What problem does it solve? Reviewing a teammate's test-automation pull request fairly requires knowing the repo's actual documented conventions, reading the real diff instead of the PR description, and separating genuine defects from stylistic deviations — a process that is easy to get wrong when done from memory or generic best practices. ## Core Features & Use Cases - Doctrine-grounded review: Loads the target repo's own AGENTS.md and KATA doctrine files (or probes an external repo via gh api) so every finding cites a concrete file:line or doctrine section, never an unverifiable claim. - Calibrated strictness and scoring: Runs a Flexible/Standard/Strict preflight, buckets findings into Real/Reliability, Pattern/Doctrine-deviation, and Positive categories, and produces a defensible score out of 10. - Interactive triage and safe posting: Presents findings for user triage, drafts praise-constructive-praise feedback, and posts to GitHub only after explicit confirmation. - Use Case: A QA lead asks to review a colleague's PR in an external repo (owner/repo#42); the skill probes that repo for its own doctrine, fetches per-file patches via gh api, flags a hardcoded credential as Critical with its file:line citation, notes a KATA pattern deviation as a Standard-level observation, and posts the approved feedback comment. ## Quick Start Ask the assistant to act as a QA lead and review pull request number 42 in owner/repo at Standard strictness, then triage the findings before posting any comment.
