loop-code-review

Files (1)

• SKILL.md

SKILL.md (first lines)

# Loop Code Review

Use this skill when a change is too important for a single pass: shared runtime behavior, security/privacy, data/schema, deploy, release, dependency, or broad refactor work.

## Goal

Turn review into a closed loop:

1. implement a narrow change;
2. get an independent review;
3. fix actionable findings;
4. re-run the smallest proof that covers the touched surface;
5. stop only when the latest review has no material findings or the project-specific acceptance score is met.

## Workflow

1. Inspect the current worktree and identify the intended write set.
2. Name the changed behavior, the risk surface, and the smallest useful verification commands.
3. Ask an independent reviewer or review mode to focus on bugs, regressions, missing tests, security/privacy issues, and unclear behavior.
4. Apply only actionable findings that improve correctness or maintainability.
5. Record any rejected finding with a short reason when the tradeoff matters.
6. Re-run targeted tests, type/lint/build, or browser checks that match the touched surface.
7. Repeat only while the reviewer finds material issues. Avoid cycling on subjective style comments.

## Reviewer Prompt Shape

```text
Review this diff as a bug/security/regression reviewer.
Prioritize findings over summary.
Focus on:
- changed runtime behavior;
- missing tests or weak coverage;
- auth, data, privacy, injection, logs, and dependency risk;
- rollback, deploy, and migration risk if touched.
Do not rewrite for preference. Report only actionable findings.
```

## Acceptance

The loop is complete when:

- the latest review has no high or medium actionable findings;
- required checks for the touched surface pass;
- the remaining risk is explicit and acceptable for the project phase;
- unrelated files and user changes were not included.

For high-risk work, require the project-specific stronger review threshold before merge or release.

From iKairoseki/codex-agent-coordination, MIT.

More from iKairoseki/codex-agent-coordination