kenoxa/run-architecture-audit

Read-only audit — no file writes, no code changes.

Cross-reference: references/deep-modules.md for depth heuristics, dependency categories, Design It Twice.

Phases

Reference convention: linked refs load into mainthread. Backticked paths → dispatch to subagent, do NOT Read into mainthread.

Phase Trace: Log row at scope, explore (or zero-dispatch for focused), analyze, synthesize. Include scope classification and candidate count.

All phases: orchestrate-audit.md.

| Phase | Agent | Reference | |-------|-------|-----------| | Scope | @scout | run-explore/references/explore-scout.md | | Explore | @scout | references/audit-scout.md | | Analyze | @researcher | references/analyze-researcher.md | | Synthesize | @synthesizer | references/audit-synthesis.md |

Candidate Format

Per candidate:

• Module path — repo-relative directory or file
• Coupling indicators — export count, caller count, pass-through functions
• Dependency category — in-process / local-substitutable / remote-but-owned / true-external
• Depth assessment — shallow/medium/deep with depth ratio estimate
• Test impact — current test strategy vs recommended boundary

Output Format

Write architecture-findings.md to .scratch/<session>/:

1. Summary — scope audited, candidate count, top friction areas (2-3 sentences)
2. Candidate table — module | depth | dependency category | priority
3. Per-candidate analysis — coupling indicators, depth assessment, test impact, deepening approach
4. Next step — suggest /do-frame for spec creation (large scope) or /do-design with architecture-depth lens (focused scope)

Anti-Patterns

• Proposing code changes (audit diagnoses; do-design/do-build prescribes)
• Auditing test quality in isolation (use run-review for that)
• Skipping dependency classification (every candidate needs a category)
• Conflating shallow modules with small modules (small + deep = fine)