44= go.maximumdirect.net/configkit v1.4.0/go.mod h1:ewu6blSokUDCMovy8JBXAdXRikGmlHXAQpFghnPWRfs=
go.yaml.in/yaml/v4 v4.0.0-rc.6 h1:1h7H1ohdUh93/FyE4YaDa1Zh64K6VVbjF4K6WUxMtH4= go.maximumdirect.net/releasekit v0.10.0 h1:3GOgaIigXTkvYaRkntWZASE//2NTR1+IhkoKmt64aJo=
go.yaml.in/yaml/v4 v4.0.0-rc.6/go.mod h1:aZqd9kCMsGL7AuUv/m/PvWLdg5sjJsZ4oHDEnfPPfY0= go.maximumdirect.net/releasekit v0.10.0/go.mod h1:cT5NZNEcf7SKDx4MPG40YNK9V4bNyIxcLLFuNlbGQ90=
golang.org/x/mod v0.40.0 h1:hUv+3cXcdRHz08UmSiOob7sadHig73uo5bkXxQ/tvUs= go.yaml.in/yaml/v4 v4.0.0 h1:/LG1pNO84Z4l603j/xMn69CcS5OYhMQ/f7l2LoQUGuw=
golang.org/x/mod v0.40.0/go.mod h1:0/weTWkPWGBikyTWAX3dkjVztMmBA5hM0DH6BElSupE= go.yaml.in/yaml/v4 v4.0.0/go.mod h1:aZqd9kCMsGL7AuUv/m/PvWLdg5sjJsZ4oHDEnfPPfY0=
golang.org/x/text v0.14.0 h1:ScX5w1eTa3QqT8oi6+ziP7dTV1S2+ALU0bI+0zXKWiQ= golang.org/x/mod v0.43.0 h1:MFuq5J6s35W9D3IjkwyWc9vHh0nENOrt5K8Iof4ofmg=
golang.org/x/text v0.14.0/go.mod h1:18ZOQIKpY8NJVqYksKHtTdi31H5itFRjB5/qKTNYzSU= golang.org/x/mod v0.43.0/go.mod h1:4ec1JnuyBz5+FIVhg86UsA39YG8BpnrZdJYSNHmGzHY=
golang.org/x/text v0.44.0 h1:YItdp/mQNGnsmrgpcdA5yQ3bnRxCfo/Tj/iByGW9GGo=
golang.org/x/text v0.44.0/go.mod h1:S4FR1waQ9LbjQTls97SyRjDaUUVkaM+cXwwuG/G+MwY=

View File

@@ -3,8 +3,6 @@ package artifact
import ( import (
"bytes" "bytes"
"context" "context"
"crypto/sha256"
"fmt"
"mime" "mime"
"os" "os"
"path/filepath" "path/filepath"
@@ -153,9 +151,12 @@ func checkArtifactFileRead(b *testing.B, reader Reader, ref domain.ArtifactRef,
if !bytes.Equal(artifact.Body, want) || artifact.Size != int64(len(want)) { if !bytes.Equal(artifact.Body, want) || artifact.Size != int64(len(want)) {
b.Fatalf("artifact size=%d body length=%d, want %d", artifact.Size, len(artifact.Body), len(want)) b.Fatalf("artifact size=%d body length=%d, want %d", artifact.Size, len(artifact.Body), len(want))
} }
wantHash := fmt.Sprintf("%x", sha256.Sum256(want)) inline, err := NewCompositeReader().Read(context.Background(), domain.ArtifactRef{Type: domain.ArtifactRefInline, Body: string(want)})
if artifact.Hash != wantHash { if err != nil {
b.Fatalf("artifact hash=%q, want %q", artifact.Hash, wantHash) b.Fatal(err)
}
if inline.Hash == "" || artifact.Hash != inline.Hash {
b.Fatalf("file and inline hashes differ: %q/%q", artifact.Hash, inline.Hash)
} }
} }

View File

@@ -3,9 +3,7 @@ package artifact
import ( import (
"bytes" "bytes"
"context" "context"
"crypto/sha256"
"errors" "errors"
"fmt"
"os" "os"
"path/filepath" "path/filepath"
"strings" "strings"
@@ -284,9 +282,12 @@ func TestFileReaderUsesOpenedSizeOnlyAsACapacityHint(t *testing.T) {
if artifact.Name != filepath.Base(filePath) || artifact.URI != filePath || artifact.ContentType == "" { if artifact.Name != filepath.Base(filePath) || artifact.URI != filePath || artifact.ContentType == "" {
t.Fatalf("unexpected artifact metadata: name=%q URI=%q content type=%q", artifact.Name, artifact.URI, artifact.ContentType) t.Fatalf("unexpected artifact metadata: name=%q URI=%q content type=%q", artifact.Name, artifact.URI, artifact.ContentType)
} }
wantHash := fmt.Sprintf("%x", sha256.Sum256(tc.body)) inline, err := NewCompositeReader().Read(context.Background(), domain.ArtifactRef{Type: domain.ArtifactRefInline, Body: string(tc.body)})
if artifact.Hash != wantHash { if err != nil {
t.Fatalf("artifact hash=%q, want %q", artifact.Hash, wantHash) t.Fatal(err)
}
if inline.Hash == "" || artifact.Hash != inline.Hash {
t.Fatalf("file and inline hashes differ: %q/%q", artifact.Hash, inline.Hash)
} }
}) })
} }

View File

@@ -27,17 +27,13 @@ type Registry struct {
// NewEffectiveRegistry stores already-validated effective backend definitions // NewEffectiveRegistry stores already-validated effective backend definitions
// under their exact configuration keys. Configuration validation and // under their exact configuration keys. Configuration validation and
// normalization precede this ownership boundary. // normalization precede this ownership boundary.
func NewEffectiveRegistry(definitions map[string]domain.Backend) (*Registry, error) { func NewEffectiveRegistry(definitions map[string]domain.Backend) *Registry {
registry := &Registry{backends: make(map[string]domain.Backend, len(definitions))} registry := &Registry{backends: make(map[string]domain.Backend, len(definitions))}
for id, definition := range definitions { for id, definition := range definitions {
definition.ID = id definition.ID = id
cloned, err := domain.CloneBackend(definition) registry.backends[id] = domain.CloneBackend(definition)
if err != nil {
return nil, fmt.Errorf("backend %q: %w", id, err)
}
registry.backends[id] = cloned
} }
return registry, nil return registry
} }
// GetBackend returns a defensive copy of the backend registered with id. // GetBackend returns a defensive copy of the backend registered with id.
@@ -49,11 +45,7 @@ func (r *Registry) GetBackend(id string) (domain.Backend, error) {
if !ok { if !ok {
return domain.Backend{}, fmt.Errorf("%w: %q", ErrBackendNotFound, id) return domain.Backend{}, fmt.Errorf("%w: %q", ErrBackendNotFound, id)
} }
cloned, err := domain.CloneBackend(definition) return domain.CloneBackend(definition), nil
if err != nil {
return domain.Backend{}, fmt.Errorf("copy backend %q: %w", id, err)
}
return cloned, nil
} }
// CapacityPolicies returns a copy of the normalized policies for limited // CapacityPolicies returns a copy of the normalized policies for limited

View File

@@ -22,16 +22,13 @@ func TestEffectiveRegistryOwnsDefinitionsAndCapacityPolicies(t *testing.T) {
Kind: &authenticationKind, Endpoint: &authenticationEndpoint, Kind: &authenticationKind, Endpoint: &authenticationEndpoint,
}, },
} }
registry, err := backend.NewEffectiveRegistry(map[string]domain.Backend{ registry := backend.NewEffectiveRegistry(map[string]domain.Backend{
" local.v1 ": { " local.v1 ": {
Endpoint: "https://backend.example/v1", ExtraParams: params, Endpoint: "https://backend.example/v1", ExtraParams: params,
ConcurrencyLimit: 2, QueueCapacity: 3, QueueCapacitySet: true, ConcurrencyLimit: 2, QueueCapacity: 3,
SmokeTests: smokeTests, SmokeTests: smokeTests,
}, },
}) })
if err != nil {
t.Fatal(err)
}
params["nested"].(map[string]any)["enabled"] = false params["nested"].(map[string]any)["enabled"] = false
defaultProfile = "mutated-profile" defaultProfile = "mutated-profile"
reachabilityEndpoint = "https://mutated.example/ready" reachabilityEndpoint = "https://mutated.example/ready"
@@ -79,10 +76,7 @@ func TestEffectiveRegistryOwnsDefinitionsAndCapacityPolicies(t *testing.T) {
} }
func TestEffectiveRegistryMissingBackend(t *testing.T) { func TestEffectiveRegistryMissingBackend(t *testing.T) {
registry, err := backend.NewEffectiveRegistry(nil) registry := backend.NewEffectiveRegistry(nil)
if err != nil {
t.Fatal(err)
}
if _, err := registry.GetBackend("missing"); !errors.Is(err, backend.ErrBackendNotFound) { if _, err := registry.GetBackend("missing"); !errors.Is(err, backend.ErrBackendNotFound) {
t.Fatalf("missing backend error = %v", err) t.Fatalf("missing backend error = %v", err)
} }

View File

@@ -1,40 +0,0 @@
package capacity
import (
"context"
"go.maximumdirect.net/promptkit/internal/domain"
"go.maximumdirect.net/promptkit/internal/llm"
)
type client struct {
manager *Manager
next llm.Client
}
// NewClient wraps next with configured active-generation limits. A nil manager
// leaves next unchanged.
func NewClient(manager *Manager, next llm.Client) llm.Client {
if manager == nil {
return next
}
return &client{
manager: manager,
next: next,
}
}
func (c *client) Generate(
ctx context.Context,
req domain.GenerateRequest,
) (*domain.GenerateResponse, error) {
if c.manager.getPool(req.Target.BackendID) == nil {
return c.next.Generate(ctx, req)
}
release, err := c.manager.Acquire(ctx, req.Target.BackendID)
if err != nil {
return nil, err
}
defer release()
return c.next.Generate(ctx, req)
}

View File

@@ -3,7 +3,6 @@ package capacity
import ( import (
"context" "context"
"errors" "errors"
"reflect"
"runtime" "runtime"
"sync" "sync"
"sync/atomic" "sync/atomic"
@@ -24,6 +23,18 @@ type clientFunc func(
domain.GenerateRequest, domain.GenerateRequest,
) (*domain.GenerateResponse, error) ) (*domain.GenerateResponse, error)
// newTestClient exercises one acquisition around a controlled call.
func newTestClient(manager *Manager, next llm.Client) llm.Client {
return clientFunc(func(ctx context.Context, req domain.GenerateRequest) (*domain.GenerateResponse, error) {
release, err := manager.Acquire(ctx, req.Target.BackendID)
if err != nil {
return nil, err
}
defer release()
return next.Generate(ctx, req)
})
}
func (f clientFunc) Generate( func (f clientFunc) Generate(
ctx context.Context, ctx context.Context,
req domain.GenerateRequest, req domain.GenerateRequest,
@@ -167,7 +178,7 @@ func TestClientLimitsPeakConcurrencyAndServesWaitersFIFO(t *testing.T) {
"second": secondRelease, "second": secondRelease,
"third": thirdRelease, "third": thirdRelease,
}) })
client := NewClient(manager, next) client := newTestClient(manager, next)
first := generateAsync(client, context.Background(), "limited", "first") first := generateAsync(client, context.Background(), "limited", "first")
if got := receiveStarted(t, next.started); got != "first" { if got := receiveStarted(t, next.started); got != "first" {
@@ -212,7 +223,7 @@ func TestClientPeakConcurrencyDoesNotExceedConfiguredLimit(t *testing.T) {
releases[string(rune('a'+i))] = gate releases[string(rune('a'+i))] = gate
} }
next := newBlockingClient(releases) next := newBlockingClient(releases)
client := NewClient(manager, next) client := newTestClient(manager, next)
results := make([]<-chan generateResult, 0, len(releases)) results := make([]<-chan generateResult, 0, len(releases))
for id := range releases { for id := range releases {
@@ -257,7 +268,7 @@ func TestClientRemovesCanceledWaiters(t *testing.T) {
"three": make(chan struct{}), "three": make(chan struct{}),
} }
next := newBlockingClient(releases) next := newBlockingClient(releases)
client := NewClient(manager, next) client := newTestClient(manager, next)
holder := generateAsync(client, context.Background(), "limited", "holder") holder := generateAsync(client, context.Background(), "limited", "holder")
if got := receiveStarted(t, next.started); got != "holder" { if got := receiveStarted(t, next.started); got != "holder" {
@@ -321,7 +332,7 @@ func TestClientGrantCancellationRaceDoesNotLeakPermit(t *testing.T) {
} }
return &domain.GenerateResponse{Content: req.Prompt.SessionID}, nil return &domain.GenerateResponse{Content: req.Prompt.SessionID}, nil
}) })
client := NewClient(manager, next) client := newTestClient(manager, next)
holder := generateAsync(client, context.Background(), "limited", "holder") holder := generateAsync(client, context.Background(), "limited", "holder")
waitForActiveCount(t, manager, "limited", 1) waitForActiveCount(t, manager, "limited", 1)
@@ -394,7 +405,7 @@ func TestClientUsesIndependentPoolsAndUnlimitedFastPaths(t *testing.T) {
"alpha": alphaRelease, "alpha": alphaRelease,
"beta": betaRelease, "beta": betaRelease,
}) })
client := NewClient(manager, next) client := newTestClient(manager, next)
alpha := generateAsync(client, context.Background(), "alpha", "alpha") alpha := generateAsync(client, context.Background(), "alpha", "alpha")
beta := generateAsync(client, context.Background(), "beta", "beta") beta := generateAsync(client, context.Background(), "beta", "beta")
@@ -423,63 +434,6 @@ func TestClientUsesIndependentPoolsAndUnlimitedFastPaths(t *testing.T) {
t.Fatalf("unlimited backend %q response=(%#v, %v)", backendID, response, err) t.Fatalf("unlimited backend %q response=(%#v, %v)", backendID, response, err)
} }
} }
if got := NewClient(nil, next); got != next {
t.Fatal("nil manager did not return the wrapped client unchanged")
}
}
func TestClientPreservesRequestsResponsesAndErrors(t *testing.T) {
manager := newTestManager(t, map[string]domain.BackendCapacityPolicy{
"limited": {ConcurrencyLimit: 1},
})
request := domain.GenerateRequest{
Prompt: domain.RenderedPrompt{
SessionID: "session",
Messages: []domain.RenderedMessage{
{Role: "user", Content: "content"},
},
},
Target: domain.ExecutionTarget{
BackendID: "limited",
Model: "model",
ExtraParams: map[string]any{"key": "value"},
},
}
response := &domain.GenerateResponse{
Content: "output",
Usage: domain.TokenUsage{TotalTokens: 7},
}
collaboratorErr := errors.New("collaborator failure")
tests := []struct {
name string
response *domain.GenerateResponse
err error
}{
{name: "successful response", response: response},
{name: "nil response"},
{name: "collaborator error", response: response, err: collaboratorErr},
}
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
var captured domain.GenerateRequest
next := clientFunc(func(
_ context.Context,
req domain.GenerateRequest,
) (*domain.GenerateResponse, error) {
captured = req
return tc.response, tc.err
})
gotResponse, gotErr := NewClient(manager, next).Generate(context.Background(), request)
if !reflect.DeepEqual(captured, request) {
t.Fatalf("request changed: %#v", captured)
}
if gotResponse != tc.response || gotErr != tc.err {
t.Fatalf("response=(%p, %v), want (%p, %v)",
gotResponse, gotErr, tc.response, tc.err)
}
})
}
} }
func TestClientReleasesPermitDuringPanicUnwinding(t *testing.T) { func TestClientReleasesPermitDuringPanicUnwinding(t *testing.T) {
@@ -496,7 +450,7 @@ func TestClientReleasesPermitDuringPanicUnwinding(t *testing.T) {
} }
return &domain.GenerateResponse{Content: "recovered"}, nil return &domain.GenerateResponse{Content: "recovered"}, nil
}) })
client := NewClient(manager, next) client := newTestClient(manager, next)
request := domain.GenerateRequest{ request := domain.GenerateRequest{
Target: domain.ExecutionTarget{BackendID: "limited"}, Target: domain.ExecutionTarget{BackendID: "limited"},
} }

View File

@@ -13,7 +13,7 @@ import (
const benchmarkDefinitionCount = 128 const benchmarkDefinitionCount = 128
var ( var (
constructionResult *loaded constructionResult *testLoaded
selectedPrompt *domain.PromptDefinition selectedPrompt *domain.PromptDefinition
selectedPromptHash string selectedPromptHash string
selectedPromptInspection *domain.PromptInspection selectedPromptInspection *domain.PromptInspection

View File

@@ -0,0 +1,16 @@
package configuration
import (
"embed"
"go.maximumdirect.net/configkit"
)
//go:embed bundled/openrouter.yaml
var bundledFiles embed.FS
// BundledOpenRouterUnderlay returns the built-in OpenRouter configuration source.
func BundledOpenRouterUnderlay() configkit.Underlay {
filesystem := configkit.NewFilesystem("Promptkit bundled OpenRouter", bundledFiles)
return configkit.Underlay{Inputs: []configkit.Input{{Source: filesystem.File("bundled/openrouter.yaml"), Destination: configkit.Keys("promptkit")}}}
}

View File

@@ -0,0 +1,60 @@
backends:
- id: openrouter
endpoint: https://openrouter.ai/api/v1
api_key_env: OPENROUTER_API_KEY
smoke_tests:
default_profile: gpt-luna-latest
authentication:
kind: openrouter_key
profiles:
- id: gpt-astra-latest
backend: openrouter
model: ~openai/gpt-astra-latest
- id: gpt-sol-latest
backend: openrouter
model: ~openai/gpt-sol-latest
- id: gpt-terra-latest
backend: openrouter
model: ~openai/gpt-terra-latest
- id: gpt-luna-latest
backend: openrouter
model: ~openai/gpt-luna-latest
- id: claude-fable-latest
backend: openrouter
model: ~anthropic/claude-fable-latest
- id: claude-opus-latest
backend: openrouter
model: ~anthropic/claude-opus-latest
- id: claude-sonnet-latest
backend: openrouter
model: ~anthropic/claude-sonnet-latest
- id: claude-haiku-latest
backend: openrouter
model: ~anthropic/claude-haiku-latest
- id: gemini-pro-latest
backend: openrouter
model: ~google/gemini-pro-latest
- id: gemini-flash-latest
backend: openrouter
model: ~google/gemini-flash-latest
- id: auto
backend: openrouter
model: openrouter/auto
- id: free
backend: openrouter
model: openrouter/free
- id: auto-low
extends: {id: auto}
extra_params: {plugins: [{id: auto-router, cost_tier: low}]}
- id: auto-medium
extends: {id: auto}
extra_params: {plugins: [{id: auto-router, cost_tier: medium}]}
- id: auto-high
extends: {id: auto}
extra_params: {plugins: [{id: auto-router, cost_tier: high}]}
- id: auto-xhigh
extends: {id: auto}
extra_params: {plugins: [{id: auto-router, cost_tier: xhigh}]}
- id: auto-max
extends: {id: auto}
extra_params: {plugins: [{id: auto-router, cost_tier: max}]}

View File

@@ -0,0 +1,54 @@
package configuration
import (
"context"
"fmt"
"reflect"
"strings"
"testing"
"go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit/internal/defaults"
)
func TestBundledOpenRouterUnderlay(t *testing.T) {
state, err := load(configkit.Input{Source: configkit.NoSource()}, []configkit.Underlay{BundledOpenRouterUnderlay()})
if err != nil {
t.Fatal(err)
}
backend, err := state.GetBackend("openrouter")
if err != nil || backend.Endpoint != "https://openrouter.ai/api/v1" || backend.APIKeyEnv != "OPENROUTER_API_KEY" || backend.SmokeTests == nil || backend.SmokeTests.DefaultProfile == nil || *backend.SmokeTests.DefaultProfile != "gpt-luna-latest" || backend.SmokeTests.Authentication == nil || backend.SmokeTests.Authentication.Kind == nil || *backend.SmokeTests.Authentication.Kind != "openrouter_key" {
t.Fatalf("backend = %+v, %v", backend, err)
}
if !reflect.DeepEqual(state.backendIDs, []string{"openrouter"}) || len(state.promptIDs) != 0 || len(state.collections) != 0 {
t.Fatalf("bundle inventories: backends=%v prompts=%v collections=%v", state.backendIDs, state.promptIDs, state.collections)
}
models := map[string]string{
"gpt-astra-latest": "~openai/gpt-astra-latest", "gpt-sol-latest": "~openai/gpt-sol-latest", "gpt-terra-latest": "~openai/gpt-terra-latest", "gpt-luna-latest": "~openai/gpt-luna-latest",
"claude-fable-latest": "~anthropic/claude-fable-latest", "claude-opus-latest": "~anthropic/claude-opus-latest", "claude-sonnet-latest": "~anthropic/claude-sonnet-latest", "claude-haiku-latest": "~anthropic/claude-haiku-latest",
"gemini-pro-latest": "~google/gemini-pro-latest", "gemini-flash-latest": "~google/gemini-flash-latest", "auto": "openrouter/auto", "free": "openrouter/free",
}
for _, tier := range []string{"low", "medium", "high", "xhigh", "max"} {
models["auto-"+tier] = "openrouter/auto"
}
if len(state.profileIDs) != len(models) {
t.Fatalf("profiles = %v", state.profileIDs)
}
for id, model := range models {
profile, err := state.GetProfile(context.Background(), id)
if err != nil {
t.Fatalf("profile %s = %+v, %v", id, profile, err)
}
want := defaults.ExecutionTargetDefault()
want.Model, want.BackendID, want.Endpoint, want.APIKeyEnv = model, "openrouter", "https://openrouter.ai/api/v1", "OPENROUTER_API_KEY"
if tier, ok := strings.CutPrefix(id, "auto-"); ok {
want.ExtraParams = map[string]any{"plugins": []any{map[string]any{"id": "auto-router", "cost_tier": tier}}}
}
if !reflect.DeepEqual(*profile.ResolvedTarget, want) {
t.Fatalf("profile %s target = %+v, want %+v", id, *profile.ResolvedTarget, want)
}
}
if !strings.Contains(fmt.Sprint(state.inspection.Sources()), "Promptkit bundled OpenRouter") {
t.Fatalf("source evidence = %+v", state.inspection.Sources())
}
}

View File

@@ -0,0 +1,151 @@
package capturetest
import (
"context"
"io/fs"
"testing"
"testing/fstest"
"go.maximumdirect.net/configkit"
"go.maximumdirect.net/promptkit"
"go.maximumdirect.net/promptkit/internal/configuration"
)
type application struct {
PromptKit promptkit.Settings `yaml:"promptkit"`
Other string `yaml:"other"`
}
func TestCaptureUsesEffectiveSourceContexts(t *testing.T) {
baseFiles := fstest.MapFS{
"catalog/base.yaml": {Data: []byte("promptkit:\n prompts:\n - id: base\n messages: [{role: user, content_file: original.txt}]\n output: {format: json, validation_mode: json_schema, schema_file: schema.json}\n")},
"catalog/original.txt": {Data: []byte("original")},
"catalog/schema.json": {Data: []byte(`{"type":"object"}`)},
}
overrideFiles := fstest.MapFS{
"operator/main.yaml": {Data: []byte("promptkit:\n prompts:\n - id: child\n extends: {id: base}\n messages: [{role: user, content_file: replacement.txt}]\n")},
"operator/replacement.txt": {Data: []byte("replacement")},
}
base := configkit.NewFilesystem("base", baseFiles)
override := configkit.NewFilesystem("operator", overrideFiles)
definition := promptkit.Configure(configkit.Definition[application]{}, func(value *application) *promptkit.Settings { return &value.PromptKit })
result, err := configkit.LoadLayered(configkit.Input{Source: override.File("operator/main.yaml"), Filesystem: &configkit.FilesystemContext{FS: &override, Root: "."}}, configkit.LayeredOptions[application]{
Underlays: []configkit.Underlay{{Inputs: []configkit.Input{{Source: base.File("catalog/base.yaml"), Filesystem: &configkit.FilesystemContext{FS: &base, Root: "."}}}}}, Definition: definition,
})
if err != nil {
t.Fatal(err)
}
view, ok := result.InspectSubtree(configkit.Keys("promptkit"))
if !ok {
t.Fatal("inspection unavailable")
}
state, err := configuration.Capture(&result.Config.PromptKit, view, nil)
if err != nil {
t.Fatal(err)
}
child, _, err := state.GetPromptSnapshot(context.Background(), "child")
if err != nil || child.Templates[0].Content != "replacement" || child.SchemaFile == nil || child.SchemaFile.Path != "catalog/schema.json" {
t.Fatalf("child = %+v, %v", child, err)
}
data, err := fs.ReadFile(child.SchemaFile.FS, child.SchemaFile.Path)
if err != nil || string(data) != `{"type":"object"}` {
t.Fatalf("schema = %q, %v", data, err)
}
}
func TestCapturedInspectionUsesNativeView(t *testing.T) {
type anotherApplication struct {
PromptKit promptkit.Settings `yaml:"promptkit"`
Sibling int `yaml:"sibling"`
}
definition := promptkit.Configure(configkit.Definition[application]{}, func(value *application) *promptkit.Settings { return &value.PromptKit })
first, err := configkit.Load(configkit.Bytes("first", []byte("promptkit: {application_field: first}\n")), configkit.Options[application]{Definition: definition})
if err != nil {
t.Fatal(err)
}
secondDefinition := promptkit.Configure(configkit.Definition[anotherApplication]{}, func(value *anotherApplication) *promptkit.Settings { return &value.PromptKit })
second, err := configkit.Load(configkit.Bytes("second", []byte("promptkit: {application_field: second}\n")), configkit.Options[anotherApplication]{Definition: secondDefinition})
if err != nil {
t.Fatal(err)
}
leftView, _ := first.InspectSubtree(configkit.Keys("promptkit"))
rightView, _ := second.InspectSubtree(configkit.Keys("promptkit"))
diagnostics := []configkit.Diagnostic{{Message: "warning", Primary: &configkit.Location{SourceID: "first"}, Related: []configkit.Location{{SourceID: "related"}}}}
left, err := configuration.Capture(&first.Config.PromptKit, leftView, diagnostics)
if err != nil {
t.Fatal(err)
}
right, err := configuration.Capture(&second.Config.PromptKit, rightView, nil)
if err != nil {
t.Fatal(err)
}
diagnostics[0].Primary.SourceID = "changed"
diagnostics[0].Related[0].SourceID = "changed"
got := left.Inspection().Diagnostics()
if got[0].Primary.SourceID != "first" || got[0].Related[0].SourceID != "related" {
t.Fatalf("captured diagnostics = %+v", got)
}
got[0].Primary.SourceID = "again"
got[0].Related[0].SourceID = "again"
got = left.Inspection().Diagnostics()
if got[0].Primary.SourceID != "first" || got[0].Related[0].SourceID != "related" {
t.Fatalf("returned diagnostics aliased: %+v", got)
}
differences, err := configuration.Diff(left.Inspection(), right.Inspection())
if err != nil || len(differences) == 0 || len(left.Inspection().HistoricalPaths()) == 0 {
t.Fatalf("diff = %+v, %v", differences, err)
}
}
func TestCaptureOwnsEffectiveResources(t *testing.T) {
definition := promptkit.Configure(configkit.Definition[application]{}, func(value *application) *promptkit.Settings { return &value.PromptKit })
result, err := configkit.Load(configkit.Bytes("application", []byte(`other: application
promptkit:
backends:
- id: local
endpoint: https://example.test/v1
extra_params: {nested: {value: original}}
profiles:
- id: profile
backend: local
model: original
prompts:
- id: prompt
inputs: [{name: document, required: false, description: text}]
messages: [{role: user, content: hello}]
output:
format: json
validation_mode: json_schema
schema: {type: object, properties: {answer: {type: string}}}
`)), configkit.Options[application]{Definition: definition})
if err != nil {
t.Fatal(err)
}
view, ok := result.InspectSubtree(configkit.Keys("promptkit"))
if !ok {
t.Fatal("inspection unavailable")
}
state, err := configuration.Capture(&result.Config.PromptKit, view, result.Diagnostics())
if err != nil {
t.Fatal(err)
}
settings := &result.Config.PromptKit
*settings.Backends[0].Endpoint = "https://changed.test/v1"
settings.Backends[0].ExtraParams["nested"].(map[string]any)["value"] = "changed"
*settings.Profiles[0].Model = "changed"
*settings.Prompts[0].Messages[0].Content = "changed"
settings.Prompts[0].Inputs[0].Required = true
settings.Prompts[0].Output.Schema.(map[string]any)["type"] = "changed"
backend, err := state.InspectBackend(context.Background(), "local")
if err != nil || backend.Endpoint != "https://example.test/v1" || backend.ExtraParams["nested"].(map[string]any)["value"] != "original" {
t.Fatalf("backend = %+v, %v", backend, err)
}
profile, err := state.GetProfile(context.Background(), "profile")
if err != nil || profile.ResolvedTarget.Model != "original" {
t.Fatalf("profile = %+v, %v", profile, err)
}
prompt, err := state.InspectPrompt(context.Background(), "prompt")
if err != nil || prompt.Messages[0].Content != "hello" || prompt.Output.Schema.(map[string]any)["type"] != "object" || prompt.Inputs == nil || (*prompt.Inputs)[0].Required == nil || *(*prompt.Inputs)[0].Required {
t.Fatalf("prompt = %+v, %v", prompt, err)
}
}

View File

@@ -8,24 +8,30 @@ import (
"go.maximumdirect.net/configkit" "go.maximumdirect.net/configkit"
) )
// Inspection is the read-only subset of the retained ConfigKit result used by // Inspection is the read-only subset of the native ConfigKit view used by
// the facade. It does not expose typed configuration or resource contexts. // the facade. It does not expose typed configuration or resource contexts.
type Inspection interface { type Inspection interface {
Metadata() configkit.Metadata Metadata() configkit.Metadata
Sources() []configkit.SourceInfo Sources() []configkit.SourceInfo
Diagnostics() []configkit.Diagnostic Diagnostics() []configkit.Diagnostic
HistoricalPaths() []configkit.Path
Explain(configkit.Path) (configkit.Explanation, bool) Explain(configkit.Path) (configkit.Explanation, bool)
WriteEffective(io.Writer) error WriteEffective(io.Writer) error
WriteExplanation(io.Writer, configkit.Path) error WriteExplanation(io.Writer, configkit.Path) error
capturedResult() *configkit.Result[document] nativeView() configkit.Inspection
} }
type inspectionResult struct { type inspectionResult struct {
*configkit.Result[document] configkit.Inspection
diagnostics []configkit.Diagnostic
} }
func (inspection inspectionResult) capturedResult() *configkit.Result[document] { func (inspection inspectionResult) nativeView() configkit.Inspection {
return inspection.Result return inspection.Inspection
}
func (inspection inspectionResult) Diagnostics() []configkit.Diagnostic {
return ownDiagnostics(inspection.diagnostics)
} }
// Inspection returns only the retained result's inspection operations. // Inspection returns only the retained result's inspection operations.
@@ -33,27 +39,25 @@ func (state *loaded) Inspection() Inspection {
if state == nil { if state == nil {
return nil return nil
} }
return inspectionResult{Result: state.result} return inspectionResult{Inspection: state.view, diagnostics: state.diagnostics}
} }
// Diff compares two captured ConfigKit results without reconstructing them. // Diff compares two captured ConfigKit results without reconstructing them.
func Diff(left, right Inspection) ([]configkit.Difference, error) { func Diff(left, right Inspection) ([]configkit.Difference, error) {
var leftResult, rightResult *configkit.Result[document] return left.nativeView().Diff(right.nativeView())
if left != nil {
leftResult = left.capturedResult()
}
if right != nil {
rightResult = right.capturedResult()
}
return configkit.Diff(leftResult, rightResult)
} }
func secretProviderPath(path configkit.Path) bool { // Sensitive marks credential-shaped provider values beneath PromptKit settings.
func Sensitive(path configkit.Path) bool {
pointer := path.String() pointer := path.String()
if pointer == "" { if pointer == "" {
return false return false
} }
parts := strings.Split(strings.TrimPrefix(pointer, "/"), "/") parts := strings.Split(strings.TrimPrefix(pointer, "/"), "/")
if len(parts) == 0 || parts[0] != "promptkit" {
return false
}
parts = parts[1:]
if !providerExtraParamsPath(parts) { if !providerExtraParamsPath(parts) {
return false return false
} }

View File

@@ -12,20 +12,20 @@ func TestSecretProviderPathRecognizesResourceExtraParams(t *testing.T) {
path configkit.Path path configkit.Path
want bool want bool
}{ }{
{name: "backend nested secret", path: configkit.Keys("backends").Index(0).Key("extra_params").Key("nested").Key("Token"), want: true}, {name: "backend nested secret", path: configkit.Keys("promptkit", "backends").Index(0).Key("extra_params").Key("nested").Key("Token"), want: true},
{name: "profile nested secret", path: configkit.Keys("profiles").Index(1).Key("extra_params").Key("authorization"), want: true}, {name: "profile nested secret", path: configkit.Keys("promptkit", "profiles").Index(1).Key("extra_params").Key("authorization"), want: true},
{name: "collection backend nested secret", path: configkit.Keys("collections", "provider", "backends").Index(2).Key("extra_params").Key("nested").Key("Token"), want: true}, {name: "collection backend nested secret", path: configkit.Keys("promptkit", "collections", "provider", "backends").Index(2).Key("extra_params").Key("nested").Key("Token"), want: true},
{name: "collection profile secret", path: configkit.Keys("collections", "assistant", "profiles").Index(0).Key("extra_params").Key("authorization"), want: true}, {name: "collection profile secret", path: configkit.Keys("promptkit", "collections", "assistant", "profiles").Index(0).Key("extra_params").Key("authorization"), want: true},
{name: "unrelated nested mapping", path: configkit.Keys("application", "backends", "provider", "extra_params", "token"), want: false}, {name: "unrelated nested mapping", path: configkit.Keys("application", "backends", "provider", "extra_params", "token"), want: false},
{name: "extra params outside resource map", path: configkit.Keys("backends", "provider", "options", "extra_params", "token"), want: false}, {name: "extra params outside resource map", path: configkit.Keys("promptkit", "backends", "provider", "options", "extra_params", "token"), want: false},
{name: "collection extra params outside resource map", path: configkit.Keys("collections", "provider", "settings", "backends", "default", "extra_params", "token"), want: false}, {name: "collection extra params outside resource map", path: configkit.Keys("promptkit", "collections", "provider", "settings", "backends", "default", "extra_params", "token"), want: false},
{name: "missing resource key", path: configkit.Keys("backends", "", "extra_params", "token"), want: false}, {name: "missing resource key", path: configkit.Keys("promptkit", "backends", "", "extra_params", "token"), want: false},
{name: "ordinary provider value", path: configkit.Keys("profiles").Index(0).Key("extra_params").Key("provider").Key("mode"), want: false}, {name: "ordinary provider value", path: configkit.Keys("promptkit", "profiles").Index(0).Key("extra_params").Key("provider").Key("mode"), want: false},
} }
for _, test := range tests { for _, test := range tests {
t.Run(test.name, func(t *testing.T) { t.Run(test.name, func(t *testing.T) {
if got := secretProviderPath(test.path); got != test.want { if got := Sensitive(test.path); got != test.want {
t.Fatalf("secretProviderPath(%q) = %t, want %t", test.path, got, test.want) t.Fatalf("Sensitive(%q) = %t, want %t", test.path, got, test.want)
} }
}) })
} }

View File

@@ -4,11 +4,12 @@ import (
"context" "context"
"strings" "strings"
"testing" "testing"
"testing/fstest"
"go.maximumdirect.net/configkit" "go.maximumdirect.net/configkit"
) )
func TestTargetInterpretationPreservesComposedNestedValues(t *testing.T) { func TestTargetResolutionNormalizesNestedValues(t *testing.T) {
state, err := load(mapping("padded target", map[string]any{ state, err := load(mapping("padded target", map[string]any{
"backends": []any{map[string]any{ "backends": []any{map[string]any{
"id": "local", "endpoint": " HTTPS://example.invalid/v1 ", "concurrency_limit": 1, "id": "local", "endpoint": " HTTPS://example.invalid/v1 ", "concurrency_limit": 1,
@@ -32,41 +33,125 @@ func TestTargetInterpretationPreservesComposedNestedValues(t *testing.T) {
if err != nil || profile.ResolvedTarget.ServiceTier != "standard" || profile.ResolvedTarget.APIKeyEnv != "TOKEN" { if err != nil || profile.ResolvedTarget.ServiceTier != "standard" || profile.ResolvedTarget.APIKeyEnv != "TOKEN" {
t.Fatalf("interpreted profile = %+v, %v", profile, err) t.Fatalf("interpreted profile = %+v, %v", profile, err)
} }
composed := state.result.Config.Backends[0] composed := state.backendRecords["local"]
if *composed.Endpoint != " HTTPS://example.invalid/v1 " || composed.QueueCapacity != nil || if *composed.Endpoint != "https://example.invalid/v1" || composed.QueueCapacity == nil || *composed.QueueCapacity != 1024 ||
*composed.SmokeTests.Reachability.Endpoint != " https://example.invalid/health " || *composed.SmokeTests.Reachability.Endpoint != "https://example.invalid/health" ||
*composed.SmokeTests.Authentication.Kind != " NONE " || *composed.SmokeTests.Authentication.Endpoint != " https://example.invalid/key " || *composed.SmokeTests.Authentication.Kind != "none" || *composed.SmokeTests.Authentication.Endpoint != "https://example.invalid/key" ||
*state.result.Config.Profiles[0].ServiceTier != " standard " || *state.result.Config.Profiles[0].APIKeyEnv != " TOKEN " { *state.profileRecords["partial"].ServiceTier != "standard" || *state.profileRecords["partial"].APIKeyEnv != "TOKEN" {
t.Fatalf("composed values were mutated: %+v", state.result.Config) t.Fatalf("resolved values were not normalized: %+v", composed)
} }
path := configkit.Keys("backends").Index(0).Key("smoke_tests").Key("authentication").Key("endpoint") path := configkit.Keys("promptkit", "backends").Index(0).Key("smoke_tests").Key("authentication").Key("endpoint")
field, found := state.result.Metadata().Lookup(path) field, found := state.inspection.Metadata().Lookup(path)
if !found || len(field.Origin.Locations) == 0 || field.Origin.Operation == configkit.ChangeNormalization { if !found || field.Origin.Operation != configkit.ChangeNormalization {
t.Fatalf("nested field origin changed: %+v", field) t.Fatalf("nested field origin changed: %+v", field)
} }
for _, path := range []configkit.Path{
configkit.Keys("promptkit", "backends").Index(0).Key("endpoint"),
configkit.Keys("promptkit", "backends").Index(0).Key("smoke_tests").Key("reachability").Key("endpoint"),
configkit.Keys("promptkit", "backends").Index(0).Key("smoke_tests").Key("authentication").Key("kind"),
configkit.Keys("promptkit", "profiles").Index(0).Key("service_tier"),
configkit.Keys("promptkit", "profiles").Index(0).Key("api_key_env"),
} {
field, found := state.inspection.Metadata().Lookup(path)
if !found || field.Origin.Operation != configkit.ChangeNormalization {
t.Errorf("normalization at %v: %+v", path, field)
}
}
} }
func TestPromptInterpretationPreservesComposedOutputAndRejectsDuplicateNames(t *testing.T) { func TestPromptResolutionNormalizesOutputAndRejectsDuplicateNames(t *testing.T) {
definition := map[string]any{ definition := map[string]any{
"id": "example", "inputs": []any{map[string]any{"name": " document "}}, "id": "example", "inputs": []any{map[string]any{"name": " document "}},
"messages": []any{map[string]any{"role": "user", "content": "hello"}}, "messages": []any{map[string]any{"role": "user", "content": "hello"}},
"output": map[string]any{"format": "json", "validation_mode": "json_schema", "schema_path": " schema.json "}, "output": map[string]any{"format": "json", "validation_mode": "json_schema", "schema_path": " schema.json "},
} }
state, err := load(mapping("prompt metadata", map[string]any{"prompts": []any{definition}}), nil) filesystem := configkit.NewFilesystem("schemas", fstest.MapFS{})
input := mapping("prompt metadata", map[string]any{"schema_root": ".", "prompts": []any{definition}})
input.Filesystem = &configkit.FilesystemContext{FS: &filesystem, Root: ".", BaseDir: "."}
state, err := load(input, nil)
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
selected, inspection, err := state.prompt("example") selected, inspection, err := state.prompt("example")
if err != nil || selected.Validation.SchemaPath != "schema.json" || inspection.OutputContract.SchemaPath != "schema.json" || selected.Inputs[0].Name != "document" { if err != nil || selected.Validation.SchemaPath != "schema.json" || selected.Inputs[0].Name != "document" {
t.Fatalf("interpreted prompt = %+v, %+v, %v", selected, inspection, err) t.Fatalf("interpreted prompt = %+v, %+v, %v", selected, inspection, err)
} }
composed := state.result.Config.Prompts[0] composed := state.prompts["example"].resource
if *composed.Output.SchemaPath != " schema.json " || composed.Inputs[0].Name != " document " { if *composed.Output.SchemaPath != "schema.json" || composed.Inputs[0].Name != "document" {
t.Fatalf("composed prompt changed: %+v", composed) t.Fatalf("prompt was not normalized: %+v", composed)
} }
definition["inputs"] = []any{map[string]any{"name": " document "}, map[string]any{"name": "document"}} definition["inputs"] = []any{map[string]any{"name": " document "}, map[string]any{"name": "document"}}
_, err = load(mapping("duplicate inputs", map[string]any{"prompts": []any{definition}}), nil) _, err = load(mapping("duplicate inputs", map[string]any{"prompts": []any{definition}}), nil)
if err == nil || !strings.Contains(err.Error(), "/prompts/0/inputs/1") || !strings.Contains(err.Error(), "duplicated") { if err == nil || !strings.Contains(err.Error(), "/promptkit/prompts/0/inputs/1") || !strings.Contains(err.Error(), "duplicated") {
t.Fatalf("duplicate normalized names were not rejected: %v", err) t.Fatalf("duplicate normalized names were not rejected: %v", err)
} }
} }
func TestPromptResolutionRecordsMessageFields(t *testing.T) {
loadPrompt := func(role, cacheType, ttl string) *testLoaded {
t.Helper()
state, err := load(mapping("prompt", map[string]any{"prompts": []any{map[string]any{
"id": "example", "inputs": []any{map[string]any{"name": " input ", "content_type": nil, "description": " text "}},
"messages": []any{map[string]any{"role": role, "content": "", "cache_control": map[string]any{"type": cacheType, "ttl": ttl}}},
"output": map[string]any{"format": "text", "validation_mode": "basic"},
}}}), nil)
if err != nil {
t.Fatal(err)
}
return state
}
padded := loadPrompt(" USER ", " ephemeral ", " 1h ")
canonical := loadPrompt("user", "ephemeral", "1h")
definition, inspection, err := padded.prompt("example")
if err != nil {
t.Fatal(err)
}
other, otherInspection, err := canonical.prompt("example")
if err != nil || inspection.PromptHash != otherInspection.PromptHash || definition.Templates[0].Role != other.Templates[0].Role || definition.Templates[0].Content != "" {
t.Fatalf("equivalent prompt mismatch: %+v %+v %v", definition, other, err)
}
base := configkit.Keys("promptkit", "prompts").Index(0)
for _, path := range []configkit.Path{base.Key("inputs").Index(0).Key("name"), base.Key("inputs").Index(0).Key("description"), base.Key("messages").Index(0).Key("role"), base.Key("messages").Index(0).Key("cache_control").Key("type"), base.Key("messages").Index(0).Key("cache_control").Key("ttl")} {
field, found := padded.inspection.Metadata().Lookup(path)
if !found || field.Origin.Operation != configkit.ChangeNormalization {
t.Errorf("normalization at %v: %+v", path, field)
}
}
field, ok := padded.inspection.Metadata().Lookup(base.Key("inputs").Index(0).Key("content_type"))
if !ok || !field.Null {
t.Fatal("null input metadata was not retained")
}
}
func TestResolutionRecordsTargetDependencies(t *testing.T) {
backend := func(id string) map[string]any {
return map[string]any{"id": id, "endpoint": " HTTPS://example.invalid/v1 ", "concurrency_limit": 2, "queue_capacity": nil}
}
state, err := load(mapping("targets", map[string]any{
"backends": []any{backend("root")},
"collections": map[string]any{"scoped": map[string]any{"backends": []any{backend("local")}}},
}), nil)
if err != nil {
t.Fatal(err)
}
for _, base := range []configkit.Path{configkit.Keys("promptkit", "backends").Index(0), configkit.Keys("promptkit", "collections", "scoped", "backends").Index(0)} {
endpoint, ok := state.inspection.Explain(base.Key("endpoint"))
if !ok || !strings.Contains(endpoint.Effective.YAML, "https://example.invalid/v1") {
t.Fatalf("endpoint explanation: %+v", endpoint)
}
history := state.inspection.Metadata().History(base.Key("queue_capacity"))
found := false
for _, change := range history {
if change.Origin.Operation == configkit.ChangeDerivedDefault {
for _, dependency := range change.Dependencies {
if dependency.String() == base.Key("concurrency_limit").String() {
found = true
}
}
}
}
if !found {
t.Fatalf("missing queue dependency: %+v", history)
}
}
}

View File

@@ -2,36 +2,66 @@ package configuration
import ( import (
"context" "context"
"sort"
"strings"
"go.maximumdirect.net/promptkit/internal/domain" "go.maximumdirect.net/promptkit/internal/domain"
) )
// IdentityFilter selects all identities, one namespace, or unqualified IDs.
type IdentityFilter struct {
All bool
Namespace string
}
func selectIdentities[T any](ctx context.Context, entries []T, filter IdentityFilter, id func(T) string) ([]T, error) {
if err := ctx.Err(); err != nil {
return nil, err
}
if filter.All {
result := append([]T(nil), entries...)
if err := ctx.Err(); err != nil {
return nil, err
}
return result, nil
}
if filter.Namespace != "" {
prefix := filter.Namespace + "/"
start := sort.Search(len(entries), func(i int) bool { return id(entries[i]) >= prefix })
end := start + sort.Search(len(entries)-start, func(i int) bool { return !strings.HasPrefix(id(entries[start+i]), prefix) })
result := append([]T(nil), entries[start:end]...)
if err := ctx.Err(); err != nil {
return nil, err
}
return result, nil
}
var result []T
for _, entry := range entries {
if err := ctx.Err(); err != nil {
return nil, err
}
if !strings.Contains(id(entry), "/") {
result = append(result, entry)
}
}
if err := ctx.Err(); err != nil {
return nil, err
}
return result, nil
}
// CollectionIdentity is the captured metadata for one final collection entry. // CollectionIdentity is the captured metadata for one final collection entry.
type CollectionIdentity struct { type CollectionIdentity struct {
Namespace string Namespace string
ReleaseVersion *string ReleaseVersion *string
} }
func (state *loaded) ListProfiles(ctx context.Context) ([]string, error) { func (state *loaded) ListProfiles(ctx context.Context, filter IdentityFilter) ([]string, error) {
if err := ctx.Err(); err != nil { return selectIdentities(ctx, state.profileIDs, filter, func(id string) string { return id })
return nil, err
}
profiles := append([]string(nil), state.profileIDs...)
if err := ctx.Err(); err != nil {
return nil, err
}
return profiles, nil
} }
func (state *loaded) ListBackends(ctx context.Context) ([]string, error) { func (state *loaded) ListBackends(ctx context.Context, filter IdentityFilter) ([]string, error) {
if err := ctx.Err(); err != nil { return selectIdentities(ctx, state.backendIDs, filter, func(id string) string { return id })
return nil, err
}
backends := append([]string(nil), state.backendIDs...)
if err := ctx.Err(); err != nil {
return nil, err
}
return backends, nil
} }
func (state *loaded) ListNamespaces(ctx context.Context, kinds []string) ([]string, error) { func (state *loaded) ListNamespaces(ctx context.Context, kinds []string) ([]string, error) {
@@ -72,13 +102,6 @@ func (state *loaded) ListCollections(ctx context.Context) ([]CollectionIdentity,
return collections, nil return collections, nil
} }
func (state *loaded) ListPrompts(ctx context.Context) ([]domain.PromptIdentity, error) { func (state *loaded) ListPrompts(ctx context.Context, filter IdentityFilter) ([]domain.PromptIdentity, error) {
if err := ctx.Err(); err != nil { return selectIdentities(ctx, state.promptIDs, filter, func(entry domain.PromptIdentity) string { return entry.ID })
return nil, err
}
prompts := append([]domain.PromptIdentity(nil), state.promptIDs...)
if err := ctx.Err(); err != nil {
return nil, err
}
return prompts, nil
} }

View File

@@ -11,7 +11,7 @@ import (
) )
func TestLoadedListsSortedEffectiveIdentities(t *testing.T) { func TestLoadedListsSortedEffectiveIdentities(t *testing.T) {
state, err := Load(mapping("identities", map[string]any{ state, err := load(mapping("identities", map[string]any{
"backends": []any{map[string]any{"id": "local", "endpoint": "http://localhost:8080/v1"}}, "backends": []any{map[string]any{"id": "local", "endpoint": "http://localhost:8080/v1"}},
"collections": map[string]any{ "collections": map[string]any{
"z": map[string]any{"profiles": []any{map[string]any{"id": "profile", "backend": "local", "model": "z-model"}}, "prompts": []any{listingPromptDefinition("prompt")}}, "z": map[string]any{"profiles": []any{map[string]any{"id": "profile", "backend": "local", "model": "z-model"}}, "prompts": []any{listingPromptDefinition("prompt")}},
@@ -23,7 +23,7 @@ func TestLoadedListsSortedEffectiveIdentities(t *testing.T) {
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
backends, err := state.ListBackends(context.Background()) backends, err := state.ListBackends(context.Background(), IdentityFilter{All: true})
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
@@ -32,7 +32,7 @@ func TestLoadedListsSortedEffectiveIdentities(t *testing.T) {
t.Fatalf("backend IDs = %#v, want %#v", backends, wantBackends) t.Fatalf("backend IDs = %#v, want %#v", backends, wantBackends)
} }
profiles, err := state.ListProfiles(context.Background()) profiles, err := state.ListProfiles(context.Background(), IdentityFilter{All: true})
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
@@ -41,7 +41,7 @@ func TestLoadedListsSortedEffectiveIdentities(t *testing.T) {
t.Fatalf("profile IDs = %#v, want %#v", profiles, wantProfiles) t.Fatalf("profile IDs = %#v, want %#v", profiles, wantProfiles)
} }
prompts, err := state.ListPrompts(context.Background()) prompts, err := state.ListPrompts(context.Background(), IdentityFilter{All: true})
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
@@ -56,7 +56,7 @@ func TestLoadedListsSortedEffectiveIdentities(t *testing.T) {
} }
func TestLoadedListsAreIndependent(t *testing.T) { func TestLoadedListsAreIndependent(t *testing.T) {
state, err := Load(mapping("identities", map[string]any{ state, err := load(mapping("identities", map[string]any{
"backends": []any{map[string]any{"id": "backend", "endpoint": "https://example.test/v1"}}, "backends": []any{map[string]any{"id": "backend", "endpoint": "https://example.test/v1"}},
"profiles": []any{map[string]any{"id": "profile"}}, "profiles": []any{map[string]any{"id": "profile"}},
"prompts": []any{listingPromptDefinition("example")}, "prompts": []any{listingPromptDefinition("example")},
@@ -65,36 +65,36 @@ func TestLoadedListsAreIndependent(t *testing.T) {
t.Fatal(err) t.Fatal(err)
} }
profiles, err := state.ListProfiles(context.Background()) profiles, err := state.ListProfiles(context.Background(), IdentityFilter{All: true})
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
profiles = append(profiles, "caller profile") profiles = append(profiles, "caller profile")
profiles[0] = "changed profile" profiles[0] = "changed profile"
backends, err := state.ListBackends(context.Background()) backends, err := state.ListBackends(context.Background(), IdentityFilter{All: true})
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
backends[0] = "changed backend" backends[0] = "changed backend"
prompts, err := state.ListPrompts(context.Background()) prompts, err := state.ListPrompts(context.Background(), IdentityFilter{All: true})
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
prompts[0].ID = "changed prompt" prompts[0].ID = "changed prompt"
profiles, err = state.ListProfiles(context.Background()) profiles, err = state.ListProfiles(context.Background(), IdentityFilter{All: true})
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
if !reflect.DeepEqual(profiles, []string{"profile"}) { if !reflect.DeepEqual(profiles, []string{"profile"}) {
t.Fatalf("caller mutation changed profile inventory: %#v", profiles) t.Fatalf("caller mutation changed profile inventory: %#v", profiles)
} }
backends, err = state.ListBackends(context.Background()) backends, err = state.ListBackends(context.Background(), IdentityFilter{All: true})
if err != nil || !reflect.DeepEqual(backends, []string{"backend"}) { if err != nil || !reflect.DeepEqual(backends, []string{"backend"}) {
t.Fatalf("caller mutation changed backend inventory: %#v, %v", backends, err) t.Fatalf("caller mutation changed backend inventory: %#v, %v", backends, err)
} }
prompts, err = state.ListPrompts(context.Background()) prompts, err = state.ListPrompts(context.Background(), IdentityFilter{All: true})
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
@@ -104,27 +104,73 @@ func TestLoadedListsAreIndependent(t *testing.T) {
} }
} }
func TestLoadedListsSelectNamespaces(t *testing.T) {
ids := []string{"a", "a-b/one", "a/one", "a/two", "b", "z/one"}
state := &loaded{profileIDs: append([]string(nil), ids...), backendIDs: append([]string(nil), ids...)}
for _, id := range ids {
state.promptIDs = append(state.promptIDs, domain.PromptIdentity{ID: id})
}
cases := []struct {
name string
filter IdentityFilter
want []string
}{
{"all", IdentityFilter{All: true}, ids},
{"unqualified", IdentityFilter{}, []string{"a", "b"}},
{"namespace", IdentityFilter{Namespace: "a"}, []string{"a/one", "a/two"}},
{"prefix sharing", IdentityFilter{Namespace: "a-b"}, []string{"a-b/one"}},
{"missing", IdentityFilter{Namespace: "absent"}, nil},
}
for _, tc := range cases {
t.Run(tc.name, func(t *testing.T) {
profiles, err := state.ListProfiles(context.Background(), tc.filter)
if err != nil || !reflect.DeepEqual(profiles, tc.want) {
t.Fatalf("profiles = %v, %v; want %v", profiles, err, tc.want)
}
backends, err := state.ListBackends(context.Background(), tc.filter)
if err != nil || !reflect.DeepEqual(backends, tc.want) {
t.Fatalf("backends = %v, %v; want %v", backends, err, tc.want)
}
prompts, err := state.ListPrompts(context.Background(), tc.filter)
if err != nil || len(prompts) != len(tc.want) {
t.Fatalf("prompts = %v, %v; want %v", prompts, err, tc.want)
}
for i := range prompts {
if prompts[i].ID != tc.want[i] {
t.Fatalf("prompt ID = %q; want %q", prompts[i].ID, tc.want[i])
}
}
if len(profiles) > 0 {
profiles[0], backends[0], prompts[0].ID = "changed", "changed", "changed"
}
})
}
if !reflect.DeepEqual(state.profileIDs, ids) || !reflect.DeepEqual(state.backendIDs, ids) || state.promptIDs[0].ID != ids[0] {
t.Fatal("caller mutation changed captured identities")
}
}
func TestLoadedListsEmptyInventory(t *testing.T) { func TestLoadedListsEmptyInventory(t *testing.T) {
state, err := Load(configkit.Input{Source: configkit.NoSource()}, nil) state, err := load(configkit.Input{Source: configkit.NoSource()}, nil)
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
profiles, err := state.ListProfiles(context.Background()) profiles, err := state.ListProfiles(context.Background(), IdentityFilter{All: true})
if err != nil || len(profiles) != 0 { if err != nil || len(profiles) != 0 {
t.Fatalf("empty profile listing = %#v, %v", profiles, err) t.Fatalf("empty profile listing = %#v, %v", profiles, err)
} }
backends, err := state.ListBackends(context.Background()) backends, err := state.ListBackends(context.Background(), IdentityFilter{All: true})
if err != nil || len(backends) != 0 { if err != nil || len(backends) != 0 {
t.Fatalf("empty backend listing = %#v, %v", backends, err) t.Fatalf("empty backend listing = %#v, %v", backends, err)
} }
prompts, err := state.ListPrompts(context.Background()) prompts, err := state.ListPrompts(context.Background(), IdentityFilter{All: true})
if err != nil || len(prompts) != 0 { if err != nil || len(prompts) != 0 {
t.Fatalf("empty prompt listing = %#v, %v", prompts, err) t.Fatalf("empty prompt listing = %#v, %v", prompts, err)
} }
} }
func TestLoadedListsHonorCanceledContext(t *testing.T) { func TestLoadedListsHonorCanceledContext(t *testing.T) {
state, err := Load(mapping("identities", map[string]any{ state, err := load(mapping("identities", map[string]any{
"profiles": []any{map[string]any{"id": "partial"}}, "profiles": []any{map[string]any{"id": "partial"}},
"prompts": []any{listingPromptDefinition("example")}, "prompts": []any{listingPromptDefinition("example")},
}), nil) }), nil)
@@ -134,17 +180,19 @@ func TestLoadedListsHonorCanceledContext(t *testing.T) {
ctx, cancel := context.WithCancel(context.Background()) ctx, cancel := context.WithCancel(context.Background())
cancel() cancel()
profiles, err := state.ListProfiles(ctx) for _, filter := range []IdentityFilter{{All: true}, {}, {Namespace: "example"}} {
if !errors.Is(err, context.Canceled) || len(profiles) != 0 { profiles, err := state.ListProfiles(ctx, filter)
t.Fatalf("canceled profile listing = %#v, %v", profiles, err) if !errors.Is(err, context.Canceled) || len(profiles) != 0 {
} t.Fatalf("canceled profile listing = %#v, %v", profiles, err)
backends, err := state.ListBackends(ctx) }
if !errors.Is(err, context.Canceled) || len(backends) != 0 { backends, err := state.ListBackends(ctx, filter)
t.Fatalf("canceled backend listing = %#v, %v", backends, err) if !errors.Is(err, context.Canceled) || len(backends) != 0 {
} t.Fatalf("canceled backend listing = %#v, %v", backends, err)
prompts, err := state.ListPrompts(ctx) }
if !errors.Is(err, context.Canceled) || len(prompts) != 0 { prompts, err := state.ListPrompts(ctx, filter)
t.Fatalf("canceled prompt listing = %#v, %v", prompts, err) if !errors.Is(err, context.Canceled) || len(prompts) != 0 {
t.Fatalf("canceled prompt listing = %#v, %v", prompts, err)
}
} }
} }

View File

@@ -4,7 +4,6 @@ package configuration
import ( import (
"fmt" "fmt"
"net/http"
"regexp" "regexp"
"sort" "sort"
"strings" "strings"
@@ -14,118 +13,26 @@ import (
"go.maximumdirect.net/promptkit/internal/domain" "go.maximumdirect.net/promptkit/internal/domain"
"go.maximumdirect.net/promptkit/internal/jsonvalue" "go.maximumdirect.net/promptkit/internal/jsonvalue"
"go.maximumdirect.net/promptkit/internal/llm" "go.maximumdirect.net/promptkit/internal/llm"
"go.maximumdirect.net/promptkit/internal/settings"
) )
type document struct { type collectionConfig = settings.CollectionSettings
Version *string `yaml:"version"` type backendConfig = settings.BackendSettings
Backends []backendConfig `yaml:"backends"` type profileConfig = settings.ProfileSettings
Profiles []profileConfig `yaml:"profiles"` type promptConfig = settings.PromptSettings
Prompts []promptConfig `yaml:"prompts"` type promptInputConfig = settings.PromptInputSettings
Collections map[string]collectionConfig `yaml:"collections"` type promptMessageConfig = settings.PromptMessageSettings
} type cacheControlConfig = settings.PromptCacheControlSettings
type promptOutputConfig = settings.PromptOutputSettings
type collectionConfig struct {
ReleaseVersion *string `yaml:"release_version"`
Backends []backendConfig `yaml:"backends"`
Profiles []profileConfig `yaml:"profiles"`
Prompts []promptConfig `yaml:"prompts"`
}
type backendConfig struct {
ID string `yaml:"id"`
Endpoint *string `yaml:"endpoint"`
APIKeyEnv *string `yaml:"api_key_env"`
ConcurrencyLimit *int `yaml:"concurrency_limit"`
QueueCapacity *int `yaml:"queue_capacity"`
ExtraParams map[string]any `yaml:"extra_params"`
SmokeTests *backendSmokeTestConfig `yaml:"smoke_tests"`
}
type backendSmokeTestConfig struct {
DefaultProfile *string `yaml:"default_profile"`
Reachability *reachabilityProbeConfig `yaml:"reachability"`
Authentication *authenticationProbeConfig `yaml:"authentication"`
}
type reachabilityProbeConfig struct {
Endpoint *string `yaml:"endpoint"`
}
type authenticationProbeConfig struct {
Kind *string `yaml:"kind"`
Endpoint *string `yaml:"endpoint"`
}
type profileConfig struct {
ID string `yaml:"id"`
Backend *string `yaml:"backend"`
Endpoint *string `yaml:"endpoint"`
Model *string `yaml:"model"`
Temperature *float64 `yaml:"temperature"`
TopP *float64 `yaml:"top_p"`
MaxTokens *int `yaml:"max_tokens"`
TimeoutSeconds *int `yaml:"timeout_seconds"`
ServiceTier *string `yaml:"service_tier"`
ReasoningEffort *string `yaml:"reasoning_effort"`
APIKeyEnv *string `yaml:"api_key_env"`
APIKeyRequired *bool `yaml:"api_key_required"`
ExtraParams map[string]any `yaml:"extra_params"`
}
type promptConfig struct {
ID string `yaml:"id"`
DefaultProfile *string `yaml:"default_profile"`
Description *string `yaml:"description"`
SessionID *string `yaml:"session_id"`
Inputs []promptInputConfig `yaml:"inputs"`
Messages []promptMessageConfig `yaml:"messages"`
Output *promptOutputConfig `yaml:"output"`
}
type promptInputConfig struct {
Name string `yaml:"name"`
Required bool `yaml:"required"`
ContentType *string `yaml:"content_type"`
Description *string `yaml:"description"`
}
type promptMessageConfig struct {
Role *string `yaml:"role"`
Content *string `yaml:"content"`
ContentFile *string `yaml:"content_file"`
CacheControl *cacheControlConfig `yaml:"cache_control"`
}
type cacheControlConfig struct {
Type *string `yaml:"type"`
TTL *string `yaml:"ttl"`
}
type promptOutputConfig struct {
Format *string `yaml:"format"`
ValidationMode *string `yaml:"validation_mode"`
SchemaPath *string `yaml:"schema_path"`
RepairAttempts int `yaml:"repair_attempts"`
}
// fieldState distinguishes omission, an explicit value, and an explicit null.
// Inherited values are explicit because ConfigKit has already expanded them.
type fieldState uint8
const (
fieldAbsent fieldState = iota
fieldValue
fieldNull
)
type loaded struct { type loaded struct {
result *configkit.Result[document] view configkit.Inspection
interpreted document diagnostics []configkit.Diagnostic
backends map[string]map[string]fieldState promptRequired map[string][]bool
profiles map[string]map[string]fieldState
promptMetadata map[string][]promptInputPresence
prompts map[string]*promptEntry prompts map[string]*promptEntry
registry *backend.Registry registry *backend.Registry
backendRecords map[string]backendConfig
profileRecords map[string]profileConfig
indexedProfiles map[string]domain.ExecutionProfile indexedProfiles map[string]domain.ExecutionProfile
backendIDs []string backendIDs []string
profileIDs []string profileIDs []string
@@ -134,6 +41,9 @@ type loaded struct {
collections []CollectionIdentity collections []CollectionIdentity
} }
// Snapshot holds configuration resources captured for engine assembly.
type Snapshot = loaded
type resourceBases struct { type resourceBases struct {
backends configkit.Path backends configkit.Path
profiles configkit.Path profiles configkit.Path
@@ -151,20 +61,20 @@ type resourceSet struct {
func rootResourceBases() resourceBases { func rootResourceBases() resourceBases {
return resourceBases{ return resourceBases{
backends: configkit.Keys("backends"), backends: configkit.Keys("promptkit", "backends"),
profiles: configkit.Keys("profiles"), profiles: configkit.Keys("promptkit", "profiles"),
prompts: configkit.Keys("prompts"), prompts: configkit.Keys("promptkit", "prompts"),
} }
} }
func resourceSets(config *document) []resourceSet { func resourceSets(config *settings.Settings) []resourceSet {
sets := make([]resourceSet, 0, len(config.Collections)+1) sets := make([]resourceSet, 0, len(config.Collections)+1)
sets = append(sets, resourceSet{ sets = append(sets, resourceSet{
bases: rootResourceBases(), backends: config.Backends, profiles: config.Profiles, prompts: config.Prompts, bases: rootResourceBases(), backends: config.Backends, profiles: config.Profiles, prompts: config.Prompts,
}) })
for _, namespace := range sortedKeys(config.Collections) { for _, namespace := range sortedKeys(config.Collections) {
collection := config.Collections[namespace] collection := config.Collections[namespace]
base := configkit.Keys("collections", namespace) base := configkit.Keys("promptkit", "collections", namespace)
sets = append(sets, resourceSet{ sets = append(sets, resourceSet{
isCollection: true, isCollection: true,
namespace: namespace, namespace: namespace,
@@ -192,7 +102,7 @@ func indexedResourceID(namespace, localKey string) string {
return namespace + "/" + localKey return namespace + "/" + localKey
} }
func indexedBackends(config *document) map[string]backendConfig { func indexedBackends(config *settings.Settings) map[string]backendConfig {
backends := make(map[string]backendConfig) backends := make(map[string]backendConfig)
for _, resources := range resourceSets(config) { for _, resources := range resourceSets(config) {
for _, definition := range resources.backends { for _, definition := range resources.backends {
@@ -202,7 +112,7 @@ func indexedBackends(config *document) map[string]backendConfig {
return backends return backends
} }
func indexedProfiles(config *document) map[string]profileConfig { func indexedProfiles(config *settings.Settings) map[string]profileConfig {
profiles := make(map[string]profileConfig) profiles := make(map[string]profileConfig)
for _, resources := range resourceSets(config) { for _, resources := range resourceSets(config) {
for _, definition := range resources.profiles { for _, definition := range resources.profiles {
@@ -214,67 +124,61 @@ func indexedProfiles(config *document) map[string]profileConfig {
var envName = regexp.MustCompile(`^[A-Za-z_][A-Za-z0-9_]*$`) var envName = regexp.MustCompile(`^[A-Za-z_][A-Za-z0-9_]*$`)
var backendFields = []string{"endpoint", "api_key_env", "concurrency_limit", "queue_capacity", "extra_params", "smoke_tests"} // SequenceIdentities returns the component's native sequence registrations.
var profileFields = []string{"backend", "endpoint", "model", "temperature", "top_p", "max_tokens", "timeout_seconds", "service_tier", "reasoning_effort", "api_key_env", "api_key_required", "extra_params"} func SequenceIdentities() []configkit.SequenceIdentity {
return []configkit.SequenceIdentity{
// Load captures one immutable effective configuration for internal consumers. configkit.SequenceIdentityAt(configkit.Keys("promptkit", "backends"), "id"),
func Load(main configkit.Input, underlays []configkit.Underlay) (*loaded, error) { configkit.SequenceIdentityAt(configkit.Keys("promptkit", "profiles"), "id"),
return load(main, underlays) configkit.SequenceIdentityAt(configkit.Keys("promptkit", "prompts"), "id"),
configkit.SequenceIdentityMatching(configkit.SequenceSelector{configkit.SequenceLiteralKey("promptkit"), configkit.SequenceLiteralKey("collections"), configkit.SequenceAnyMappingKey(), configkit.SequenceLiteralKey("backends")}, "id"),
configkit.SequenceIdentityMatching(configkit.SequenceSelector{configkit.SequenceLiteralKey("promptkit"), configkit.SequenceLiteralKey("collections"), configkit.SequenceAnyMappingKey(), configkit.SequenceLiteralKey("profiles")}, "id"),
configkit.SequenceIdentityMatching(configkit.SequenceSelector{configkit.SequenceLiteralKey("promptkit"), configkit.SequenceLiteralKey("collections"), configkit.SequenceAnyMappingKey(), configkit.SequenceLiteralKey("prompts")}, "id"),
}
} }
func configurationDefinition(state *loaded) configkit.Definition[document] { // Resolve normalizes the selected effective settings in place.
return configkit.Definition[document]{ func Resolve(config *settings.Settings, resolution *configkit.Resolution) {
SequenceIdentities: []configkit.SequenceIdentity{ if config == nil {
configkit.SequenceIdentityAt(configkit.Keys("backends"), "id"), return
configkit.SequenceIdentityAt(configkit.Keys("profiles"), "id"), }
configkit.SequenceIdentityAt(configkit.Keys("prompts"), "id"), for _, resources := range resourceSets(config) {
configkit.SequenceIdentityMatching(configkit.SequenceSelector{configkit.SequenceLiteralKey("collections"), configkit.SequenceAnyMappingKey(), configkit.SequenceLiteralKey("backends")}, "id"), resolveBackends(resources.backends, resources.bases.backends, resolution)
configkit.SequenceIdentityMatching(configkit.SequenceSelector{configkit.SequenceLiteralKey("collections"), configkit.SequenceAnyMappingKey(), configkit.SequenceLiteralKey("profiles")}, "id"), resolveProfiles(resources.profiles, resources.bases.profiles, resolution)
configkit.SequenceIdentityMatching(configkit.SequenceSelector{configkit.SequenceLiteralKey("collections"), configkit.SequenceAnyMappingKey(), configkit.SequenceLiteralKey("prompts")}, "id"), resolvePrompts(resources.prompts, resources.bases.prompts, resolution)
}, }
Version: &configkit.VersionRule{ }
Path: configkit.Keys("version"),
Check: checkConfigurationVersion, // Validate checks the selected effective settings.
}, func Validate(config *settings.Settings, validation *configkit.Validation) {
Sensitive: secretProviderPath, if config != nil {
Resolve: func(config *document, resolution *configkit.Resolution) error { validateEffective(config, validation)
for _, resources := range resourceSets(config) { }
captureFields(resources.backends, func(v backendConfig) string { return v.ID }, resources.bases.backends, resources.namespace, backendFields, resolution, state.backends) }
captureFields(resources.profiles, func(v profileConfig) string { return v.ID }, resources.bases.profiles, resources.namespace, profileFields, resolution, state.profiles)
capturePromptMetadata(resources.prompts, resources.bases.prompts, resources.namespace, resolution, state.promptMetadata) // Capture builds owned runtime resources from effective settings and native evidence.
func Capture(input *settings.Settings, view configkit.Inspection, diagnostics []configkit.Diagnostic) (*loaded, error) {
config := ownSettings(input)
state := &loaded{view: view, diagnostics: ownDiagnostics(diagnostics), promptRequired: make(map[string][]bool)}
metadata := view.Metadata()
for _, resources := range resourceSets(&config) {
for index, entry := range resources.prompts {
presence := make([]bool, len(entry.Inputs))
for i := range entry.Inputs {
path := resources.bases.prompts.Index(index).Key("inputs").Index(i)
// Typed values capture defaults; explicitness preserves source false.
// IsExplicit avoids projecting inheritance origins for a flag.
presence[i] = entry.Inputs[i].Required || metadata.IsExplicit(path.Key("required"))
} }
return nil state.promptRequired[indexedResourceID(resources.namespace, entry.ID)] = presence
}, }
Validate: func(config *document, validation *configkit.Validation) {
state.interpreted = interpretConfiguration(config, validation)
validateEffective(&state.interpreted, validation)
},
} }
} if err := state.materializePrompts(&config, metadata); err != nil {
// load performs one ConfigKit lifecycle over the selected source tiers.
func load(main configkit.Input, underlays []configkit.Underlay) (*loaded, error) {
state := &loaded{
backends: make(map[string]map[string]fieldState),
profiles: make(map[string]map[string]fieldState),
promptMetadata: make(map[string][]promptInputPresence),
}
result, err := configkit.LoadLayered(main, configkit.LayeredOptions[document]{
Underlays: underlays,
SourceVersionMode: configkit.SourceVersionModeIfPresent,
Definition: configurationDefinition(state),
})
if err != nil {
return nil, fmt.Errorf("load configuration: %w", err) return nil, fmt.Errorf("load configuration: %w", err)
} }
state.result = result if err := state.buildExecutionProfiles(&config); err != nil {
if err := state.materializePrompts(); err != nil {
return nil, fmt.Errorf("load configuration: %w", err) return nil, fmt.Errorf("load configuration: %w", err)
} }
if err := state.buildExecutionProfiles(); err != nil { state.backendIDs = sortedKeys(indexedBackends(&config))
return nil, fmt.Errorf("load configuration: %w", err)
}
state.backendIDs = sortedKeys(indexedBackends(&state.interpreted))
state.profileIDs = sortedKeys(state.indexedProfiles) state.profileIDs = sortedKeys(state.indexedProfiles)
for _, id := range sortedKeys(state.prompts) { for _, id := range sortedKeys(state.prompts) {
state.promptIDs = append(state.promptIDs, domain.PromptIdentity{ID: id}) state.promptIDs = append(state.promptIDs, domain.PromptIdentity{ID: id})
@@ -284,8 +188,8 @@ func load(main configkit.Input, underlays []configkit.Underlay) (*loaded, error)
"profile": resourceNamespaces(state.profileIDs), "profile": resourceNamespaces(state.profileIDs),
"prompt": promptNamespaces(state.promptIDs), "prompt": promptNamespaces(state.promptIDs),
} }
for _, namespace := range sortedKeys(result.Config.Collections) { for _, namespace := range sortedKeys(config.Collections) {
collection := result.Config.Collections[namespace] collection := config.Collections[namespace]
state.collections = append(state.collections, CollectionIdentity{ state.collections = append(state.collections, CollectionIdentity{
Namespace: namespace, ReleaseVersion: stringPointerValue(collection.ReleaseVersion), Namespace: namespace, ReleaseVersion: stringPointerValue(collection.ReleaseVersion),
}) })
@@ -293,6 +197,18 @@ func load(main configkit.Input, underlays []configkit.Underlay) (*loaded, error)
return state, nil return state, nil
} }
func ownDiagnostics(values []configkit.Diagnostic) []configkit.Diagnostic {
if values == nil {
return nil
}
result := append([]configkit.Diagnostic(nil), values...)
for i := range result {
result[i].Primary = copyPointer(result[i].Primary)
result[i].Related = append([]configkit.Location(nil), result[i].Related...)
}
return result
}
func resourceNamespaces(ids []string) []string { func resourceNamespaces(ids []string) []string {
namespaces := make(map[string]struct{}) namespaces := make(map[string]struct{})
for _, id := range ids { for _, id := range ids {
@@ -311,64 +227,26 @@ func promptNamespaces(prompts []domain.PromptIdentity) []string {
return resourceNamespaces(ids) return resourceNamespaces(ids)
} }
func checkConfigurationVersion(version configkit.Version, validation *configkit.Validation) error { // eligibleResourceList leaves invalid identities to the existing validation
if !version.Present { // boundaries, without mutating a list that cannot be recorded by identity.
return nil func eligibleResourceList[T any](entries []T, identity func(T) string) bool {
} seen := make(map[string]struct{}, len(entries))
if version.Null || version.Text != "2" { for _, entry := range entries {
validation.Error(configkit.Keys("version"), "unsupported configuration specification version") id := identity(entry)
} if !domain.IsValidLocalResourceKey(id) {
return nil return false
}
type promptInputPresence struct {
contentType fieldState
description fieldState
}
func capturePromptMetadata(prompts []promptConfig, basePath configkit.Path, namespace string, resolution *configkit.Resolution, destination map[string][]promptInputPresence) {
for promptIndex, prompt := range prompts {
inputs := prompt.Inputs
states := make([]promptInputPresence, len(inputs))
base := basePath.Index(promptIndex).Key("inputs")
for index := range inputs {
path := base.Index(index)
states[index] = promptInputPresence{
contentType: captureFieldState(path.Key("content_type"), resolution),
description: captureFieldState(path.Key("description"), resolution),
}
} }
destination[indexedResourceID(namespace, prompt.ID)] = states if _, exists := seen[id]; exists {
} return false
}
func captureFieldState(path configkit.Path, resolution *configkit.Resolution) fieldState {
info, exists := resolution.EntryInfo(path)
switch {
case !exists || !info.Explicit:
return fieldAbsent
case info.Null:
return fieldNull
default:
return fieldValue
}
}
func captureFields[T any](entries []T, identity func(T) string, basePath configkit.Path, namespace string, fields []string, resolution *configkit.Resolution, destination map[string]map[string]fieldState) {
for index, entry := range entries {
states := make(map[string]fieldState, len(fields))
for _, field := range fields {
info, exists := resolution.EntryInfo(basePath.Index(index).Key(field))
switch {
case !exists || !info.Explicit:
states[field] = fieldAbsent
case info.Null:
states[field] = fieldNull
default:
states[field] = fieldValue
}
} }
destination[indexedResourceID(namespace, identity(entry))] = states seen[id] = struct{}{}
}
return true
}
func recordStringNormalization(value **string, path configkit.Path, resolution *configkit.Resolution, normalized string) {
if *value != nil && **value != normalized {
configkit.Normalize(resolution, path, value, &normalized, "Normalize configured value")
} }
} }
@@ -381,119 +259,12 @@ func sortedKeys[T any](entries map[string]T) []string {
return keys return keys
} }
func interpretConfiguration(config *document, validation *configkit.Validation) document { func validateEffective(config *settings.Settings, validation *configkit.Validation) {
interpreted := *config
interpreted.Backends = interpretBackends(config.Backends, configkit.Keys("backends"), validation)
interpreted.Profiles = interpretProfiles(config.Profiles, configkit.Keys("profiles"), validation)
interpreted.Prompts = interpretPrompts(config.Prompts)
if config.Collections != nil {
interpreted.Collections = make(map[string]collectionConfig, len(config.Collections))
for namespace, original := range config.Collections {
collection := original
base := configkit.Keys("collections", namespace)
collection.Backends = interpretBackends(original.Backends, base.Key("backends"), validation)
collection.Profiles = interpretProfiles(original.Profiles, base.Key("profiles"), validation)
collection.Prompts = interpretPrompts(original.Prompts)
interpreted.Collections[namespace] = collection
}
}
return interpreted
}
func interpretBackends(source []backendConfig, basePath configkit.Path, validation *configkit.Validation) []backendConfig {
if source == nil {
return nil
}
backends := append([]backendConfig{}, source...)
for index := range backends {
backend := &backends[index]
base := basePath.Index(index)
if backend.Endpoint != nil {
endpoint, err := domain.NormalizeOpenAICompatibleBaseEndpoint(*backend.Endpoint)
if err != nil {
validation.Error(base.Key("endpoint"), err.Error())
} else {
backend.Endpoint = &endpoint
}
}
if backend.APIKeyEnv != nil {
value := strings.TrimSpace(*backend.APIKeyEnv)
backend.APIKeyEnv = &value
}
if backend.SmokeTests != nil {
smokeTests := *backend.SmokeTests
backend.SmokeTests = &smokeTests
smokePath := base.Key("smoke_tests")
if smokeTests.Reachability != nil {
reachability := *smokeTests.Reachability
smokeTests.Reachability = &reachability
normalizeDiagnosticEndpoint(&reachability.Endpoint, smokePath.Key("reachability").Key("endpoint"), validation)
}
if smokeTests.Authentication != nil {
authentication := *smokeTests.Authentication
smokeTests.Authentication = &authentication
if authentication.Kind != nil {
kind := strings.ToLower(strings.TrimSpace(*authentication.Kind))
authentication.Kind = &kind
}
normalizeDiagnosticEndpoint(&authentication.Endpoint, smokePath.Key("authentication").Key("endpoint"), validation)
}
}
if backend.ConcurrencyLimit != nil && *backend.ConcurrencyLimit > 0 && backend.QueueCapacity == nil {
capacity := 1024
backend.QueueCapacity = &capacity
}
}
return backends
}
func normalizeDiagnosticEndpoint(value **string, path configkit.Path, validation *configkit.Validation) {
if value == nil || *value == nil {
return
}
normalized := strings.TrimSpace(**value)
request, err := http.NewRequest(http.MethodGet, normalized, nil)
if normalized == "" || err != nil || request.URL == nil ||
(strings.ToLower(request.URL.Scheme) != "http" && strings.ToLower(request.URL.Scheme) != "https") ||
request.URL.Hostname() == "" {
validation.Error(path, "diagnostic endpoint must be an absolute HTTP(S) URL with a host")
return
}
*value = &normalized
}
func interpretProfiles(source []profileConfig, basePath configkit.Path, validation *configkit.Validation) []profileConfig {
if source == nil {
return nil
}
profiles := append([]profileConfig{}, source...)
for index := range profiles {
profile := &profiles[index]
base := basePath.Index(index)
if profile.Endpoint != nil && *profile.Endpoint != "" {
endpoint, err := domain.NormalizeOpenAICompatibleBaseEndpoint(*profile.Endpoint)
if err != nil {
validation.Error(base.Key("endpoint"), err.Error())
} else {
profile.Endpoint = &endpoint
}
}
for _, value := range []**string{&profile.ServiceTier, &profile.ReasoningEffort, &profile.APIKeyEnv} {
if *value != nil {
normalized := strings.TrimSpace(**value)
*value = &normalized
}
}
}
return profiles
}
func validateEffective(config *document, validation *configkit.Validation) {
sets := resourceSets(config) sets := resourceSets(config)
backendIDs := make(map[string]struct{}) backendIDs := make(map[string]struct{})
for _, resources := range sets { for _, resources := range sets {
if resources.isCollection && !domain.IsValidResourceNamespace(resources.namespace) { if resources.isCollection && !domain.IsValidResourceNamespace(resources.namespace) {
validation.Error(configkit.Keys("collections", resources.namespace), "collection namespace must be nonblank and contain no slash") validation.Error(configkit.Keys("promptkit", "collections", resources.namespace), "collection namespace must be nonblank and contain no slash")
} }
for _, backend := range resources.backends { for _, backend := range resources.backends {
backendIDs[indexedResourceID(resources.namespace, backend.ID)] = struct{}{} backendIDs[indexedResourceID(resources.namespace, backend.ID)] = struct{}{}
@@ -501,95 +272,110 @@ func validateEffective(config *document, validation *configkit.Validation) {
} }
for _, resources := range sets { for _, resources := range sets {
for index, backend := range resources.backends { for index, backend := range resources.backends {
base := resources.bases.backends.Index(index) validateBackendRecord(backend, resources.bases.backends.Index(index), validation)
if !domain.IsValidLocalResourceKey(backend.ID) {
validation.Error(base.Key("id"), "backend ID must be a nonblank local key without a slash")
}
if backend.Endpoint == nil || *backend.Endpoint == "" {
validation.Error(base.Key("endpoint"), "endpoint is required")
}
if backend.APIKeyEnv != nil && *backend.APIKeyEnv != "" && !envName.MatchString(*backend.APIKeyEnv) {
validation.Error(base.Key("api_key_env"), "environment variable name is invalid")
}
if backend.ConcurrencyLimit != nil && *backend.ConcurrencyLimit < 0 {
validation.Error(base.Key("concurrency_limit"), "concurrency limit must not be negative")
}
if backend.QueueCapacity != nil {
if *backend.QueueCapacity < 0 {
validation.Error(base.Key("queue_capacity"), "queue capacity must not be negative")
}
if backend.ConcurrencyLimit == nil || *backend.ConcurrencyLimit <= 0 {
validation.Error(base.Key("queue_capacity"), "queue capacity requires a positive concurrency limit")
} else if *backend.QueueCapacity > int(^uint(0)>>1)-*backend.ConcurrencyLimit {
validation.Error(base.Key("queue_capacity"), "total capacity overflows int")
}
}
validateExtraParams(backend.ExtraParams, base.Key("extra_params"), true, validation)
if smokeTests := backend.SmokeTests; smokeTests != nil {
smokePath := base.Key("smoke_tests")
if smokeTests.DefaultProfile != nil {
if strings.TrimSpace(*smokeTests.DefaultProfile) == "" {
validation.Error(smokePath.Key("default_profile"), "default profile must not be blank")
} else if !domain.IsValidResourceID(*smokeTests.DefaultProfile) {
validation.Error(smokePath.Key("default_profile"), "default profile reference is malformed")
}
}
if authentication := smokeTests.Authentication; authentication != nil {
kindPath := smokePath.Key("authentication").Key("kind")
if authentication.Kind == nil {
validation.Error(kindPath, "authentication kind is required")
} else if *authentication.Kind != "none" && *authentication.Kind != "openrouter_key" {
validation.Error(kindPath, "authentication kind must be none or openrouter_key")
}
}
}
} }
for index, profile := range resources.profiles { for index, profile := range resources.profiles {
base := resources.bases.profiles.Index(index) validateProfileRecord(profile, resources.bases.profiles.Index(index), backendIDs, validation)
if !domain.IsValidLocalResourceKey(profile.ID) {
validation.Error(base.Key("id"), "profile ID must be a nonblank local key without a slash")
}
if profile.Backend != nil && *profile.Backend != "" {
if !domain.IsValidResourceID(*profile.Backend) {
validation.Error(base.Key("backend"), "backend reference is malformed")
} else if _, exists := backendIDs[*profile.Backend]; !exists {
validation.Error(base.Key("backend"), "backend is not defined")
}
}
if profile.Model != nil && *profile.Model != "" && strings.TrimSpace(*profile.Model) == "" {
validation.Error(base.Key("model"), "model must not be blank")
}
if profile.Temperature != nil {
if err := domain.ValidateExecutionTargetSettings(domain.ExecutionTarget{Temperature: *profile.Temperature}); err != nil {
validation.Error(base.Key("temperature"), err.Error())
}
}
if profile.TopP != nil {
if err := domain.ValidateExecutionTargetSettings(domain.ExecutionTarget{TopP: *profile.TopP}); err != nil {
validation.Error(base.Key("top_p"), err.Error())
}
}
if profile.MaxTokens != nil {
if err := domain.ValidateExecutionTargetSettings(domain.ExecutionTarget{MaxTokens: *profile.MaxTokens}); err != nil {
validation.Error(base.Key("max_tokens"), err.Error())
}
}
if profile.TimeoutSeconds != nil {
if err := domain.ValidateExecutionTargetSettings(domain.ExecutionTarget{TimeoutSeconds: *profile.TimeoutSeconds}); err != nil {
validation.Error(base.Key("timeout_seconds"), err.Error())
}
}
validateExtraParams(profile.ExtraParams, base.Key("extra_params"), false, validation)
} }
for index, prompt := range resources.prompts { for index, prompt := range resources.prompts {
base := resources.bases.prompts.Index(index) validatePromptRecord(prompt, resources.bases.prompts.Index(index), validation)
if !domain.IsValidLocalResourceKey(prompt.ID) { }
validation.Error(base.Key("id"), "prompt ID must be a nonblank local key without a slash") }
} }
if prompt.DefaultProfile != nil && *prompt.DefaultProfile != "" && !domain.IsValidResourceID(*prompt.DefaultProfile) {
validation.Error(base.Key("default_profile"), "default profile reference is malformed") func validateBackendRecord(backend backendConfig, base configkit.Path, validation *configkit.Validation) {
validateBackendEndpoints(backend, base, validation)
if !domain.IsValidLocalResourceKey(backend.ID) {
validation.Error(base.Key("id"), "backend ID must be a nonblank local key without a slash")
}
if backend.Endpoint == nil || *backend.Endpoint == "" {
validation.Error(base.Key("endpoint"), "endpoint is required")
}
if backend.APIKeyEnv != nil && *backend.APIKeyEnv != "" && !envName.MatchString(*backend.APIKeyEnv) {
validation.Error(base.Key("api_key_env"), "environment variable name is invalid")
}
if backend.ConcurrencyLimit != nil && *backend.ConcurrencyLimit < 0 {
validation.Error(base.Key("concurrency_limit"), "concurrency limit must not be negative")
}
if backend.QueueCapacity != nil {
if *backend.QueueCapacity < 0 {
validation.Error(base.Key("queue_capacity"), "queue capacity must not be negative")
}
if backend.ConcurrencyLimit == nil || *backend.ConcurrencyLimit <= 0 {
validation.Error(base.Key("queue_capacity"), "queue capacity requires a positive concurrency limit")
} else if *backend.QueueCapacity > int(^uint(0)>>1)-*backend.ConcurrencyLimit {
validation.Error(base.Key("queue_capacity"), "total capacity overflows int")
}
}
validateExtraParams(backend.ExtraParams, base.Key("extra_params"), true, validation)
if smokeTests := backend.SmokeTests; smokeTests != nil {
smokePath := base.Key("smoke_tests")
if smokeTests.DefaultProfile != nil {
if strings.TrimSpace(*smokeTests.DefaultProfile) == "" {
validation.Error(smokePath.Key("default_profile"), "default profile must not be blank")
} else if !domain.IsValidResourceID(*smokeTests.DefaultProfile) {
validation.Error(smokePath.Key("default_profile"), "default profile reference is malformed")
} }
} }
if authentication := smokeTests.Authentication; authentication != nil {
kindPath := smokePath.Key("authentication").Key("kind")
if authentication.Kind == nil {
validation.Error(kindPath, "authentication kind is required")
} else if *authentication.Kind != "none" && *authentication.Kind != "openrouter_key" {
validation.Error(kindPath, "authentication kind must be none or openrouter_key")
}
}
}
}
func validateProfileRecord(profile profileConfig, base configkit.Path, backendIDs map[string]struct{}, validation *configkit.Validation) {
if profile.Endpoint != nil && *profile.Endpoint != "" {
if _, err := domain.NormalizeOpenAICompatibleBaseEndpoint(*profile.Endpoint); err != nil {
validation.Error(base.Key("endpoint"), err.Error())
}
}
if !domain.IsValidLocalResourceKey(profile.ID) {
validation.Error(base.Key("id"), "profile ID must be a nonblank local key without a slash")
}
if profile.Backend != nil && *profile.Backend != "" {
if !domain.IsValidResourceID(*profile.Backend) {
validation.Error(base.Key("backend"), "backend reference is malformed")
} else if _, exists := backendIDs[*profile.Backend]; !exists {
validation.Error(base.Key("backend"), "backend is not defined")
}
}
if profile.Model != nil && *profile.Model != "" && strings.TrimSpace(*profile.Model) == "" {
validation.Error(base.Key("model"), "model must not be blank")
}
if profile.Temperature != nil {
if err := domain.ValidateExecutionTargetSettings(domain.ExecutionTarget{Temperature: *profile.Temperature}); err != nil {
validation.Error(base.Key("temperature"), err.Error())
}
}
if profile.TopP != nil {
if err := domain.ValidateExecutionTargetSettings(domain.ExecutionTarget{TopP: *profile.TopP}); err != nil {
validation.Error(base.Key("top_p"), err.Error())
}
}
if profile.MaxTokens != nil {
if err := domain.ValidateExecutionTargetSettings(domain.ExecutionTarget{MaxTokens: *profile.MaxTokens}); err != nil {
validation.Error(base.Key("max_tokens"), err.Error())
}
}
if profile.TimeoutSeconds != nil {
if err := domain.ValidateExecutionTargetSettings(domain.ExecutionTarget{TimeoutSeconds: *profile.TimeoutSeconds}); err != nil {
validation.Error(base.Key("timeout_seconds"), err.Error())
}
}
validateExtraParams(profile.ExtraParams, base.Key("extra_params"), false, validation)
}
func validatePromptRecord(prompt promptConfig, base configkit.Path, validation *configkit.Validation) {
if !domain.IsValidLocalResourceKey(prompt.ID) {
validation.Error(base.Key("id"), "prompt ID must be a nonblank local key without a slash")
}
if prompt.DefaultProfile != nil && *prompt.DefaultProfile != "" && !domain.IsValidResourceID(*prompt.DefaultProfile) {
validation.Error(base.Key("default_profile"), "default profile reference is malformed")
} }
} }

View File

@@ -28,7 +28,10 @@ func TestLoadIdentifiedResourceLists(t *testing.T) {
if err != nil { if err != nil {
t.Fatalf("load identified resources with header %t: %v", header, err) t.Fatalf("load identified resources with header %t: %v", header, err)
} }
if loaded.result.Config.Backends[0].ID != "local" || loaded.result.Config.Collections["team"].Backends[0].ID != "local" { if _, rootErr := loaded.GetBackend("local"); rootErr != nil {
t.Fatal(rootErr)
}
if _, collectionErr := loaded.GetBackend("team/local"); collectionErr != nil {
t.Fatal("identified root and collection resources were not retained") t.Fatal("identified root and collection resources were not retained")
} }
} }
@@ -153,13 +156,58 @@ func TestIdentifiedResourcesRejectDuplicateStrictPeers(t *testing.T) {
} }
} }
func TestInvalidIdentitiesWithNormalizableFieldsAreRejected(t *testing.T) {
for _, kind := range []string{"backends", "profiles", "prompts"} {
for _, id := range []string{" ", "bad/id"} {
t.Run(kind+"/"+id, func(t *testing.T) {
record := map[string]any{"id": id}
switch kind {
case "backends":
record["endpoint"] = " HTTPS://example.invalid/v1 "
record["concurrency_limit"] = 1
case "profiles":
record["service_tier"] = " standard "
case "prompts":
record["description"] = " padded "
}
_, err := load(mapping("invalid", map[string]any{kind: []any{record}}), nil)
if err == nil || strings.Contains(strings.ToLower(err.Error()), "unrecorded") {
t.Fatalf("invalid identity error = %v", err)
}
})
}
}
}
func TestDuplicateIdentitiesWithNormalizableFieldsAreRejected(t *testing.T) {
for _, kind := range []string{"backends", "profiles", "prompts"} {
t.Run(kind, func(t *testing.T) {
record := map[string]any{"id": "same"}
switch kind {
case "backends":
record["endpoint"] = " HTTPS://example.invalid/v1 "
record["concurrency_limit"] = 1
case "profiles":
record["service_tier"] = " standard "
case "prompts":
record["description"] = " padded "
}
_, err := load(mapping("duplicates", map[string]any{kind: []any{record, record}}), nil)
if err == nil || strings.Contains(strings.ToLower(err.Error()), "unrecorded") {
t.Fatalf("duplicate identity error = %v", err)
}
})
}
}
func TestFinalIdentitiesAreCheckedAfterApplicationResolution(t *testing.T) { func TestFinalIdentitiesAreCheckedAfterApplicationResolution(t *testing.T) {
for _, kind := range []string{"backends", "profiles", "prompts"} { for _, kind := range []string{"backends", "profiles", "prompts"} {
for _, namespace := range []string{"", "*"} { for _, namespace := range []string{"", "*"} {
t.Run(kind+"/"+namespace, func(t *testing.T) { t.Run(kind+"/"+namespace, func(t *testing.T) {
definition := configurationDefinition(&loaded{}) definition := testDefinition()
definition.Resolve = func(config *document, resolution *configkit.Resolution) error { definition.Resolve = func(root *testDocument, resolution *configkit.Resolution) error {
path := configkit.Keys(kind) config := &root.PromptKit
path := configkit.Keys("promptkit", kind)
if namespace == "" { if namespace == "" {
switch kind { switch kind {
case "backends": case "backends":
@@ -180,7 +228,7 @@ func TestFinalIdentitiesAreCheckedAfterApplicationResolution(t *testing.T) {
collection.Prompts = []promptConfig{{ID: "same"}, {ID: "same"}} collection.Prompts = []promptConfig{{ID: "same"}, {ID: "same"}}
} }
config.Collections = map[string]collectionConfig{namespace: collection} config.Collections = map[string]collectionConfig{namespace: collection}
path = configkit.Keys("collections", namespace, kind) path = configkit.Keys("promptkit", "collections", namespace, kind)
} }
resolution.RecordNormalization(path, "supply resolved records") resolution.RecordNormalization(path, "supply resolved records")
return nil return nil
@@ -191,7 +239,8 @@ func TestFinalIdentitiesAreCheckedAfterApplicationResolution(t *testing.T) {
namespace: map[string]any{kind: []any{}}, namespace: map[string]any{kind: []any{}},
}}) }})
} }
_, err := configkit.LoadLayered(input, configkit.LayeredOptions[document]{Definition: definition}) input.Destination = configkit.Keys("promptkit")
_, err := configkit.LoadLayered(input, configkit.LayeredOptions[testDocument]{Definition: definition})
var loadErr *configkit.LoadError var loadErr *configkit.LoadError
if !errors.As(err, &loadErr) { if !errors.As(err, &loadErr) {
t.Fatalf("load error = %v, want a final identity diagnostic", err) t.Fatalf("load error = %v, want a final identity diagnostic", err)
@@ -228,14 +277,14 @@ func TestIdentifiedResourceLayerOperations(t *testing.T) {
values := map[string]any{"profiles": test.records} values := map[string]any{"profiles": test.records}
if test.mode != "" { if test.mode != "" {
values["config"] = map[string]any{"sequences": []any{map[string]any{ values["config"] = map[string]any{"sequences": []any{map[string]any{
"destination": []any{"profiles"}, "mode": test.mode, "destination": []any{"promptkit", "profiles"}, "mode": test.mode,
}}} }}}
} }
state, err := load(mapping("overlay profiles", values), []configkit.Underlay{{Inputs: []configkit.Input{base}}}) state, err := load(mapping("overlay profiles", values), []configkit.Underlay{{Inputs: []configkit.Input{base}}})
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
profiles := indexedProfiles(&state.result.Config) profiles := state.profileRecords
if len(profiles) != len(test.models) { if len(profiles) != len(test.models) {
t.Fatalf("profiles = %+v, want IDs and models %+v", profiles, test.models) t.Fatalf("profiles = %+v, want IDs and models %+v", profiles, test.models)
} }
@@ -271,7 +320,7 @@ func TestIdentifiedResourceLayersRejectDisplacedParentAndDuplicateAppend(t *test
{ {
name: "append cannot duplicate an identity", name: "append cannot duplicate an identity",
values: map[string]any{ values: map[string]any{
"config": map[string]any{"sequences": []any{map[string]any{"destination": []any{"profiles"}, "mode": "append"}}}, "config": map[string]any{"sequences": []any{map[string]any{"destination": []any{"promptkit", "profiles"}, "mode": "append"}}},
"profiles": []any{map[string]any{"id": "shared", "model": "other"}}, "profiles": []any{map[string]any{"id": "shared", "model": "other"}},
}, },
code: configkit.CodeSequenceConflict, code: configkit.CodeSequenceConflict,
@@ -293,62 +342,11 @@ func TestIdentifiedResourceLayersRejectDisplacedParentAndDuplicateAppend(t *test
} }
} }
func TestLoadChecksOptionalConfigurationVersion(t *testing.T) { func TestLoadRejectsLegacyResourceMappings(t *testing.T) {
for _, test := range []struct {
name string
version any
include bool
valid bool
}{
{name: "string two", version: "2", include: true, valid: true},
{name: "integer two", version: int64(2), include: true, valid: true},
{name: "omitted", valid: true},
{name: "explicit null", include: true},
{name: "unsupported string", version: "1", include: true},
} {
t.Run(test.name, func(t *testing.T) {
values := map[string]any{}
if test.include {
values["version"] = test.version
}
loaded, err := load(mapping("main", values), nil)
if test.valid {
if err != nil {
t.Fatalf("load accepted version: %v", err)
}
if test.include && (loaded.result.Config.Version == nil || *loaded.result.Config.Version != "2") {
t.Fatalf("typed version = %v, want 2", loaded.result.Config.Version)
}
if !test.include && loaded.result.Config.Version != nil {
t.Fatalf("omitted version = %q, want nil", *loaded.result.Config.Version)
}
return
}
var loadErr *configkit.LoadError
if err == nil || !errors.As(err, &loadErr) {
t.Fatalf("load error = %v, want ConfigKit load error", err)
}
found := false
for _, diagnostic := range loadErr.Diagnostics() {
if diagnostic.Path.String() == "/version" && diagnostic.Message == "unsupported configuration specification version" {
found = true
}
}
if !found {
t.Fatalf("load diagnostics = %+v, want safe version diagnostic", loadErr.Diagnostics())
}
})
}
}
func TestLoadRejectsMalformedHeadersAndLegacyResourceMappings(t *testing.T) {
for name, values := range map[string]map[string]any{ for name, values := range map[string]map[string]any{
"boolean header": {"version": true}, "backend mapping": {"backends": map[string]any{"local": map[string]any{"endpoint": "https://example.invalid/v1"}}},
"fractional header": {"version": 2.5}, "profile mapping": {"profiles": map[string]any{"local": map[string]any{}}},
"sequence header": {"version": []any{2}}, "prompt mapping": {"prompts": map[string]any{"local": map[string]any{}}},
"backend mapping": {"backends": map[string]any{"local": map[string]any{"endpoint": "https://example.invalid/v1"}}},
"profile mapping": {"profiles": map[string]any{"local": map[string]any{}}},
"prompt mapping": {"prompts": map[string]any{"local": map[string]any{}}},
} { } {
t.Run(name, func(t *testing.T) { t.Run(name, func(t *testing.T) {
_, err := load(mapping(name, values), nil) _, err := load(mapping(name, values), nil)
@@ -376,23 +374,10 @@ func TestNestedConfigRemainsApplicationData(t *testing.T) {
} }
} }
func TestLoadConsumesAcceptedUnderlayVersionAndValidatesHeaderlessPayload(t *testing.T) { func TestLoadValidatesHeaderlessPromptFragment(t *testing.T) {
underlay := mapping("declared underlay", map[string]any{ invalid := mapping("prompt fragment", map[string]any{"prompts": []any{map[string]any{"id": "incomplete", "description": "still validated"}}})
"version": int64(2), if _, err := load(configkit.Input{Source: configkit.NoSource()}, []configkit.Underlay{{Inputs: []configkit.Input{invalid}}}); err == nil || !strings.Contains(err.Error(), "at least one message is required") {
}) t.Fatalf("invalid fragment error = %v", err)
loaded, err := load(mapping("headerless main", map[string]any{}), []configkit.Underlay{{Inputs: []configkit.Input{underlay}}})
if err != nil {
t.Fatalf("load accepted non-main declaration: %v", err)
}
if loaded.result.Config.Version != nil {
t.Fatalf("consumed non-main declaration remained in effective payload: %q", *loaded.result.Config.Version)
}
invalidFragment := mapping("headerless prompt fragment", map[string]any{
"prompts": []any{map[string]any{"id": "incomplete", "description": "still validated"}},
})
if _, err := load(configkit.Input{Source: configkit.NoSource()}, []configkit.Underlay{{Inputs: []configkit.Input{invalidFragment}}}); err == nil || !strings.Contains(err.Error(), "at least one message is required") {
t.Fatalf("headerless fragment validation error = %v, want final prompt validation", err)
} }
} }
@@ -440,19 +425,16 @@ func TestLoadComposesLiteralKeysAndEffectivePresence(t *testing.T) {
} }
t.Fatal(err) t.Fatal(err)
} }
profile := loaded.result.Config.Collections["base"].Profiles[1] profile := loaded.profileRecords["base/child.v1"]
if profile.Temperature == nil || *profile.Temperature != 0 || profile.TopP != nil || profile.APIKeyRequired == nil || *profile.APIKeyRequired { if profile.Temperature == nil || *profile.Temperature != 0 || profile.TopP != nil || profile.APIKeyRequired == nil || *profile.APIKeyRequired {
t.Fatalf("effective values were not retained: %+v", profile) t.Fatalf("effective values were not retained: %+v", profile)
} }
if loaded.profiles["base/child.v1"]["temperature"] != fieldValue || loaded.profiles["base/child.v1"]["top_p"] != fieldNull || loaded.profiles["base/child.v1"]["api_key_required"] != fieldValue {
t.Fatalf("effective presence was lost: %+v", loaded.profiles["base/child.v1"])
}
provider := profile.ExtraParams["provider"].(map[string]any) provider := profile.ExtraParams["provider"].(map[string]any)
if provider["first"] != int64(1) || provider["second"] != int64(3) { if provider["first"] != int64(1) || provider["second"] != int64(3) {
t.Fatalf("recursive overlay failed: %+v", provider) t.Fatalf("recursive overlay failed: %+v", provider)
} }
backend := loaded.result.Config.Backends[0] backend := loaded.backendRecords["local.v1"]
if backend.QueueCapacity == nil || *backend.QueueCapacity != 0 || loaded.backends["local.v1"]["queue_capacity"] != fieldValue { if backend.QueueCapacity == nil || *backend.QueueCapacity != 0 {
t.Fatalf("explicit zero queue capacity was lost: %+v", backend) t.Fatalf("explicit zero queue capacity was lost: %+v", backend)
} }
if _, err := loaded.registry.GetBackend("base/local.v1"); err != nil { if _, err := loaded.registry.GetBackend("base/local.v1"); err != nil {
@@ -462,13 +444,11 @@ func TestLoadComposesLiteralKeysAndEffectivePresence(t *testing.T) {
if !exists || collectionProfile.ID != "base/namespaced" || collectionProfile.ResolvedTarget == nil || collectionProfile.ResolvedTarget.BackendID != "base/local.v1" { if !exists || collectionProfile.ID != "base/namespaced" || collectionProfile.ResolvedTarget == nil || collectionProfile.ResolvedTarget.BackendID != "base/local.v1" {
t.Fatalf("collection profile snapshot = %+v, exists %t", collectionProfile, exists) t.Fatalf("collection profile snapshot = %+v, exists %t", collectionProfile, exists)
} }
if loaded.result.Config.Collections["invoice"].Prompts[0].Messages[0].Content == nil { if loaded.prompts["invoice/summary"].definition.Templates[0].Content != "Summarize." {
t.Fatal("literal prompt content was not decoded") t.Fatal("literal prompt content was not decoded")
} }
if loaded.result.Config.Version == nil || *loaded.result.Config.Version != "2" || if len(loaded.collections) != 2 || loaded.collections[0].ReleaseVersion == nil || *loaded.collections[0].ReleaseVersion != "" || loaded.collections[1].ReleaseVersion == nil || *loaded.collections[1].ReleaseVersion != "r7" {
loaded.result.Config.Collections["base"].ReleaseVersion == nil || *loaded.result.Config.Collections["base"].ReleaseVersion != "" || t.Fatalf("collection version metadata was not retained: %+v", loaded.collections)
loaded.result.Config.Collections["invoice"].ReleaseVersion == nil || *loaded.result.Config.Collections["invoice"].ReleaseVersion != "r7" {
t.Fatalf("root or collection version metadata was not retained: %+v", loaded.result.Config)
} }
if _, _, err := loaded.prompt("invoice/version"); err != nil { if _, _, err := loaded.prompt("invoice/version"); err != nil {
t.Fatalf("resource named version was not indexed under its qualified identity: %v", err) t.Fatalf("resource named version was not indexed under its qualified identity: %v", err)
@@ -533,18 +513,23 @@ func TestLoadInterpretsBackendNormalizationAndNullDefault(t *testing.T) {
backend.APIKeyEnv != "TOKEN_NAME" || backend.QueueCapacity != 1024 { backend.APIKeyEnv != "TOKEN_NAME" || backend.QueueCapacity != 1024 {
t.Fatalf("backend normalization/default was not applied: %+v", backend) t.Fatalf("backend normalization/default was not applied: %+v", backend)
} }
composed := loaded.result.Config.Backends[0] composed := loaded.backendRecords["local"]
if composed.Endpoint == nil || *composed.Endpoint != " HTTP://localhost:8080/v1 " || if composed.Endpoint == nil || *composed.Endpoint != "http://localhost:8080/v1" ||
composed.APIKeyEnv == nil || *composed.APIKeyEnv != " TOKEN_NAME " || composed.QueueCapacity != nil { composed.APIKeyEnv == nil || *composed.APIKeyEnv != "TOKEN_NAME" || composed.QueueCapacity == nil || *composed.QueueCapacity != 1024 {
t.Fatalf("composed backend was changed: %+v", composed) t.Fatalf("resolved backend was not normalized: %+v", composed)
} }
if loaded.backends["local"]["queue_capacity"] != fieldNull { field, found := loaded.inspection.Metadata().Lookup(configkit.Keys("promptkit", "backends").Index(0).Key("endpoint"))
t.Fatalf("explicit null was not retained: %+v", loaded.backends["local"]) if !found || field.Origin.Operation != configkit.ChangeNormalization {
}
field, found := loaded.result.Metadata().Lookup(configkit.Keys("backends").Index(0).Key("endpoint"))
if !found || field.Origin.Operation == configkit.ChangeNormalization || len(field.Origin.Locations) == 0 {
t.Fatalf("endpoint origin was not preserved: %+v", field) t.Fatalf("endpoint origin was not preserved: %+v", field)
} }
field, found = loaded.inspection.Metadata().Lookup(configkit.Keys("promptkit", "backends").Index(0).Key("queue_capacity"))
if !found || field.Origin.Operation != configkit.ChangeDerivedDefault {
t.Fatalf("queue capacity default was not recorded: %+v", field)
}
field, found = loaded.inspection.Metadata().Lookup(configkit.Keys("promptkit", "backends").Index(0).Key("concurrency_limit"))
if !found || field.Origin.Operation == configkit.ChangeNormalization || field.Origin.Operation == configkit.ChangeDerivedDefault {
t.Fatalf("unchanged concurrency acquired a resolution record: %+v", field)
}
} }
func TestLoadBackendSmokeTestsOmissionAndNull(t *testing.T) { func TestLoadBackendSmokeTestsOmissionAndNull(t *testing.T) {
@@ -578,7 +563,7 @@ func TestLoadBackendSmokeTestsOmissionAndNull(t *testing.T) {
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
if got := loaded.result.Config.Backends[0].SmokeTests; got != nil { if got := loaded.backendRecords["local"].SmokeTests; got != nil {
t.Fatalf("smoke_tests = %+v, want nil", got) t.Fatalf("smoke_tests = %+v, want nil", got)
} }
selected, err := loaded.registry.GetBackend("local") selected, err := loaded.registry.GetBackend("local")
@@ -651,32 +636,32 @@ func TestLoadRejectsInvalidFinalBackendSmokeTests(t *testing.T) {
{ {
name: "blank default profile", name: "blank default profile",
data: map[string]any{"default_profile": " \t "}, data: map[string]any{"default_profile": " \t "},
path: "/backends/0/smoke_tests/default_profile", path: "/promptkit/backends/0/smoke_tests/default_profile",
}, },
{ {
name: "authentication kind required", name: "authentication kind required",
data: map[string]any{"authentication": map[string]any{}}, data: map[string]any{"authentication": map[string]any{}},
path: "/backends/0/smoke_tests/authentication/kind", path: "/promptkit/backends/0/smoke_tests/authentication/kind",
}, },
{ {
name: "unknown authentication kind", name: "unknown authentication kind",
data: map[string]any{"authentication": map[string]any{"kind": "bearer"}}, data: map[string]any{"authentication": map[string]any{"kind": "bearer"}},
path: "/backends/0/smoke_tests/authentication/kind", path: "/promptkit/backends/0/smoke_tests/authentication/kind",
}, },
{ {
name: "non-http reachability URL", name: "non-http reachability URL",
data: map[string]any{"reachability": map[string]any{"endpoint": "ftp://probe.example/ready"}}, data: map[string]any{"reachability": map[string]any{"endpoint": "ftp://probe.example/ready"}},
path: "/backends/0/smoke_tests/reachability/endpoint", path: "/promptkit/backends/0/smoke_tests/reachability/endpoint",
}, },
{ {
name: "blank authentication URL", name: "blank authentication URL",
data: map[string]any{"authentication": map[string]any{"kind": "none", "endpoint": " "}}, data: map[string]any{"authentication": map[string]any{"kind": "none", "endpoint": " "}},
path: "/backends/0/smoke_tests/authentication/endpoint", path: "/promptkit/backends/0/smoke_tests/authentication/endpoint",
}, },
{ {
name: "relative authentication URL", name: "relative authentication URL",
data: map[string]any{"authentication": map[string]any{"kind": "none", "endpoint": "/key"}}, data: map[string]any{"authentication": map[string]any{"kind": "none", "endpoint": "/key"}},
path: "/backends/0/smoke_tests/authentication/endpoint", path: "/promptkit/backends/0/smoke_tests/authentication/endpoint",
}, },
} { } {
t.Run(test.name, func(t *testing.T) { t.Run(test.name, func(t *testing.T) {
@@ -711,7 +696,7 @@ func TestLoadDefaultsAbsentQueueCapacity(t *testing.T) {
t.Fatal(err) t.Fatal(err)
} }
backend, err := loaded.GetBackend("local") backend, err := loaded.GetBackend("local")
if err != nil || backend.APIKeyEnv != "" || backend.QueueCapacity != 1024 || loaded.result.Config.Backends[0].QueueCapacity != nil || loaded.backends["local"]["queue_capacity"] != fieldAbsent { if err != nil || backend.APIKeyEnv != "" || backend.QueueCapacity != 1024 || loaded.backendRecords["local"].QueueCapacity == nil || *loaded.backendRecords["local"].QueueCapacity != 1024 {
t.Fatalf("absent queue capacity did not receive its default: %+v", backend) t.Fatalf("absent queue capacity did not receive its default: %+v", backend)
} }
} }
@@ -723,13 +708,12 @@ func TestLoadCapturesProfilePresence(t *testing.T) {
if err != nil { if err != nil {
t.Fatal(err) t.Fatal(err)
} }
profile := loaded.result.Config.Profiles[0] profile := loaded.profileRecords["local"]
if profile.Temperature == nil || *profile.Temperature != 0 || profile.TopP != nil || profile.APIKeyRequired == nil || *profile.APIKeyRequired { if profile.Temperature == nil || *profile.Temperature != 0 || profile.TopP != nil || profile.APIKeyRequired == nil || *profile.APIKeyRequired {
t.Fatalf("profile values were not preserved: %+v", profile) t.Fatalf("profile values were not preserved: %+v", profile)
} }
fields := loaded.profiles["local"] if profile.Model != nil {
if fields["temperature"] != fieldValue || fields["top_p"] != fieldNull || fields["api_key_required"] != fieldValue || fields["model"] != fieldAbsent { t.Fatalf("omitted model was not preserved: %+v", profile)
t.Fatalf("profile presence was not preserved: %+v", fields)
} }
} }
@@ -754,7 +738,7 @@ func TestLoadValidatesOnlyFinalDomainValues(t *testing.T) {
if err != nil { if err != nil {
t.Fatalf("replaced lower-layer domain value should not fail: %v", err) t.Fatalf("replaced lower-layer domain value should not fail: %v", err)
} }
if loaded.result.Config.Profiles[0].Model != nil { if loaded.profileRecords["partial"].Model != nil {
t.Fatal("unused partial profile unexpectedly gained a model") t.Fatal("unused partial profile unexpectedly gained a model")
} }
@@ -765,7 +749,7 @@ func TestLoadValidatesOnlyFinalDomainValues(t *testing.T) {
if !errors.As(err, &loadError) { if !errors.As(err, &loadError) {
t.Fatalf("expected path-aware effective backend failure, got %v", err) t.Fatalf("expected path-aware effective backend failure, got %v", err)
} }
if diagnostics := loadError.Diagnostics(); len(diagnostics) != 1 || diagnostics[0].Path.String() != "/backends/0/endpoint" { if diagnostics := loadError.Diagnostics(); len(diagnostics) != 1 || diagnostics[0].Path.String() != "/promptkit/backends/0/endpoint" {
t.Fatalf("expected endpoint diagnostic, got %+v", diagnostics) t.Fatalf("expected endpoint diagnostic, got %+v", diagnostics)
} }
} }

View File

@@ -0,0 +1,145 @@
package configuration
import (
"go.maximumdirect.net/promptkit/internal/jsonvalue"
"go.maximumdirect.net/promptkit/internal/settings"
)
func copyPointer[T any](value *T) *T {
if value == nil {
return nil
}
copy := *value
return &copy
}
func copyProviderMap(value map[string]any) map[string]any {
if value == nil {
return nil
}
return jsonvalue.CloneTrusted(value).(map[string]any)
}
func ownBackendSettings(in settings.BackendSettings) settings.BackendSettings {
in.Endpoint = copyPointer(in.Endpoint)
in.APIKeyEnv = copyPointer(in.APIKeyEnv)
in.ConcurrencyLimit = copyPointer(in.ConcurrencyLimit)
in.QueueCapacity = copyPointer(in.QueueCapacity)
in.ExtraParams = copyProviderMap(in.ExtraParams)
in.Application = nil
if in.SmokeTests != nil {
smoke := *in.SmokeTests
smoke.DefaultProfile = copyPointer(smoke.DefaultProfile)
if smoke.Reachability != nil {
reach := *smoke.Reachability
reach.Endpoint = copyPointer(reach.Endpoint)
smoke.Reachability = &reach
}
if smoke.Authentication != nil {
auth := *smoke.Authentication
auth.Kind = copyPointer(auth.Kind)
auth.Endpoint = copyPointer(auth.Endpoint)
smoke.Authentication = &auth
}
in.SmokeTests = &smoke
}
return in
}
func ownProfileSettings(in settings.ProfileSettings) settings.ProfileSettings {
in.Backend = copyPointer(in.Backend)
in.Endpoint = copyPointer(in.Endpoint)
in.Model = copyPointer(in.Model)
in.Temperature = copyPointer(in.Temperature)
in.TopP = copyPointer(in.TopP)
in.MaxTokens = copyPointer(in.MaxTokens)
in.TimeoutSeconds = copyPointer(in.TimeoutSeconds)
in.ServiceTier = copyPointer(in.ServiceTier)
in.ReasoningEffort = copyPointer(in.ReasoningEffort)
in.APIKeyEnv = copyPointer(in.APIKeyEnv)
in.APIKeyRequired = copyPointer(in.APIKeyRequired)
in.ExtraParams = copyProviderMap(in.ExtraParams)
in.Application = nil
return in
}
func ownPromptSettings(in settings.PromptSettings) settings.PromptSettings {
in.DefaultProfile = copyPointer(in.DefaultProfile)
in.Description = copyPointer(in.Description)
in.SessionID = copyPointer(in.SessionID)
in.Application = nil
if in.Inputs != nil {
in.Inputs = append(make([]settings.PromptInputSettings, 0, len(in.Inputs)), in.Inputs...)
for i := range in.Inputs {
in.Inputs[i].ContentType = copyPointer(in.Inputs[i].ContentType)
in.Inputs[i].Description = copyPointer(in.Inputs[i].Description)
}
}
if in.Messages != nil {
in.Messages = append(make([]settings.PromptMessageSettings, 0, len(in.Messages)), in.Messages...)
for i := range in.Messages {
message := &in.Messages[i]
message.Role = copyPointer(message.Role)
message.Content = copyPointer(message.Content)
message.ContentFile = copyPointer(message.ContentFile)
if message.CacheControl != nil {
cache := *message.CacheControl
cache.Type = copyPointer(cache.Type)
cache.TTL = copyPointer(cache.TTL)
message.CacheControl = &cache
}
}
}
if in.Output != nil {
output := *in.Output
output.Format = copyPointer(output.Format)
output.ValidationMode = copyPointer(output.ValidationMode)
output.Schema = jsonvalue.CloneTrusted(output.Schema)
output.SchemaFile = copyPointer(output.SchemaFile)
output.SchemaPath = copyPointer(output.SchemaPath)
output.RepairAttempts = copyPointer(output.RepairAttempts)
in.Output = &output
}
return in
}
func ownSettings(in *settings.Settings) settings.Settings {
if in == nil {
return settings.Settings{}
}
out := settings.Settings{SchemaRoot: copyPointer(in.SchemaRoot)}
copyResources := func(backends []settings.BackendSettings, profiles []settings.ProfileSettings, prompts []settings.PromptSettings) ([]settings.BackendSettings, []settings.ProfileSettings, []settings.PromptSettings) {
var b []settings.BackendSettings
if backends != nil {
b = make([]settings.BackendSettings, len(backends))
for i := range backends {
b[i] = ownBackendSettings(backends[i])
}
}
var p []settings.ProfileSettings
if profiles != nil {
p = make([]settings.ProfileSettings, len(profiles))
for i := range profiles {
p[i] = ownProfileSettings(profiles[i])
}
}
var q []settings.PromptSettings
if prompts != nil {
q = make([]settings.PromptSettings, len(prompts))
for i := range prompts {
q[i] = ownPromptSettings(prompts[i])
}
}
return b, p, q
}
out.Backends, out.Profiles, out.Prompts = copyResources(in.Backends, in.Profiles, in.Prompts)
if in.Collections != nil {
out.Collections = make(map[string]settings.CollectionSettings, len(in.Collections))
for name, collection := range in.Collections {
copy := settings.CollectionSettings{ReleaseVersion: copyPointer(collection.ReleaseVersion), SchemaRoot: copyPointer(collection.SchemaRoot)}
copy.Backends, copy.Profiles, copy.Prompts = copyResources(collection.Backends, collection.Profiles, collection.Prompts)
out.Collections[name] = copy
}
}
return out
}

Some files were not shown because too many files have changed in this diff Show More