Resolve PR review bot findings on current PR. Fetches unanswered bot comments, evaluates each finding, fixes real bugs, dismisses false positives, replies to every comment, and watches for new findings until bots go quiet.

Automatically resolve findings from PR review bots (Copilot, Cursor Bugbot, CodeRabbit, etc.) on the current PR. Uses a two phase workflow: fix all existing issues, then poll for new ones until bots go quiet. Prerequisites All commands below use npx agent reviews . If the project uses a different package manager, substitute the appropriate runner (e.g., pnpm dlx agent reviews for pnpm, yarn dlx agent reviews for Yarn, bunx agent reviews for Bun). Honor the user's package manager preference throughout. Cloud environments only (e.g., Codespaces, remote agents): verify git author identity so CI checks can map commits to the user. Run git config global get user.email and if empty or a placeholder, set it manually. Skip this check in local environments. Phase 1: FETCH & FIX (synchronous) Step 1: Fetch All Bot Comments (Expanded) Run npx agent reviews bots only unanswered expanded The CLI auto detects the current branch, finds the associated PR, and authenticates via gh CLI or environment variables. If anything fails (no token, no PR, CLI not installed), it exits with a clear error message. This shows only unanswered bot comments with full detail: complete comment body (no truncation), diff hunk (code context), and all replies. Each comment shows its ID in brackets (e.g., [12345678] ). If zero comments are returned, print "No unanswered bot comments found" and skip to Phase 2. Step 3: Process Each Unanswered Comment For each comment from the expanded output: A. Evaluate the Finding Read the referenced code and determine: 1. TRUE POSITIVE A real bug that needs fixing 2. FALSE POSITIVE Not actually a bug (intentional behavior, bot misunderstanding) 3. UNCERTAIN Not sure; ask the user Likely TRUE POSITIVE: Code obviously violates stated behavior Missing null checks on potentially undefined values Type mismatches or incorrect function signatures Logic errors in conditionals Missing error handling for documented failure cases Likely FALSE POSITIVE: Bot doesn't understand the framework/library patterns Code is intentionally structured that way (with comments explaining why) Bot is flagging style preferences, not bugs The "bug" is actually a feature or intentional behavior Bot misread the code flow When UNCERTAIN ask the user: The fix would require architectural changes You're genuinely unsure if the behavior is intentional The "bug" relates to business logic you don't fully understand Multiple valid interpretations exist The fix could have unintended side effects B. Act on Evaluation If TRUE POSITIVE: Fix the code. Track the comment ID and a brief description of the fix. If FALSE POSITIVE: Do NOT change the code. Track the comment ID and the reason it's not a real bug. If UNCERTAIN: Ask the user. If they say skip, track it as skipped. Do NOT reply to comments yet. Replies happen after the commit (Step 5). Step 4: Commit and Push After evaluating and fixing ALL unanswered comments: 1. Run your project's lint and type check 2. Stage, commit, and push: 3. Capture the commit hash from the output. Step 5: Reply to All Comments Before posting to an inline review comment, fetch npx agent reviews detail <comment id and inspect the latest replies. PR level issue comments and review summaries have no threaded reply list; an empty list for these types is not evidence that no outcome was recorded. The CLI skips replies to those types. Record that skip and do not create a replacement top level comment or use another tool to bypass it. If the same outcome is already recorded, do not post it again. On a resumed session, use the fresh unanswered list instead of replaying a prior session's work queue. Revisit a decision only when new discussion or a user request calls for it. For multiline or complex replies, write the exact reply text to a uniquely named UTF 8 file in a temporary directory outside the repository and use npx agent reviews reply <comment id body file <path to reply file (adding resolve only where appropriate below). The CLI accepts absolute and relative paths; prefer an absolute path for the outside repository temporary file. Use the file path directly, without shell command substitution. Do not also pass a positional message. Remove the temporary file after the reply succeeds; keep reply files outside the repository even if posting fails so later git add A commands cannot commit them. Now that the commit hash exists, reply to every processed comment. The resolve flag marks the review thread as resolved on GitHub. Use it only when closing the conversation (false positives and user skipped findings), not for fixes that reviewers should still verify. For each TRUE POSITIVE (we fixed it; leave the thread open so reviewers/bots can verify): Run npx agent reviews reply <comment id "Fixed in {hash}. {Brief description of the fix}" For each FALSE POSITIVE (closing the conversation): Run npx agent reviews reply <comment id "Won't fix: {reason}. {Explanation of why this is intentional or not applicable}" resolve For each SKIPPED (user chose not to address it): Run npx agent reviews reply <comment id "Skipped per user request" resolve Non threadable comments: If a finding was posted as a PR level issue comment or review summary submission (not an inline review comment on a specific line), the CLI cannot post a threaded reply and will print ⊘ Skipped reply: comment is not part of a review thread . That's expected. Record the outcome in the summary and move on; do not retry. DO NOT start Phase 2 until all replies are posted. Phase 2: WAIT FOR THE HEAD COMMIT'S CHECKS (loop until clean) Most review bots (Greptile, Cursor Bugbot, CodeRabbit and others) report as a check on the commit, beside CI, so a review is finished when the commit's checks are. wait checks waits for exactly that, then prints only what needs you. Step 6: Wait, then act on the report Repeat until a pass exits 0: 6a. Run npx agent reviews wait checks bots only unanswered as a background task, and carry on with other work until it finishes. Never poll with gh pr checks watch , sleep loops or repeated listings. Add expect "<check name " for each reviewer that registers its check late (for example one that starts only after a preview deploys), so the wait does not end before it reports. Expected checks are not waited for on a draft PR, since review bots usually skip drafts. 6b. Read the report. The first line names the commit and the outcome; below it are failed checks with links, expected checks that never reported, unanswered inline comments, and PR level comments. PR level comments (issue comments, review summaries) have no thread to reply in, so they are listed on every run: process only IDs you have not handled in this run. The exit code says what to do: 0 : every check passed and no comment waits. Move to the Summary Report. 1 : a check failed, an expected check never reported, or inline comments wait. Process comments exactly as in Phase 1, Steps 3 5 (evaluate, fix, commit, push, reply). A failed check is a finding too: read its log ( gh run view <id log failed for GitHub Actions; for a review bot or an external status, open its link), fix it, commit and push. Then go back to Step 6a . If the only problem is an expected check that never reported, do not wait again: name it in the Summary Report and stop. 2 : the PR conflicts with its base branch, so CI will not run. Merge or rebase the base branch, push, and go back to Step 6a. 3 : timed out with checks still running (default 30 minutes; timeout <seconds ). Report which checks are pending. 4 : no check registered on the commit. Report it; the repository may have no CI for this branch. If you push while it waits, it follows the new head commit on its own. Round cap. Each fix push in this phase is a round, whether its findings came from 6b or from the comment watcher below. If new findings keep landing on lines your own earlier fix commits changed, each fix is opening the next edge case, and more patches will not converge. After the second such round, stop: do not push another fix. Report the remaining findings to the user together with what each fix round changed, and suggest reviewing the design as a whole (for example with an adversarial design review) before continuing. Findings on code the earlier fixes did not touch do not count toward the cap. A reviewer that posts without a check (Copilot, for example) is the one case for the comment watcher. After a clean wait checks , run npx agent reviews watch bots only unanswered timeout 180 once. If it returns findings, process them as in Phase 1, Steps 3 5, and go back to Step 6a , so the new commit's checks are waited for too. Summary Report After both phases complete, provide a summary: Important Notes Response Policy Every finding gets a response No silent ignores Responses help train bots and document decisions "Won't fix" responses prevent the same false positive from being re raised User Interaction Ask the user when uncertain about a finding Don't guess on architectural or business logic questions It's better to ask than to make a wrong fix or wrong dismissal Best Practices Verify findings before fixing bots have false positives Keep fixes minimal and focused don't refactor unrelated code Ensure type check and lint pass before committing Group related fixes into a single commit Copilot suggestion blocks often contain ready to use fixes

resolve-agent-reviews