jokerman89 /code-review
Reviews the current branch's diff before landing. Lighter than /plan-eng-review (which reviews a plan/design doc). Use when there's no plan but you want a code review pass before /ship.
Auto-scales: small diffs get fast review; large diffs (200+ lines) additionally get Codex structured review with P1 gate.
When to use
- Branch is feature-complete, no design doc exists
- Quick correctness pass before
/ship - Small fixes / refactors that don't warrant
/plan-eng-review /shipblocked by "Eng Review NOT CLEARED" and the work is too small for full plan-eng-review
When NOT to use
- Plan exists — use
/plan-eng-reviewagainst the plan instead - Diff too small to be reviewable (<5 lines, trivial fix) — skip review
- Tests fail — fix first, review after
Inputs
- Optional
--base <branch>— compare against this base instead of default branch (default:main) - Optional
--codex— force Codex structured review even on small diffs - Optional
--no-codex— skip Codex even on large diffs
Workflow
- Diff stats —
git diff <base>...HEAD --stat. Auto-classify:- <50 lines changed: SMALL (Claude-only review)
- 50-200: MEDIUM (Claude + optional Codex)
- 200+: LARGE (Claude + mandatory Codex with P1 gate, unless --no-codex)
- Read diff content —
git diff <base>...HEAD. Identify changed files + functions + tests. - 4-dimension review (lighter than
/plan-eng-review's 4 sections — no per-issue AskUserQuestion overhead):- Architecture impact (does the diff respect existing boundaries?)
- Code quality (DRY violations, error handling, edge cases)
- Test coverage (does the diff add tests for new code paths? regression risk?)
- Performance (N+1, memory, slow paths introduced)
- Codex pass (if LARGE diff or --codex) — invoke Codex with structured review prompt. P1 findings BLOCK ship.
- Persist via
bin/li-review-logwithskill: code-review(distinct from Phase-6/reviewandplan-eng-review). - Output: findings list + severity + suggested fixes.
Report format
Review Status: <branch> vs <base>
Diff scope: 173 lines changed across 8 files (MEDIUM)
Codex pass: optional, skipped this run
## Findings (3)
[P2] (confidence: 8/10) src/services/billing.ts:47 — N+1 query in refund flow
Loop calls payment.fetch() per refund; batch via .fetchMany()
Suggested: ~5 line refactor
[P3] (confidence: 7/10) src/utils/format.ts:12 — DRY violation
formatDate duplicated in format.ts and date-helpers.ts; consolidate
[P3] (confidence: 9/10) tests/billing.test.ts:NEW — Missing edge case
refundPayment with 0 amount: no test. Add test or document why allowed.
## Verdict
- P1 count: 0 — no block
- P2 count: 1 — should fix before /ship
- P3 count: 2 — recommended fixes
Run /ship when P2+ resolved.
Persist via bin/li-review-log:
bin/li-review-log '{"skill":"code-review","timestamp":"...","status":"...","findings":N,"findings_fixed":N,"gate":"P1_clean","commit":"..."}'
Compliance integration
- Sanity scan on diff content (secrets, customer-data, PII) — fast pre-check. Blocks if hit.
cli_support+voicefrontmatter validation on any new/changed skill or agent files in the diff.
Confidence scoring
Every finding gets a 1-10 confidence:
- 9-10: verified by reading specific code
- 7-8: high-confidence pattern match
- 5-6: medium — show with "verify this is actually an issue" caveat
- 3-4: suppressed from main report, appendix only
- 1-2: speculation only (rarely reported)
Failure modes
- Diff empty: no changes to review. Report + exit.
- Diff too large to fully analyze: spot-check + warn operator that some files weren't deeply reviewed.
- Codex unavailable: fall back to Claude adversarial subagent. Note in report.
bin/li-review-logunavailable: still print findings, but no dashboard update.
Examples
Small clean diff:
> /code-review
Diff scope: 23 lines, 2 files (SMALL)
✓ No findings. Clean to /ship.
Medium diff with findings:
> /code-review
[3 findings reported]
P1: 0, P2: 1, P3: 2
Action: fix P2 before /ship.
Large diff with Codex P1 gate:
> /code-review
Diff: 412 lines, 18 files (LARGE)
Codex pass: P1 found — race condition in payment-handler.ts:108
✗ /ship BLOCKED until P1 resolved
See also
/plan-eng-review— heavier plan-stage review (use when design doc exists)/investigate— debugging when /code-review finds something broken/ship— reads /code-review's dashboard entry as ship-gate signal (within 7 days, current commit)