What problem does it solve? Pull requests authored by an AI agent need an independent reviewer that does not trust the author's claims, and teams need a consistent, evidence-based gate before code lands on the main branch. ## Core Features & Use Cases - Contextless adversarial review: A fresh reviewer agent reads only the PR, linked issue, and diff, then works a ten-point attack list covering spec fidelity, silent failures, concurrency, security, and style-guide rules. - Full verification pipeline: Checks CI status, runs unit and integration tests, enforces coverage thresholds (80% or no regression), scans for secrets, and walks the PR's suggested test steps. - Structured verdicts and merge: Posts Changes Requested, Decomposition Requested, Approved, or Escalation comment templates, files deferred issues for non-blocking findings, and squash-merges only after a clean review. - Use Case: After an agent opens a PR closing issue #42, dispatch this skill in a subagent to independently verify the fix, run the test suite, request changes on findings, and squash-merge once the review is clean. ## Quick Start Ask the agent to review pull request number 42 using the github-pr-review skill and merge it only if the review comes back clean.
