Code Review

The agent running this skill is the orchestrator. It resolves the review scope, allocates exactly three isolated reviewer executions from the capabilities available in its environment, collects their candidate findings, verifies the evidence, and synthesizes one human-controlled summary.md checklist.

Reviewers may be Claude, Codex, Cursor, repeated isolated sessions of the same provider, or native sub-agents supplied by the host environment. Provider diversity is preferred but not required. Review-only by default: never modify product code.

Commands

• review (default): resolve scope, fan out to all available agents, write one <reviewer-id>-<provider>.md per reviewer, then synthesize summary.md.
• synthesize: only merge existing code_review/<feature>/*.md review files into summary.md. Do not re-run agents.