| name | review |
| description | Use when the user wants a code review of the current diff or recent changes. Triggered by /review, 'review this', 'review the diff', 'do a review', 'review my changes'. Wraps the workflow in agent_docs/review_process.md — runs automated checks, evaluates all P0/P1/P2 categories, fixes findings inline, writes deferred items to BACKLOG.md. |
| disallowed-tools | AskUserQuestion |
Review — Diff / Recent-Changes Code Review
When to Use
- User says "/review", "review this", "review the diff", "review my changes", "do a review"
- After implementing a feature — done-skill does NOT run review automatically (it stays predictable). User invokes review skill explicitly when they want it.
Source of Truth
The full review workflow lives in agent_docs/review_process.md. Do not duplicate it here. This skill is a thin invocation wrapper:
- Read
agent_docs/review_process.md to load the current rules (severity definitions, categories, fixing rules, commit gate).
- Determine review scope: diff vs full-read vs large-scale (>30 changed files).
- Execute the workflow as defined in the process file.
- Produce the standard
### Code Review Results table.
- Write deferred findings to
BACKLOG.md per agent_docs/backlog_process.md.
Workflow
1. git status + git diff -> identify changed files
2. Read agent_docs/review_process.md -> load rules
3. Run automated checks (lint / typecheck / test / build) per CLAUDE.md
4. (If GitNexus available) gitnexus_impact on changed symbols
5. Re-read every changed file completely
6. Evaluate against P0/P1/P2 categories
7. Fix P0/P1 inline; defer P2 only if non-trivial
8. Regression & complexity QA pass
9. UI review (only if UI changed)
10. Output standard Code Review Results table
11. Write deferred findings to BACKLOG.md
Rules
- Do not run automatically. This skill is on-demand only. Done-skill must NOT trigger it.
- No security-deep-dive here — for OWASP / injection / secrets / auth focus, use the separate
security-review skill. Generic review still covers Security at the P0 level per review_process.md, but security-review goes deeper.
- Output format is fixed — must produce the table defined in
review_process.md so consumers (CI dashboards, follow-up prompts) can parse it.
- Re-read files, never review from memory.
Report
Use the exact format from review_process.md:
### Code Review Results
| # | Category | Sev | Status | Finding | Action |
|---|----------|-----|--------|---------|--------|
| ... |
Summary: X categories checked | Y fixed | Z deferred -> Backlog
Append a one-line skill footer:
review skill -- full process: agent_docs/review_process.md
