requesting-code-review · On-Demand Code Review

requesting-code-review brings in a second pair of eyes before merge: it packages review context to a template (change description + requirements/plan + BASE/HEAD commit SHAs) and dispatches a code-reviewer subagent with no session history for a fresh read of the diff, returning severity-graded findings.
Mandatory after each task, after major features, and before merging; recommended when stuck, before refactoring, or after fixing a complex bug. Critical issues are fixed immediately, Important before proceeding, Minor noted for later — and a wrong verdict can be pushed back on with evidence.
Example invocation: "Review my changes before I merge."
Full brief
Positioning
requesting-code-review is the quality gate before merge: a reviewer subagent with none of your session history reads the diff between two commits with fresh eyes, catching issues before they cascade. It writes no code — it returns findings and handling advice.
Core capabilities
- Precise review context: the
code-reviewer.mdtemplate packages description, requirements/plan, BASE_SHA, and HEAD_SHA — exactly what the reviewer needs, never your session history. - When to request: mandatory — after each task, after major features, before merging to main; recommended — when stuck (fresh perspective), before refactoring (baseline), after fixing a complex bug.
- Graded handling: Critical — fix immediately; Important — fix before proceeding; Minor — note for later.
- Push back allowed: a wrong verdict gets a technical rebuttal with code/test evidence, not blind compliance.
- No-excuse list: never skip review for being "too simple", never ignore Critical, never proceed with unfixed Important.
Workflow
- Get git SHAs:
BASE_SHA(start, e.g.HEAD~1or the merge-base with main),HEAD_SHA(end) - Dispatch the reviewer subagent with the filled template
- Act on feedback by severity: fix Critical/Important, log Minor
Inputs & outputs
| Input | Notes |
|---|---|
| Change description | What this change does |
| Requirements / plan | What it should achieve |
| BASE_SHA / HEAD_SHA | The review range |
Output: findings (strengths + severity-graded issues) and handling advice (fix / log / rebut).
Boundaries with adjacent skills
- brainstorming / executing-plans: design and execution; this skill is the review gate after execution.
- executing-plans calls this skill in its closing whole-branch review stage.
Fit
- Quality check before merging to main
- Whole-feature review after major work
- Fresh eyes on your own blind spots
- Confirming a complex bug fix introduced nothing new
Before you start
- No API key; changes must be committed to git (computable BASE/HEAD SHAs required).
- The template lives at
code-reviewer.mdin the skill directory. - Graded handling is a hard rule: Critical blocks the merge, Important blocks progress.
requesting-code-review is part of the Aiglade Skill library. Invoke it from the Aiglade chat box in plain language.