codex-cross-tool-reviewer

Codex Cross-Tool Reviewer

Use this skill when Codex is asked to review work written by Claude Code or another agent runtime.

Required Checks

1. Inspect .agent-loop/coordination/ownership.json.
2. Confirm the authoring agent edited only claimed files.
3. Check for active overlapping claims between agent runtimes.
4. Review correctness, security, tests, maintainability, factual drift, README freshness, and writing clarity when relevant.
5. For cross-tool review, run the complete two-round protocol. Codex reviews, Claude addresses comments, Codex reviews again, and Claude addresses the second round. Stop after two rounds unless the user explicitly asks for more.
6. Record each stage with:

& <python> <plugin-root>\scripts\compound_orchestrator.py cross-review --target <project-root> --task-id <task-id> --reviewer-tool codex --author-tool claude --stage round-1-review --summary "<findings summary>"
& <python> <plugin-root>\scripts\compound_orchestrator.py cross-review --target <project-root> --task-id <task-id> --reviewer-tool codex --author-tool claude --stage round-2-review --summary "<findings summary>"

For planning gates, verify that review and response artifacts cover prd.html, planning.html, spec.html, test-cases.html, architecture.html, and users.html.

Review Stance

Lead with findings ordered by severity. If there are no findings, say that clearly and list any residual risk or missing verification.

Do not edit files during cross-review unless the lead integrator explicitly assigns a patch and the intended files have been claimed.