What problem does it solve? Reviewing pull requests thoroughly requires checking many concerns — architecture, security, test coverage, spec conformance — which is slow and inconsistent when done manually. This Skill automates single-agent PR review across triage-selected domains while keeping posting and code changes behind explicit authorization gates. ## Core Features & Use Cases - Domain-based PR review: Dispatches one LLM agent (claude, codex, or gemini) across selected review domains such as security, architecture, and test coverage, with per-domain defaults configurable in config.json. - Safe-by-default authorization: Reviews are read-only unless the user explicitly passes --post to publish findings and --fix to run the fix loop that edits, tests, commits, and pushes to the PR branch. - Classified findings and fix loop: The TypeScript dispatcher classifies findings as fix, noise, or false positive, optionally fixes medium+ issues, verifies with a trusted test command, and records round history. - Use Case: Run a fast security-and-test-coverage pass on PR #42 with --domains security,test-coverage, review the findings locally, then re-run with --post --fix to publish the review and push verified fixes. ## Quick Start Ask the assistant to review pull request 42 in the current repository using the stark-review skill with default triage-selected domains and no posting.
