deepreview
Installation
SKILL.md
Deep Review
Multi-stage code review pipeline using parallel subagents. The entire review runs outside the main context. A gatherer agent collects context and runs static analysis, six reviewer agents execute parallel review passes, a verifier agent checks critical findings, then the gatherer formats the final report.
Languages: Go, Rust, TypeScript, Svelte, Astro.
Pipeline
The main session is the orchestrator. It does NOT run any review stages itself — it dispatches all work to subagents and synthesizes results.
Agent Roles
| Role | Count | Work | Model guidance |
|---|---|---|---|
| gatherer | 1 | Stages 1+2 (context gathering, static analysis), Stage 5 (final formatting) | Fast/cheap model (e.g. Sonnet, GPT-4o-mini) |
| reviewer | 6 | Stage 3 (one review pass each) | Strong reasoning model (e.g. Opus, GPT-5, o3) |
| verifier | 1 | Stage 4 (verify CRITICAL/HIGH findings) | Strong reasoning model |