Skip to main content

Detect Skills

Evidence-based repo audits with confidence-scored findings.

Pipeline​

All detect skills support --light (default) and --full modes. Light mode produces 1 summary file per domain; full mode produces detailed multi-file output.

Light Mode (default)​

/detect-dev        → $JAAN_OUTPUTS_DIR/detect/dev/summary.md        (1 file)
/detect-design     → $JAAN_OUTPUTS_DIR/detect/design/summary.md     (1 file)
/detect-writing    → $JAAN_OUTPUTS_DIR/detect/writing/summary.md    (1 file)
/detect-product    → $JAAN_OUTPUTS_DIR/detect/product/summary.md    (1 file)
/detect-ux         → $JAAN_OUTPUTS_DIR/detect/ux/summary.md         (1 file)
                                    ↓
/detect-pack       → $JAAN_OUTPUTS_DIR/detect/summary.md            (1 file)

Full Mode (--full)​

/detect-dev --full        → $JAAN_OUTPUTS_DIR/detect/dev/*.md        (9 files)
/detect-design --full     → $JAAN_OUTPUTS_DIR/detect/design/*.md     (6 files)
/detect-writing --full    → $JAAN_OUTPUTS_DIR/detect/writing/*.md    (6 files)
/detect-product --full    → $JAAN_OUTPUTS_DIR/detect/product/*.md    (7 files)
/detect-ux --full         → $JAAN_OUTPUTS_DIR/detect/ux/*.md         (7 files)
                                           ↓
/detect-pack --full       → $JAAN_OUTPUTS_DIR/detect/pack/*.md       (4+ files)

detect-pack handles mixed inputs: domains can be light or full independently.

Available Skills​

SkillDescriptionLightFull
/detect-devEngineering audit with OpenSSF-style scoring1 file9 files
/detect-designDesign system detection with drift findings1 file6 files
/detect-writingWriting system extraction with NNg tone scoring1 file6 files
/detect-productProduct reality extraction with 3-layer evidence1 file7 files
/detect-uxUX audit with Nielsen heuristics and journey mapping1 file7 files
/detect-packConsolidate all detect outputs into scored index1 file4+ files

Multi-Platform Support​

All detect skills automatically detect and analyze multi-platform monorepos:

  • Platform auto-detection: Scans folder structure (web/, backend/, mobile/, etc.) using configurable patterns
  • Platform-scoped filenames: stack-web.md, stack-backend.md instead of nested folders
  • Evidence ID prefixing: E-DEV-WEB-001 for multi-platform, E-DEV-001 for single-platform
  • Cross-platform linking: Use related_evidence field to link findings across platforms
  • "Detect and Report N/A" pattern: All domains always produce output (even if N/A for that platform)
  • Merged pack: detect-pack consolidates all platforms into cross-platform risk heatmap

See individual skill documentation for platform-specific behavior.

Shared Standards​

All detect skills share:

  • Evidence format: SARIF-compatible blocks with id, type, confidence, location (uri, startLine, snippet), method
  • Evidence ID namespaces:
    • Single-platform: E-DEV-NNN, E-DSN-NNN, E-WRT-NNN, E-PRD-NNN, E-UX-NNN
    • Multi-platform: E-DEV-{PLATFORM}-NNN (e.g., E-DEV-WEB-001, E-DSN-BACKEND-023)
  • 4-level confidence: Confirmed (0.95–1.00) / Firm (0.80–0.94) / Tentative (0.50–0.79) / Uncertain (0.20–0.49)
  • Universal frontmatter: title, id, version, status, target.commit, target.platform, tool metadata, confidence_scheme, findings_summary, overall_score (0–10), lifecycle_phase
  • Document structure (Diataxis): Executive Summary → Scope/Methodology → Findings → Recommendations → Appendices
  • Output path: $JAAN_OUTPUTS_DIR/detect/{domain}/ — flat files with platform suffixes, overwritten each run (no IDs)

Seed Reconciliation​

After detection completes, each skill checks findings against project seed files (context/tech.md, tone-of-voice.template.md, integrations.md, localization.template.md) and reports discrepancies (version drift, missing references, contradictions, stale patterns).

detect-pack goes further: it uses consolidated detection data as source of truth to actively update seed files with a diff-style preview and user approval workflow ([y/all/n/pick]). A reconciliation report is written to $JAAN_OUTPUTS_DIR/detect/seed-reconciliation.md.

See seed-reconciliation-reference.md for full comparison rules and update protocol.

Reference​

Back to Skills