What problem does it solve? Self-reviewing your own code misses defects because the author already knows the intent. This Skill replaces the built-in /code-review command with a project-scoped implementation that runs independent, fresh-context reviews of PR or working-tree diffs, filters out false positives through adversarial verification, and records every finding as a line-level inline comment on the GitHub pull request. ## Core Features & Use Cases - Parallel perspective finders: Launches independent subagents per review aspect (correctness, security, simplification, tests, documentation consistency, spec fidelity, plus conditional architecture, UI accessibility, and matching-logic scopes) so each verdict is made without contamination from other concerns. - Adversarial verification: Every finding is challenged by a counter-argument agent before being confirmed, eliminating false positives and labeling uncertain items as PLAUSIBLE. - Mandatory inline review posting: Creates a pending GitHub review, attaches per-line comments with severity and recommended fixes, and submits it even when zero findings exist, so review history is always auditable. - Use Case: After opening a pull request, invoke the review to have six-plus independent perspectives examine the diff, receive confirmed defects as inline comments on exact lines, and reply or resolve each thread through the standard review flow. ## Quick Start Ask the AI to review pull request number 42 and post the findings as inline comments on GitHub.
