| name | review-fix-loop |
| description | Use this skill when addressing review findings from the reviewer-agent. Maps every finding to a specific code change and verification evidence. |
Review Fix Loop
Close review rounds systematically by mapping every finding to a code change and proof.
Execution Loop
- Parse — categorize findings by severity (P1/P2/P3) and type (AUTO-FIX/ASK/INVESTIGATE)
- Build closure table — one row per finding, planned change identified BEFORE editing
- Implement — fix the smallest coherent batch
- Verify narrowly — run targeted tests for each fix
- Update closure table — record verification command and result
- Verify broadly — run full test suite and lint
Closure Table
Build this before making any changes:
| # | Finding | Severity | Type | File(s) | Planned Fix | Evidence Type | Verify Command | Result |
|---|
| 1 | [desc] | P1 | AUTO | [path] | [what] | [type] | [cmd] | |
Evidence types: red-test, green-test, lint, typecheck, integration, contract, manual-inspection, not-run-with-risk
Fill the Result column AFTER each fix is verified. A finding is not closed by prose — it is closed by code change plus evidence. For test-fidelity findings, closure must include either improved behavior test, reduced mock coupling, fake/contract/integration coverage, or documented justification for current test double.
Verification Ladder
Work from narrow to broad:
- Narrow — unit test for the specific function touched
- Medium — compile, lint, type-check for touched files
- Broad — full test suite, project-level checks
Handling ASK Items
Do NOT guess — batch them and report to the orchestrator with your analysis and recommendation.
Handoff
## Review Fix Report
**Findings addressed:** N of M
**Closure table:** [completed table]
**Verification evidence:** [commands and results]
**Open items:** [ASK/INVESTIGATE items pending]
**Residual risk:** [anything needing follow-up]
