What problem does it solve? When finishing a major feature or preparing a PR, a single-pass code review cannot tell whether you built the right thing (spec compliance) versus whether you built it right (code quality). This Skill separates those two concerns into sequential gated stages so spec failures surface before quality review begins. ## Core Features & Use Cases - Stage 1 Spec Compliance: Loads the intent contract from .claude/session-intent.md, checks every success criterion and boundary, and marks each as PASS, FAIL, PARTIAL, RESPECTED, or VIOLATED. - Stage 2 Code Quality: Runs five stub-detection checks (TODO markers, empty functions, null returns, low substantive line counts, mock data) plus a full quality review covering architecture, error handling, security, performance, readability, and test coverage. - Multi-Provider Review: Dispatches the diff to external providers like Codex and Antigravity via the Octopus orchestration script for parallel logic and security analysis, then synthesizes findings. - PR Integration: Detects an open PR on the current branch with the GitHub CLI and posts the combined staged review report as a comment. - Use Case: After completing a feature defined by an intent contract, run the staged review to confirm all success criteria pass, catch leftover stubs, and publish a unified verdict before merging. ## Quick Start Ask the AI to run a staged review of the current branch against the session intent contract and post the combined report to the open pull request.
