From dev
Stateful PR poller: tracks GitHub Actions CI, CodeRabbit, Codecov, and human reviews across polls. Loops with adaptive backoff internally. Run via the Bash tool with run_in_background: true so the platform notifies the agent on exit.
How this skill is triggered — by the user, by Claude, or both
Slash command
/dev:poll PR_NUMBERPR_NUMBERThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Poll PR #$ARGUMENTS for CI, CodeRabbit, and human review status, then fix what's broken.
Poll PR #$ARGUMENTS for CI, CodeRabbit, and human review status, then fix what's broken.
The script manages its own state in .claude/poll-state/pr-$ARGUMENTS.json and loops
internally with adaptive backoff (120-600s). It shows a full report on the first run,
then delta-only on subsequent polls so only new events surface.
Run the script via the Bash tool with run_in_background: true. Never use & to
background it in the shell — that detaches the process and makes the exit code invisible.
# Bash tool call — set run_in_background: true
bash .claude/scripts/poll-pr.sh $ARGUMENTS
The script sleeps and polls on its own. Do not interrupt it. When it exits, you will be notified automatically.
| Code | Meaning | Action |
|---|---|---|
| 0 | All checks pass | Ready to merge |
| 1 | CI failures | Fix code, push, re-run (background again) |
| 2 | Checks pending | Should not exit 2 unless --max-polls hit; re-run |
| 3 | Inline review comments | Address comments, push, re-run (background again) |
| 4 | Awaiting human review | Reviewer team notified; CronCreate to re-poll later |
When the script exits 4, it has already posted a comment and requested reviewers. Use CronCreate to check back automatically:
CronCreate: every 2 hours, /poll $ARGUMENTS
Delete the cron (CronDelete) once the PR is approved or merged.
Run the relevant test/lint commands per the project's AGENTS.md to reproduce
and fix failures locally. After fixing: push, then re-run the script via Bash
with run_in_background: true.
Run --once --full to see comments with reply/resolve commands:
bash .claude/scripts/poll-pr.sh $ARGUMENTS --once --full
Evaluate each comment critically — automated reviewers are often right but not always. Read the full context, understand what it's flagging, and make a genuine judgment call:
Steps:
run_in_background: truegit add -A && git commit -m "type(scope): address feedback"
git push origin <branch>
# then re-run poll via Bash tool with run_in_background: true
Repeat until exit 0.
npx claudepluginhub quay/ai-helpers --plugin devProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.