| name | gormes-review-loop |
| description | Use when Gormes work has PR feedback, CI failures, reviewer comments, static-analysis findings, or a need to iterate until a bounded validation target is green. |
Gormes Review Loop
Overview
Review feedback is a loop, not a vibes pass. Fix one verified issue at a time, prove it, and stop only when the agreed gate is green or a blocker is documented.
When to Use
Use for:
- PR review comments, GitHub checks, CI failures,
git diff --check, or reviewer confidence issues;
- “keep going until green”, “address feedback”, “make this mergeable”, or “review loop” requests;
- self-review before
gormes-git or gormes-release.
Do not use to expand scope. Large PRs should be split before looping.
Workflow
- Capture the target gate: review comments, Greptile score, GitHub check, CI annotation, test command, progress validation, or diff check.
- Read the exact failing evidence; do not infer from summaries. Preserve reviewer text, PR/check URL, file path, line/symbol, command output, and commit SHA when available.
- Pick one issue and classify it: bug, test drift, docs drift, style, blocker, or row-shaping needed.
- If the feedback is a broad external-review finding that must become canonical backlog work, route to
gormes-planner with the exact evidence. The planner must refine an existing progress row or add one builder-ready row; do not create side queues.
- For code behavior that is already builder-ready, use
gormes-tdd-slice: failing test first, minimal fix, rerun focused test.
- Re-run the target gate.
- Repeat until green, or document the blocker and pivot per workspace rules.
Quick Reference
| Evidence | Next step |
|---|
| Test failure | Reproduce focused failure before editing |
| Review comment | Map to file and exact requested behavior |
| Greptile/Grep-style review finding that changes backlog shape | Route exact evidence to gormes-planner |
| Diff whitespace | Run git diff --check after edit |
| Progress docs drift | Run go run ./cmd/progress validate |
| Full merge gate | Run go test ./... -count=1, progress validate, diff check |
Stop Conditions
- Target gate is green with command output captured.
- Same command failed twice with the same blocker.
- Fix requires scope expansion; ask Juan or route to planner.
Common Mistakes
- Letting an automated loop rewrite broad areas without focused evidence.
- Turning external review findings into private TODOs instead of progress-row refinements.
- Marking review “done” after edits but before rerunning the gate.
- Treating a confidence score as proof; command output and reviewer comments are evidence.
- Continuing after a blocker without documenting it in required locations.