Deep Review: Multi-Perspective Code Review

Overview

This skill orchestrates a multi-perspective code review using five specialized review agents. Each agent analyzes code from a distinct angle (correctness, architecture, security, production readiness, test quality). Findings are merged, deduplicated, and classified by severity. Critical and Important findings trigger an autonomous fix loop (up to 3 rounds). Results are documented in review-findings.md.

This skill is invoked by spex:review-code when the deep-review trait is enabled.

Prerequisites

The caller (review-code or ship) may provide these values. When not provided, the deep-review skill resolves them itself:

1. Stage 1 result: spec compliance score (or null if no spec)
2. Invocation context: superpowers or manual
3. Hint text: optional focus area from user (or null)
4. External tool settings: {coderabbit: true/false, copilot: true/false} (see resolution below)
5. Spec path: path to spec.md (or null)
6. Feature directory: path to the spec directory for artifact output