Review Code

Run AI code review, evaluate findings, apply fixes, and verify.

Step 1: Peer Review

Run the /peer-review skill. Pass the appropriate context:

• Uncommitted changes: --uncommitted
• Against a base branch: --base <branch>
• Specific commit: --commit <sha>

Default to reviewing against the repository's default branch (detect via gh repo view --json defaultBranchRef --jq '.defaultBranchRef.name'). If the caller specifies a different base, use that.

Capture the full review output.

Step 2: Evaluate Findings

Run the /evaluate-findings skill on the review output.

If there are additional findings to include (e.g., PR comments passed in by the caller), combine them with the review output before evaluation.

If zero actionable findings survive evaluation, report that the code looks clean and stop.

Step 3: Apply Fixes

Get the full diff and apply each actionable finding directly.

Step 4: Simplify Fixes

Run the /simplify-code skill. The diff command is git diff (the fixes are unstaged).

Step 5: Test and Lint

1. Run the test suite to confirm nothing broke
2. If tests fail, run the /investigate skill to diagnose the root cause, apply the suggested fix, and re-run tests. If investigation cannot identify a root cause, stop and report with investigation findings.
3. Run the project's linter/formatter to ensure clean output

Rules

• If the peer review tool is not available or returns malformed output, report the error and stop.
• Process findings in file order to minimize context switching.