| name | review-fix-loop |
| description | Required workflow for handling review findings with classification, focused fixes, verification, fresh-context subagent review, and PR artifact updates. |
| license | MIT |
| tags | ["review","quality","workflow"] |
| allowed-tools | ["bash","git","markdown"] |
| metadata | {"author":"laurenceputra","version":"1.0.0"} |
Review-Fix Loop
Use this skill whenever review produces a finding that may require code, docs, test, or workflow changes.
Required Triggers
- A reviewer reports a
blocking or important finding.
- A review comment identifies a correctness, security, privacy, sync, financial-calculation, CI, or regression risk.
- A previously applied fix receives follow-up review feedback.
- A finding is being declined or marked invalid and the decision needs fresh-context validation.
Minor findings may be batched, but each must still receive a disposition before PR completion.
Loop Contract
Run this sequence for each blocking or important finding:
- Capture the finding.
- Classify severity and disposition.
- Request fresh-context subagent assessment when required.
- Implement the smallest correct fix.
- Run focused QA.
- Run broader affected-surface QA.
- Request fresh-context post-fix review when required.
- Perform main-agent self-review.
- Update the PR artifact.
- Push and watch checks when the loop affects CI-covered behavior or is the final loop.
All required loop steps, including PR artifact updates, must be complete before the final push and PR completion gate.
Severity Routing
blocking: fresh assessment subagent and post-fix review subagent are required.
important: post-fix review subagent is required; fresh assessment subagent is recommended.
minor: main-agent self-review is sufficient unless the change is risky or cross-surface.
invalid or declined: fresh assessment subagent is recommended before closing the finding.
duplicate: link to the existing loop item and inherit its verification evidence.
Subagent Responsibilities
Subagents provide independent assessment and review. The main agent remains responsible for edits, commits, PR artifact updates, and check watching.
Fresh assessment prompt shape:
Review this finding with fresh context. Do not edit files.
Return:
- whether the finding is valid
- affected files and surfaces
- likely root cause
- smallest safe fix direction
- tests that should be run or added
- risks or adjacent cases likely to be missed
Post-fix review prompt shape:
Review the implemented fix with fresh context. Do not edit files.
Return findings only:
- blocking, important, or minor issues
- file and line references
- missing tests
- residual risks
- whether the review-fix loop can close
QA Expectations
- Focused QA should target the changed behavior first.
- Broad QA should cover the affected repo surface before the loop closes.
- If a check cannot run locally, record why and identify the CI check or manual evidence that covers the gap.
- Never soften, skip, or delete tests to force a green result unless the review finding is specifically about an invalid test and the rationale is recorded.
Required PR Artifact Entry
Add or update the Review Response Matrix for each finding that entered a review-fix loop or was explicitly deferred or declined:
| Finding | Severity | Disposition | Fix Location | Verification Evidence | Residual Risk |
| --- | --- | --- | --- | --- | --- |
| | | | | | |
Verification evidence should include focused QA, broader affected-surface QA, main-agent self-review, and required fresh-context subagent review when applicable.
Exit Criteria
A loop is closed only when all are true:
- The finding has a disposition:
fixed, deferred, declined with rationale, duplicate, or invalid with rationale.
- Code or docs changes are committed and pushed when applicable.
- Focused QA passed or is explicitly blocked with rationale.
- Broad QA passed or is explicitly blocked with rationale.
- Required fresh-context subagent review has completed.
- Main-agent self-review found no new
blocking or important issues.
- The PR artifact records the finding, fix, evidence, and residual risk.