Review

Review

Check the changes between HEAD and a fixed point you supply, two ways:

• Conventions - does the code follow your project's documented conventions?
• Matches the issue - does it actually do what the originating issue or spec asked for?

The two checks run separately, as parallel sub-agents, so neither colours the other. A change can pass one and fail the other - correct code that does the wrong thing, or the right behaviour built against your conventions - so they're kept apart and reported side by side.

Step 1 - Pin the fixed point

Whatever you said is the fixed point: a commit SHA, branch, tag, main, HEAD~5, or a PR number. Pass it through; don't be opinionated. If you didn't specify one, ask: "Review against what - a branch, a commit, or main?" Don't proceed without it.

Capture the diff once:

git diff <fixed-point>...HEAD      # three-dot: compares against merge-base
git log <fixed-point>..HEAD --oneline

For a PR: gh pr diff [NUMBER] and gh pr view [NUMBER].

Step 2 - Identify the spec source

Look, in order:

1. Issue references in commit messages (#123, Closes #45) - fetch with gh.
2. A path I passed as an argument.
3. A spec/PRD file under docs/, specs/, or .scratch/ matching the branch.
4. If nothing is found, ask where the spec is. If there isn't one, the Spec sub-agent skips and reports "no spec available".

Step 3 - Identify the standards sources

Anything documenting how code should be written:

• CLAUDE.md, AGENTS.md, CONTRIBUTING.md, CONTEXT.md
• docs/adr/ (architectural decisions are standards)
• STYLE.md, STANDARDS.md, STYLEGUIDE.md
• Machine-enforced config (eslint, prettier, tsconfig) - note it but don't re-check what tooling already enforces.

Step 4 - Spawn both sub-agents in parallel

One message, two Task tool calls.

Standards sub-agent:

Read these standards docs: [list]. Then read this diff: [diff].
Find every place the diff violates a documented standard. Cite the standard
(file + rule). Skip anything tooling enforces. Also flag security issues:
unsanitised input, hardcoded secrets, unvalidated endpoints, language-specific
risks (SQL injection, prototype pollution, path traversal).

Return ONLY a terse bullet list, one finding per line, no prose:
  - ❌ file:line - issue (cite standard)        # hard violation / security
  - ⚠️ file:line - issue (cite standard)        # judgement call
Then a final line: VERDICT: PASS  or  VERDICT: FAIL
If there are no findings: "- ✅ follows your conventions" then VERDICT: PASS.

Spec sub-agent:

Read the spec: [path or contents]. Then read this diff: [diff].
Find: (a) requirements missing or partial; (b) behaviour not asked for (scope
creep); (c) requirements that look implemented but wrong. Confirm tests exist
for new behaviour and test through public interfaces, not implementation details.

Return ONLY a terse bullet list, one finding per line, no prose:
  - ❌ issue - quote the spec line it relates to
  - ⚠️ issue - quote the spec line it relates to
Then a final line: VERDICT: PASS  or  VERDICT: FAIL
If everything is satisfied: "- ✅ does everything the issue asked for" then VERDICT: PASS.

If the spec is missing, skip the Spec sub-agent and note it.

Step 5 - Aggregate

Combine the two bullet lists into one compact, verdict-first report. Do not paste the sub-agent prose; use their bullets directly. Do not merge or rerank the tracks.

## Review - PR #N        (or the diff range)

**Verdict: [✅ SAFE TO MERGE | ❌ DO NOT MERGE]** · Conventions: [✅ | ❌] · Matches the issue: [✅ | ❌]

I checked it two ways - against your conventions, and against what the issue asked for.

**Conventions ([n])**
- ❌ file:line - issue (security)
- ⚠️ file:line - issue

**Matches the issue ([n])**
- ✅ does everything the issue asked for

**Biggest concern:** [one line, or "none"]
**Next:** fix and re-run /review, or merge on GitHub if it's safe

Markers: ❌ blocker/fail · ⚠️ judgement call · ✅ pass.

Rules

• A security finding on the Standards track is always DO NOT MERGE.
• A Spec FAIL (missing requirement or wrong implementation) is DO NOT MERGE.
• Don't suggest improvements beyond the two tracks - the issue had defined acceptance criteria; anything more is scope creep.
• This is a read-only review. Never modify code or merge.