← ClaudeAtlas

autonomous-pr-driverlisted

Autonomously drive a pull request to merge-ready — opening or attaching to it, then resolving automated code review (triage findings, fix the valid, reject the invalid, push, loop) and pinging a human to merge. Knows when to STOP: at diminishing returns (niche/trivial/contradictory findings, severity trending down, or review budget accruing) it declares the PR good-to-merge on substance and pauses rather than auto-looping to chase a bot to zero comments — resuming only if the user insists or a genuinely important finding appears. Use when asked to 'drive / ship / land this PR', 'get the PR green', 'resolve the PR review comments', 'address the CodeRabbit / Cursor / Bugbot / Codex findings', 'fix the code review and push', 'stop over-fixing / merge it', or to loop on PR reviews until checks pass.
stealth-engine/skills · ★ 2 · AI & Automation · score 76
Install: claude install-skill stealth-engine/skills
# Autonomous PR driver Drive a PR from change → merge-ready, resolving automated code review along the way. The hard part isn't the git mechanics — it's **judging** a stream of bot findings (some real, some stale, some wrong, some contradictory) without thrashing. This skill is the playbook for that. Deep detail lives in siblings (load on demand): [`reference/triage-playbook.md`](./reference/triage-playbook.md) (decision rules + `gh` recipes) and [`reference/known-bots.md`](./reference/known-bots.md) (per-bot cadence + @-tag behaviour snapshot). ## The loop ```text 1. OPEN/ATTACH → 2. WATCH checks → 3. RESOLVE reviews → 4. CONVERGE? ──no──┐ ▲ │ └──────── push batch ◀─────────────┘ yes → 5. HAND OFF ``` 1. **Open or attach.** If asked to ship a change: branch (repo convention — see `AGENTS.md`/`CONTRIBUTING`), commit, push, open a PR with a **Conventional Commit** title (it becomes the squash commit). If a PR already exists for the branch, **attach to it** and continue from step 2 (watch checks before triaging). 2. **Watch checks.** Poll until checks **settle** — don't triage mid-run. "Settled" = no pending checks *except* human-gated approvers (e.g. a "PR approver" agent that waits for a human). See the playbook's poll recipe. 3. **Resolve reviews.** - Enumerate **every open