What problem does it solve? Code reviews performed in the same session that wrote the code suffer from author bias, causing real bugs and missed changes to slip through. This Skill structurally removes that bias by delegating review to fresh-context subagents that never saw the conversation that produced the code. ## Core Features & Use Cases - Five-bucket parallel review: Dispatches concurrent reviewer subagents covering correctness, data flow, performance/security/observability, missed changes, and project-rule violations. - Coverage-then-precision pipeline: Subagents over-report findings with severity and confidence tags; the main session verifies, deduplicates, and severity-orders every finding before presenting it. - Multi-runtime support: Works from Claude Code via Agent tool calls or from Codex via parallel codex exec processes, with an opt-in strict maintainability rubric for deep structural audits. - Use Case: After landing a large feature branch, ask for an impartial review of the PR diff and receive a verified, severity-ranked findings list with blocking issues, dismissed false positives, and a merge recommendation. ## Quick Start Ask the assistant to impartially review the most recent commit or the current uncommitted changes using fresh reviewer subagents.
