reviewer-fix-mode

Reviewer Fix Mode

Overview

Use this skill when reviewers are allowed to fix issues they find.

Reviewer-fix mode is useful, but only if the review happens first and the fix scope is locked to documented findings.

When to Use

• A reviewer may patch issues after review.
• A review found clear defects and the owner authorises direct repair.
• A strict harness lets reviewers fix after documenting findings.
• You need to stop reviewer scope drift.
• Orchestrator must final-sign off reviewer fixes.

Non-negotiable Sequence

1. Review first. Write findings before editing.
2. Declare mode. Say reviewer-fix mode is starting.
3. Fix only documented findings. No opportunistic cleanup.
4. Verify fixes. Run targeted tests and relevant smoke checks.
5. Report separately. Keep review findings and fix evidence distinct.
6. Return to orchestrator. Reviewer does not self-merge by default.

Finding Format

### HIGH-1: [summary]
- Evidence: file.py:LN
- Regression analysis: if X changes, which test catches it?
- Fix scope: exactly what may change

Fix Report Format

## Reviewer-fix mode report

### Fixed findings
- HIGH-1: [what changed]

### Changed files
- path/file.py

### Verification
- command: result
- runtime/API/browser proof: result

### Out of scope
- [things noticed but not fixed]

### Remaining risk
- [none / listed]

Allowed Fixes

Allowed:

• direct bug fixes for documented findings;
• regression tests for documented findings;
• small refactors required to fix the finding;
• docs updates that explain changed behaviour.

Not allowed without orchestrator approval:

• new features;
• design/product changes;
• broad formatting;
• architecture rewrite;
• unrelated test cleanup;
• changing public contracts beyond the finding.

Orchestrator Final Check

The orchestrator checks:

• every fix maps to a documented finding;
• no unrelated files changed;
• tests prove the regression mode;
• production-shaped path is covered if relevant;
• reviewer did not silently become second builder.