pr-review

name	pr:review
description	Review a pull request for code quality, test coverage, and architectural alignment. Use when asked to review a PR, check code quality, provide feedback on changes, or audit a diff. Do NOT use for creating PRs (use pr:create) or for merging (use pr:merge).
argument-hint	<pr-number> [--focus security|performance|correctness] [--brief]
allowed-tools	["Read","Bash","Glob"]
user-invocable	true

Pull Request Review

Perform a structured code review on a GitHub pull request, covering correctness, test coverage, security, performance, and architectural fit.

Why This Skill Exists

Code reviews catch bugs before they reach production, but review quality varies wildly. Junior reviewers miss security issues; senior reviewers skip test coverage. This skill applies a consistent checklist so nothing falls through the cracks, while leaving subjective design decisions to the human reviewer.

Procedure

Step 1: Fetch the Diff

gh pr diff ${pr_number}

Read the diff. If --brief is set, summarize the changes in 3-5 sentences and skip to Step 5.

Gotcha: Large diffs (>2000 lines) should be reviewed file-by-file. If the diff exceeds this threshold, switch to per-file review mode automatically and note this in the output.

Step 2: Check for Common Issues

Review each changed file for:

Correctness:

• Off-by-one errors in loops and slices
• Nil/null pointer dereference risks
• Resource leaks (unclosed files, connections, channels)
• Race conditions in concurrent code

Security (prioritize if --focus security):

• SQL injection via string concatenation
• Unsanitized user input in templates or commands
• Hardcoded secrets or credentials
• Missing authentication/authorization checks

Performance (prioritize if --focus performance):

• N+1 query patterns
• Unbounded collection growth
• Missing pagination on list endpoints
• Expensive operations inside loops

Step 3: Verify Test Coverage

Check if the PR includes tests for:

• New public functions/methods
• Changed behavior (not just changed code)
• Error paths and edge cases

gh pr checks ${pr_number}

Note any failing CI checks.

Step 4: Assess Architectural Fit

Compare the changes against:

• Existing patterns in the same package/module
• The project's documented architecture (if available in docs/ or ARCHITECTURE.md)
• Separation of concerns (is business logic leaking into handlers?)

Step 5: Generate Review

Output a structured review:

## PR #${pr_number} Review

### Summary
<2-3 sentence summary of what the PR does>

### Findings

| Severity | File | Line | Issue |
|----------|------|------|-------|
| ...      | ...  | ...  | ...   |

### Test Coverage
<Assessment of test adequacy>

### Recommendation
<APPROVE | REQUEST_CHANGES | COMMENT>
<Brief justification>

Why this format? The table makes it easy to address findings one by one. The recommendation gives a clear signal. The summary helps other reviewers who arrive late.

Error Handling

Condition	Action
PR number not found	Stop: "PR #${pr_number} not found. Check the number and repo."
No diff available	Stop: "PR has no changes or is already merged."
CI checks still running	Note in review: "CI checks in progress, results may change."
Diff too large (>2000 lines)	Switch to per-file mode, note in output

Known Limitations

• Cannot assess runtime behavior or performance regressions (use profiling tools for that)
• Architecture assessment depends on documented patterns; undocumented conventions may be missed
• Security review is pattern-based, not a substitute for SAST tools like Semgrep or CodeQL