What problem does it solve? Single-reviewer code reviews miss defects because the agent that wrote the code shares the same blind spots as the one reviewing it. This Skill fires several independent reviewers that are blind to each other, plus a different model family (Codex), then reconciles their findings into one verdict before you merge. ## Core Features & Use Cases - Blind parallel panel: Launches adversarial, reality-check, test-coverage, security, and spec-alignment lenses in parallel, each reviewing the identical diff without seeing other reviewers' findings. - Cross-model jury vote: Runs the Codex CLI as a different-model-family reviewer, with a labeled same-family fallback when Codex is unavailable, and weights convergent findings most heavily. - Risk-tiered depth and triage: Scales review depth by change sensitivity (auth, secrets, billing, migrations, public APIs, infra), dedupes findings, and sorts them into fix-now, nit, and owner-decides buckets with a SHIP / FIX-FIRST / RETHINK verdict. - Use Case: Before merging a branch that touches authentication, run the panel to get a security-mandatory review, a list of must-fix findings with file:line references, and a verdict that still requires your explicit sign-off. ## Quick Start Ask the agent to run a deep review of the current branch against its base before merging, optionally with --fix to apply clear localized fixes or --pr <number> to review a GitHub pull request.