This automates the entire PR review loop: fetches all unanswered comments from bots and humans, evaluates whether each is a real issue or false positive, fixes the legit ones, commits with a single push, then replies to every comment explaining what you did. It runs in two phases: first handles existing comments, then watches for new ones until the PR goes quiet. The evaluation logic is surprisingly thoughtful, distinguishing between bot confusion and actual bugs, and it knows when to ask you instead of guessing on architectural calls. Works through npx/pnpm/yarn with the agent-reviews package. Honestly a huge time saver if you're dealing with noisy bot reviewers or just want to batch-process feedback instead of context-switching all day.
npx skills add https://github.com/pbakaus/agent-reviews --skill resolve-reviews