| name | review |
| description | Review a PR or branch using the writer/reviewer pattern — flag issues with specific line references, never rewrite the code. Use when the user asks to review someone's PR, wants a second-opinion pass on a diff, says 'critique this', 'what's wrong with this branch', 'give me review comments', or hands a branch over for inspection rather than implementation. |
Review the changes on the current branch (or the branch/PR specified in $ARGUMENTS).
Important
You are the reviewer, not the writer. Do NOT rewrite the implementation. Flag issues with specific line references.
Process
-
Determine the target:
- No argument: review current branch changes vs main
- Branch name: review that branch vs main
- PR number: fetch the PR and review its changes
-
Gather context:
git log --oneline main..HEAD
git diff main...HEAD
-
Check spec compliance - Find the relevant spec in docs/specs/. Does the implementation match?
-
Run stack-aware code review — Prefer Anthropic's official code-review skill (from claude-code-plugins) if installed. Otherwise, run /sec-review for security findings and spawn the architecture-reviewer subagent for architecture findings — in parallel. Load the matching language guide from reviews/ (go.md, rust.md, typescript.md, python.md) and pass it to each as stack-specific criteria, so findings cite idiomatic patterns and framework pitfalls.
-
Your own review - Focus on what subagents won't catch:
- Business logic correctness - Does it solve the actual problem?
- Design decisions - Are the tradeoffs right?
- What's missing - What didn't the author think of?
- Test layer coverage - Check against
docs/TECHNICAL_DESIGN_DOCUMENT.md Testing Strategy (if it exists) or infer from codebase patterns:
- Are unit tests present for new business logic?
- Are integration tests present for new API endpoints, DB operations, or service boundaries?
- Are e2e tests present for critical user flows (if applicable)?
- Do tests follow the project's conventions (location, naming, framework)?
- Are verification criteria from the spec covered at the right test layer?
- Edge cases - Are failure modes handled?
-
Consolidate findings into a single report:
# Review: [branch/PR]
## Verdict: APPROVE | REQUEST_CHANGES | COMMENT
## Spec Compliance
- [Does it match the spec? What's missing?]
## Security (from code review)
- [Summary of findings]
## Architecture (from code review)
- [Summary of findings]
## Stack-Specific (from code review)
- [Language/framework-specific findings]
## Business Logic
- [Your findings]
## Test Coverage
- **Unit:** [Present/Missing — what's covered, what's not]
- **Integration:** [Present/Missing/N/A — what's covered]
- **E2E:** [Present/Missing/N/A — what's covered]
- **Verdict:** [Adequate / Needs more coverage at X layer]
## Suggestions
- [Improvements, not blockers]