What problem does it solve? Deep codebase reviews often produce shallow, hallucinated, or unverifiable findings. This Skill enforces an evidence-gated review workflow where every claim must cite exact file paths, line numbers, and verbatim quotes, and no report ships until all coverage units are closed and a final critic pass approves the synthesis. ## Core Features & Use Cases - Phase 0 Inventory: Maps directories, entry points, manifests, dependencies, claims, public surfaces, trust boundaries, tests, UI, and AI/MCP surfaces before any review begins. - Selectable Review Tracks: Supports focused or combined tracks for functionality, security and supply chain, testing, UI/accessibility, performance/observability, AI-slop detection, and enhancement catalogs, with non-diluting depth per track. - Validation Pipeline: Explorer candidates are filtered by reviewer validation, inline critics for CRITICAL/HIGH defects, enhancement critics, and a mandatory final whole-report critic, with artifacts written to .swarm/review-v8 runs. - Use Case: Ask for a security and supply-chain audit of a repository; the skill inventories the codebase, checks dependencies against SLSA v1.2 and OWASP ASVS 5.0.0 baselines, validates each finding with quoted evidence, and produces a structured review report without modifying source files. ## Quick Start Ask the agent to run $codebase-review-swarm at the repository root to perform a full evidence-grounded codebase review.
