audita/internal/framework/llm/source_paths.go

package llm

import (
	"fmt"
	"path/filepath"
	"sort"
	"strconv"
	"strings"

	ck "go.maximumdirect.net/configkit"
	pk "go.maximumdirect.net/promptkit"
	"gopkg.in/yaml.v3"
)

func capturedSourcePaths(engine *pk.Engine, hostRoot string) ([]string, error) {
	inspection, err := engine.InspectConfiguration()
	if err != nil {
		return nil, fmt.Errorf("inspect configuration: %w", err)
	}

	sources := inspection.Sources()
	byID := make(map[string]ck.SourceInfo, len(sources))
	paths := make(map[string]struct{}, len(sources))
	hostPath := func(source ck.SourceInfo) string {
		if source.Kind == ck.SourceKindFile {
			return source.Path
		}
		if hostRoot != "" && source.Kind == ck.SourceKindFilesystem && strings.HasPrefix(source.Label, operatorFilesystem+"/") {
			return filepath.Join(hostRoot, filepath.FromSlash(source.Path))
		}
		return ""
	}
	for _, source := range sources {
		byID[source.ID] = source
		if path := hostPath(source); path != "" {
			paths[path] = struct{}{}
		}
	}

	metadata := inspection.Metadata()
	for _, path := range metadata.HistoricalPaths() {
		segments := strings.Split(strings.TrimPrefix(path.String(), "/"), "/")
		// Prompt resources are lists at the root or inside a collection.
		if len(segments) == 7 && segments[0] == "collections" {
			segments = segments[2:]
		}
		if len(segments) != 5 || segments[0] != "prompts" || segments[2] != "messages" || segments[4] != "content_file" {
			continue
		}
		if _, err := strconv.Atoi(segments[1]); err != nil {
			continue
		}
		if _, err := strconv.Atoi(segments[3]); err != nil {
			continue
		}
		field, ok := metadata.Lookup(path)
		if !ok || field.Null || len(field.Origin.Locations) != 1 {
			continue
		}
		source, ok := byID[field.Origin.Locations[0].SourceID]
		if !ok || hostPath(source) == "" {
			continue
		}
		explanation, ok := inspection.Explain(path)
		if !ok || !explanation.Effective.Present || explanation.Effective.Redacted {
			continue
		}
		var resource string
		if err := yaml.Unmarshal([]byte(explanation.Effective.YAML), &resource); err != nil {
			return nil, fmt.Errorf("decode captured message resource at %s: %w", path.String(), err)
		}
		paths[filepath.Join(filepath.Dir(hostPath(source)), filepath.FromSlash(resource))] = struct{}{}
	}

	result := make([]string, 0, len(paths))
	for path := range paths {
		result = append(result, path)
	}
	sort.Strings(result)
	return result, nil
}