audita/docs/roadmap/commandkit.md

CommandKit Adoption Outline

Updated: 2026-11-10

Status: Adoption target and design decisions resolved; the upstream release gate is satisfied and implementation is complete. The master roadmap owns shared direction and exact dependency revisions; the implementation plan owns sequencing. See the ConfigKit and PromptKit outlines for their contracts.

Baseline and Target

Before this work, Audita used application-owned CLI parsing, inspection commands, and reporting, described in the CLI contract. Adopt CommandKit core v0.6.0 with ckconfig and ckprompt v0.7.0 as the standard operator interface. Use supported command declarations, typed inputs, supplied-value tracking, help/version, reporting, and formatting. Audita handlers retain transcript processing and artifact responsibilities.

The upstream core release, ckconfig release, ckprompt release, and consumer guide define the selected set. Core uses Cobra/pflag and declares Go 1.26.0; the matched integrations require Go 1.27. They select ConfigKit v1.2.0 and PromptKit v0.21.0. The integration updates preserve their public adapters; wire processing and all inspection commands to the shared loaders and recorded application context described in the ConfigKit and PromptKit outlines.

Adoption Gate: Released Debug Binding and Reporting Mode Access

Satisfied by core v0.6.0 and the selected v0.7.0 integrations. Upstream publication, public APIs, released integration dependencies, and the required behavior were verified in the review below.

Both required capabilities are available:

• A BoolFlag named debug attaches to the shared reporting option and supplies its parsed boolean independently of the reporting mode for artifact capture.
• Invocation.ReportingMode() exposes the resolved reporting mode for quiet transcript-destination validation.

The master dependency selection records exact published versions and source revisions. No unreleased dependency, local replacement, private-state inspection, or duplicate argument parsing is needed. Audita integration and its acceptance tests remain implementation work; the released capabilities remove the upstream availability blocker.

Agreed Operator Experience

• Install the full ckprompt inspection surface: prompt/profile/backend views, resource and namespace/collection inventories, contextual requirement inspection, and captured engine configuration through its config scope. Native reports and namespace selection must remain available; a curated Audita subset is not a substitute.
• Apply the resolved profile-summary decision: retire the curated profiles view, use standard configuration/resource inspection and request preview, and add no replacement custom role summary at this stage.
• Adopt standard ckconfig viewing, validation, sources, path discovery, explanation, and comparison. Use native evidence and common text/YAML/JSON rendering. Do not recreate removed provider-specific explanation writers.
• Use standard command names and reporting flags. CommandKit owns --debug; its parsed boolean also enables Audita artifact capture for processing runs, independently of the resolved reporting mode, as resolved below.
• Keep domain transcript serialization and ordinary report files under Audita's output contracts. Generic report formatting need not transform transcript JSON. Preserve meaningful partial success: a work-unit model error is not automatically a failed CLI invocation or a reporter warning.
• Preserve safe library redaction without restricting standard inspection to today's curated fields. Loading for help/version must not trigger credentials, provider calls, or domain processing.

Debug Reporting and Artifact Capture — Resolved

Use one --debug flag for detailed diagnostics: standard CommandKit debug reporting plus retained Audita diagnostic artifacts during processing. Do not introduce --debug-artifacts, duplicate the reserved flag, or add a second parser. Artifact capture follows the parsed --debug boolean independently of the resolved reporting mode. In particular, --debug --quiet suppresses ordinary CLI output while retaining diagnostic artifacts; --debug --verbose also retains them. CommandKit continues to own CLI visibility and mode ordering.

Use the native boolean semantics: omission disables capture, --debug enables it, --debug=false disables it, and the last debug occurrence wins. Quiet and verbose flags do not rewrite this value. Aliases refer to the same input; tokens consumed as other flag values or following -- do not count as debug flags.

Inspection commands keep standard debug reporting without creating processing artifacts. Help and migration guidance must explain that supplying --debug for processing also writes diagnostic files, including when quiet suppresses output. Report artifact locations through CommandKit's normal visibility rules; artifact creation does not require an always-visible message in quiet mode. Retain Audita's artifact redaction and lifecycle responsibilities. There is no separate capture switch in this design.

Use CommandKit's released optional BoolFlag binding named debug, with a false default, to initialize capture from its typed boolean value. The resolved-mode accessor serves destination validation and reporting decisions; capture follows the bound value independently. The upstream review below records the matching released contract.

Quiet Mode and the Completion Summary — Resolved

Adopt CommandKit's standard visibility rules. Emit the completion summary and created report/debug locations as informational diagnostics through the shared reporter, using deferred information when final ordering requires it. Show them in verbose and debug modes; suppress them in ordinary and quiet modes. Keep diagnostics separate from transcript stdout and announce completion only after processing and enabled required outputs have succeeded.

Output	Ordinary	Verbose/debug	Quiet
Transcript on stdout, when selected	Shown	Shown	Rejected before processing
Completion summary and created artifact locations	Hidden	Shown	Hidden
Requested transcript and ordinary report files	Written	Written	Written
Fatal error explanation	Shown	Shown	Shown

Debug capture follows the independent boolean decision above in every reporting mode. Quiet does not cancel processing or suppress file writes. Recoverable model-operation failures retain their domain meaning and ordinary report records; do not promote them to fatal errors or warnings merely to make them visible.

This deliberately retires the always-visible completion summary. Help and hard-cutover guidance must explain that operators wanting immediate counts and locations should select verbose/debug reporting. If ordinary reports are disabled, those modes are needed to see the completion counts. Operators wanting a transcript while quiet is effective must select an explicit destination, as resolved below.

Quiet Processing Requires a Transcript Destination — Resolved

Reject a processing invocation when its resolved reporting mode is quiet and its selected transcript destination is stdout. Require an explicit non-stdout output destination, such as --output corrected.json. Deliberate disposal remains available through --output /dev/null; do not introduce a --force bypass. An empty output value that selects stdout does not satisfy the requirement.

Audita's primary product is the corrected transcript. Reports and debug evidence do not substitute for delivering it, so their enablement does not exempt a run from this check. The domain-specific restriction prevents incurring processing costs while accidentally discarding that product. It leaves CommandKit's standard quiet semantics intact and does not apply to inspection commands.

Use CommandKit's public resolved-mode accessor and Audita's selected destination. Return a usage failure with a visible explanation before creating output or debug artifacts or starting model calls. For example: "Quiet processing requires an output destination; use --output FILE, or --output /dev/null to discard the transcript deliberately." Do not scan arguments or infer mode from output. Shell redirection does not satisfy the requirement because CommandKit still suppresses stdout transcript delivery.

Check effective mode after native ordering, boolean values, and aliases: --quiet --verbose permits stdout, whereas --verbose --quiet requires an explicit destination. --debug --quiet also requires one; debug capture remains independent once the invocation passes this check.

Output-Failure Behavior — Resolved

Require successful delivery of the transcript and enabled ordinary reports. Debug artifacts remain optional even when capture was explicitly requested. Adopt CommandKit's native handling of CLI output and diagnostic write failures.

Output	Failure behavior
Transcript file or stdout	Fail the invocation.
Enabled ordinary reports	Fail the invocation if any required report cannot be written.
Debug directory or individual debug artifacts	Continue processing and submit a redacted warning through CommandKit.
CommandKit output and diagnostic streams	Preserve the native outcome and delivery-error behavior.

Ordinary reports describe incomplete work, recoverable model failures, and repairs. When enabled, delivering that account is part of completing the run. Debug artifacts provide supplemental investigative detail; their loss alone does not invalidate a delivered transcript and reports. Do not add a debug strictness option.

Detect required-destination problems before model work wherever practical through the output preparation below. If a required output fails later, retain successfully written files, fail the invocation, and attempt to finalize the remaining reports with the failure recorded. A report failure after transcript publication must not erase that transcript or repeat model calls. Preserve the original processing failure when secondary reporting or cleanup also fails. Emit the completion summary only after processing and all enabled required outputs have succeeded, including required final writes, closes, and file replacements.

Use the existing redaction and resolved visibility rules for debug warnings. In particular, --debug --quiet still attempts capture but suppresses warnings about missing optional artifacts; required-output failures retain the normal fatal-error explanation. A separate failure while writing a warning can still make the invocation fail under CommandKit's native output-failure contract. Keep that outcome under upstream ownership rather than overriding it to make diagnostic stream failures nonfatal.

Output Resource Preparation During Preflight — Resolved

Prepare output resources as part of Audita's expressly named internal/framework/preflight phase, after ConfigKit handoff and before model calls. Follow the preflight package boundary. Complete configuration, invocation, and domain-readiness checks for the entire pipeline first, including the quiet-destination rule and active destination collision checks. Only then prepare output resources, and begin execution after required preparation succeeds. Disabled report and debug destinations remain inactive.

• For a regular transcript destination, create and open a temporary file in its destination directory before processing. Preserve the existing destination until processing and writing succeed, then replace it. This also preserves input when the transcript intentionally replaces its own source; preparation must not truncate the final path.
• Initialize enabled ordinary reports before processing, using the existing in_progress report lifecycle and per-file replacement. An initialization failure stops the invocation before model work.
• When debug capture is enabled, create its directory and initial artifacts early. Create later artifacts as needed instead of opening every possible artifact in advance. Debug preparation failures follow the nonfatal artifact rule above.
• Use CommandKit's supplied writer for transcript stdout. Defer opening pipe and device destinations until output delivery; in particular, opening a named pipe during preflight could block waiting for its reader. Do not replace these destinations with regular temporary files.

The preflight package coordinates preparation through the output-session component. Reuse and adapt that component to own prepared handles, cleanup, and final publication; keep filesystem machinery out of individual readiness checks. Close owned handles and remove unpublished temporary files on failure or cancellation, while retaining published outputs and retained diagnostic evidence. The caller retains ownership of supplied standard streams.

Opening resources early establishes readiness at that point; later writes, closes, and replacements can still fail. Apply the output-failure rules throughout execution and finalization. Preserve per-file publication without adding a transaction across the transcript and reports.

Request Preview, Validation, and Smoke Commands — Resolved

Include standard ckprompt request preview, request validation, and the complete smoke-command surface in the next major release. Enable the upstream adapters and retain native commands, flags, reports, and outcomes. Smoke commands remain explicit operator-invoked operations, separate from application startup and automated release checks.

Use the standard request inputs directly where they suffice. Where Audita adds role/profile selection, shared instructions, response schemas, or domain inputs, reuse its processing request-construction helpers in the relevant adapters. Preserve native operator overrides and keep generic requests usable without Audita processing context. Request preview and validation operate on supplied inputs; representative validator candidates can come from maintained fixtures without running proposal inference as part of request construction.

All three capabilities are required adoption scope. Address concrete integration blockers upstream rather than silently omitting standard capabilities.

For standard CommandKit commands passed through by Audita, provide a concise link to the upstream documentation for the adopted version. Upstream owns their command reference and behavior explanations. Do not duplicate that documentation, add local provider-call/cost warnings, or add caveats or workarounds for perceived limitations of the standard capabilities. Audita documentation covers only its own selectors, request construction, domain inputs, and application-specific behavior. Apply this documentation boundary to the whole standard command suite.

Integration Boundaries

The native app and PromptKit configuration scopes are distinct. Both sides of native comparison use their own selected inputs. Scope validation follows native scope behavior; process preflight checks what execution needs. The same engine-loading path and automatic source contract serve processing and all standard PromptKit commands. ReleaseKit supplies the application version; CommandKit owns standard flags, formats, help, completion, and outcomes, including native comparison status 3.

Audita supplies typed process inputs, domain request construction, preflight, and output-session coordination. Glossary readiness is part of that shared phase. Standard inspection remains independent of processing inputs. Canonical Audita documentation owns application-specific behavior and the hard cutover; standard commands link to the adopted upstream reference. Exact selected dependencies and notices belong in attribution.

Acceptance and Upstream Requests

Check the complete standard inspection/configuration inventory and native formats through the assembled CLI, including failed-load evidence, redaction, source selection, and missing/partial resources. Check help/version without configuration and supplied-value handling through domain settings. Exercise stdout transcript isolation, file output, quiet/verbose/debug behavior, completion summary placement, cancellation, partial success, and output failures.

Verify that automatic prompt replacements and their origins agree between processing, resource/configuration inspection, and request preview. Apply the shared loader's source selection to validation and smoke commands as well, using controlled provider boundaries for smoke acceptance. Inspection must retain the same source behavior without requiring processing inputs or starting preflight.

Verify that the completion summary and created artifact locations appear only in verbose/debug diagnostics, never in transcript stdout. Reject quiet processing with a stdout destination; requested file outputs remain enabled in valid quiet runs. Fatal error explanations remain visible in every mode, and a required-output failure must not produce a successful completion summary. Cover ordinary reporting disabled as well as enabled, and document the loss of the former default summary.

Cover required-output preparation failure before model calls, late write/close/ replacement failures, and report failure after a transcript has been published. Verify retention of published output, preservation of the original processing failure when reporting also fails, and attempts to finalize available reports. Exercise optional debug-directory and artifact failures, including suppressed warnings in quiet mode, and the native CommandKit outcome when a diagnostic write itself fails. Keep these checks focused on Audita's integration and output responsibilities rather than duplicating upstream reporter tests.

Verify that readiness failures occur before output resources are created. Check that preparing regular transcript output preserves existing content, including an in-place transcript, and that failure or cancellation closes owned resources and removes unpublished temporary files. Cover enabled report initialization, inactive disabled destinations, and debug artifacts created as needed. Confirm that preflight does not open pipe/device destinations and that supplied standard streams remain caller-owned.

Cover omitted and empty output selections in effective quiet mode, including with reports or debug capture enabled: reject before model calls and artifact creation. Accept valid explicit file and discard destinations. Exercise native mode ordering, explicit false, and aliases so the check follows the resolved mode. Keep quiet inspection valid and help/version free of processing setup.

Verify the independent capture and visibility decisions through CommandKit's parser for processing commands with a valid explicit transcript destination:

Arguments after the command path	Reporting mode	Capture artifacts
None	Ordinary	No
--debug	Debug	Yes
--debug --quiet	Quiet	Yes
--quiet --debug	Debug	Yes
--debug --verbose	Verbose	Yes
--debug --debug=false	Ordinary	No
--debug=false --debug	Debug	Yes
--debug --quiet --debug=false	Quiet	No

Check aliases, explicit false values, and option-looking operands or values through native parsing. Inspection must not create processing artifacts even with debug enabled. Check artifact-location reporting subject to normal visibility and help text explaining capture under quiet mode.

Verify preview, validation, and every standard smoke operation through the assembled CLI using fixtures and controlled providers. Focus on Audita's command registration, engine loading, input forwarding, shared request construction, and native report/outcome handoff. Keep integration coverage proportionate rather than duplicating upstream capability tests.

Tagged core and integration suites passed during review, but do not establish these Audita behaviors. Use offline tests and controlled providers.

CommandKit and integrations are early-stage. Freely propose upstream improvements where reporting, declaration, or integration APIs are problematic or incomplete. The reporting request is fulfilled by the released capabilities described below. The resolved summary and destination decisions use standard reporting behavior; they do not require an ordinary-mode summary channel or an exception to quiet suppression. Record further concrete integration gaps as they arise and prefer shared upstream improvements over a permanent local output layer.

Upstream Review: Shared Debug Input and Reporting Mode

On 2026-11-09, verified published upstream tags for core, ckconfig, and ckprompt v0.6.0, the then-selected release set. Reviewed the core source changes and external public-API tests. Compared with the previous reviewed integration releases, ckconfig and ckprompt change dependency pins and release notes without changing integration implementation.

The released binding contract and invocation API support the selected design. The shared binding preserves supplied-value tracking, explicit false, repeated occurrences, and aliases. Application defaults affect the bound input, not reporting-mode selection; Audita selects a false default. Invocation.ReportingMode() reports the selection independently of delivery success and remains available after finalization. --debug --quiet therefore retains a true capture input while the reporting mode is quiet.

Using Go 1.27.3, all three tagged module suites passed with GOWORK=off and released dependencies. An isolated consumer importing all three published v0.6.0 modules, without local replacements, passed 33 cases spanning the roadmap's capture/mode combinations, aliases, and quiet-destination policy. The resolved graph selected the intended ConfigKit v1.1.0 and PromptKit v0.20.0 dependencies.

On 2026-11-10, the newly selected ckconfig and ckprompt v0.7.0 tagged suites also passed independently with released dependencies. Their changes update dependency pins and release documentation without changing adapter implementation. The new graph selects ConfigKit v1.2.0, PromptKit v0.21.0, and core v0.6.0. These checks establish the upstream gate; they do not replace Audita's assembled integration acceptance. No additional upstream request was identified.