What problem does it solve? Pull requests on PolyKybd repos attract five automated reviewers whose output is not uniformly trustworthy: bots post long walkthroughs without actually reviewing, rate-limit refusals render as green checks, and completed reviews can be pinned to a pre-push commit. This Skill determines which reviewers actually read the current head, verifies every finding against the real code, and drives the feedback to a single resolution reply. ## Core Features & Use Cases - Reviewer verification: Fetches PR comments and reviews via the GitHub MCP tools and distinguishes real reviews from rate-limit notices, summaries, refusals, and stale reviews pinned to old commits. - Finding validation: Reproduces or refutes each bot finding against the actual code before acting, fixing true findings and declining false ones with evidence. - Single consolidated reply: Posts one comment with a verdict table covering every finding, plus a section explaining declined items. - Use Case: After opening a PR and waiting for the bots, ask the assistant to check the PR for review feedback; it will report which of CodeRabbit, Sourcery, Qodo, Greptile, and CodeQL actually reviewed the head commit and address their verified findings. ## Quick Start Ask the assistant to check the pull request for review feedback and address what CodeRabbit, Sourcery, Qodo, Greptile, and CodeQL said.