| name | cross-review |
| description | Run code reviews in parallel using Claude Opus and GitHub Copilot (codex), then produce a unified report. Use when the user requests a cross-review, xreview, multi-LLM review, or wants diverse perspectives on code changes. Accepts optional PR number or file paths as arguments. |
| allowed-tools | ["Agent","Bash(cat *)","Bash(gh pr diff *)","Bash(git diff *)","Read"] |
Cross Review
Run two independent code reviews in parallel and merge results into a unified report.
Arguments
| Argument | Required | Description |
|---|
| PR number | no | Pull request number (e.g. #42) via gh pr diff |
| File paths | no | Scope review to specific files |
No arguments: review current uncommitted diff (fallback to last commit if clean).
Workflow
Step 1: Get the diff
- PR number given:
gh pr diff <number>
- File paths given:
git diff HEAD -- <paths>
- Otherwise:
git diff HEAD (if empty, git diff HEAD~1)
If no diff found, ask the user.
Step 2: Launch parallel reviews
Both reviews MUST be launched in a single message with run_in_background: true.
Review A -- Claude Opus:
Launch a code-reviewer subagent with model: opus. Include full diff, request review of correctness, security, performance, maintainability with file:line fix suggestions.
Review B -- GitHub Copilot codex:
Pipe a review prompt via heredoc to the bundled script. Add --lang <language> for non-English codebases.
cat <<'EOF' | bash <skill_path>/scripts/run_copilot.sh
...review prompt with diff...
EOF
After launching both, inform the user that reviews are running.
Step 3: Merge and report
Wait for both results. If one reviewer fails (timeout, crash, empty output), produce the report from the successful reviewer alone and note which reviewer failed and why. If both fail, report the errors to the user.
Read references/report-format.md for the merge strategy and report template, then produce the unified report.