Analyze

Analyze Skill

Run a structured codebase analysis and produce actionable findings.

Steps

1. Orient — read CLAUDE.md, Cargo.toml (workspace members), and recent git log --oneline -20.

2. Spawn parallel sub-agents for independent analysis streams:

   • Architecture agent: Check for hexagonal boundary violations (infra types leaking into domain, missing trait abstractions), SOLID principle adherence, module cohesion
   • Test coverage agent: Find untested public functions, missing edge cases, snapshot tests that need updating
   • Tech debt agent: Find TODOs/FIXMEs, dead code (#[allow(dead_code)]), large files (>500 lines), duplicated logic

3. Quality gates — run and report:

   • cargo clippy --all-targets 2>&1 | grep -E "^error|^warning" | head -30
   • cargo test --quiet 2>&1 | tail -20

4. Synthesize findings into a prioritized report:

   ## Critical (block next PR)
   - ...
   
   ## High (fix this sprint)
   - ...
   
   ## Low (backlog)
   - ...
   

5. If devkit is available, also run devkit health and devkit repo-review for scored reports.

6. Offer to create doob tasks from findings (bd create) if the user wants to track them.

Rules

• Focus on actionable findings, not general observations.
• Cap output at ~40 findings total — prioritize ruthlessly.
• Do not make code changes. This skill is read-only.