What problem does it solve? Automated PR reviewers (Codex, Copilot) generate findings that are sometimes wrong, duplicated, or only partially fixed, and branches merging to main risk silently reverting newer main-side changes. This Skill provides a disciplined playbook for verifying each finding against the actual code, sweeping entire defect classes instead of patching single lines, and closing every review thread before merge. ## Core Features & Use Cases - Finding triage and verification: Classify each bot comment as real, partial, or false-positive with evidence, then sweep the whole class across the feature rather than fixing only the cited line. - Thread resolution protocol: Reply in-thread with the resolution and commit SHA, react to valid comments, and resolve threads via GraphQL until zero unresolved threads remain. - Pre-merge main comparison audit: Detect branch-side reverts of main content, foreign or stacked commits from composed integration branches, and overlapping file changes using git diff, merge-base, and the check_branch_scope.py script. - Use Case: Before merging a long-running feature branch into a release branch, run the branch-scope check, triage all Copilot and Codex comments into defect classes, fix every instance of each class, and close the round with zero unresolved threads. ## Quick Start Ask the agent to process all open review comments on the current pull request using the audit-review skill, verifying each finding, sweeping its class, and resolving every thread.
