| name | pactkit-audit |
| description | H1-H7 AI Readiness Assessment — harness audit scoring and hotspot analysis |
| model | sonnet |
PactKit Audit
H1-H7 AI Coding Harness readiness assessment. Scans project structure against a 7-layer model, produces a Harness Score (0-100), and identifies code hotspots.
When Invoked
- Done Phase 8 (Harness Audit Refresh): Auto-invoked via
pactkit audit --append.
- Standalone: Ad-hoc project health assessment.
- Headless / Frontend integration: Called as
/pactkit-audit via Claude Code headless mode.
The 7 Layers
| Layer | Name | What It Checks |
|---|
| H1 | Prompt Engineering | CLAUDE.md, rules, agents, commands |
| H2 | Context Engineering | Specs, test cases, architecture graphs |
| H3 | Process Governance | Sprint board, PDCA workflow artifacts |
| H4 | Tool Governance | Skills, CLI tools, MCP integration |
| H5 | Safety & Guardrails | Pre-commit hooks, security rules, credential guards |
| H6 | Observability | Logging, cost tracking, lessons, self-audit rules |
| H7 | Evolution | Version management, changelog, CI/CD publish |
Scoring
- Each layer scores L0-L3 (None → Basic → Structured → Advanced).
- Harness Score = sum(all 7 levels) / 21 × 100.
- AI Ready = min(all 7 levels) ≥ L1.
Protocol
1. Run Audit
pactkit audit
- Outputs: scorecard + top hotspots (concise by default).
- Output file:
docs/architecture/governance/harness_audit.json.
2. Options
| Flag | Purpose |
|---|
--json | JSON output only (machine-readable) |
--layer H1-H7 | Check a single layer |
--append | Silent update for /project-done integration |
--if-needed STORY_ID | Only refresh when harness_audit.json exists and its story_id matches; skips otherwise |
--verbose | Full detail: findings + insights + hotspots |
3. Interpret Results
- Score < 50: Major gaps — address weakest layer first.
- Score 50-80: Functional but room for improvement.
- Score > 80: Strong harness — focus on hotspot refinement.
- Hotspots: Files ranked by composite score (complexity, blast radius, fan-in, test coverage, smells).
4. Suggested Tasks
When hotspots are detected and developer is configured in pactkit.yaml, the audit generates suggested task entries that can be added to the sprint board.
Usage Scenarios
/project-done: Run pactkit audit --append --if-needed {STORY_ID} — only refreshes if the audit belongs to this story; silently skips otherwise.
- Sprint planning: Run
pactkit audit --verbose to identify technical debt priorities.
- CI integration: Run
pactkit audit --json for machine-readable pipeline checks.
- Frontend dashboard: Call
/pactkit-audit via headless mode for real-time project health.