work-with-pr — Full PR Lifecycle

What it does

Orchestrates the whole path from "here's a task" to "it's merged":

• Phase 0 — Setup: decomposes one task into the smallest atomic, independently mergeable PRs, then creates a branch and a sibling git worktree per PR. Independent slices are built concurrently via background subagents or a team, so the user's main working tree is never touched.
• Phase 1 — Implement: all coding flows through the ulw-loop skill, which forces every success criterion to be proven by evidence-bound manual QA written to disk (tmux / HTTP / browser / GUI). No evidence file → no commit, no push. Includes minimum-commit ratios based on files changed.
• Phase 2 — PR creation: a reviewer-friendly English PR body template (Summary / Changes / QA & Evidence / Risks & Residuals / Related Issues), plus guardrails against pasting secret-bearing logs or committing temp screenshots.
• Phase 3 — Verification loop: unbounded iteration over Gate A (gh pr checks) and Gate B (the cubic-dev-ai[bot] review). Any failing gate routes back to Phase 1 for a scoped fix and fresh QA; Cubic may only be SKIPPED when its quota is exhausted, never because it found issues.
• Phase 4 — Merge & cleanup: arms gh pr merge --merge --auto, blocks until the PR is actually MERGED, syncs .omo state back to the main repo, then removes and prunes the worktree. A failure-recovery section deliberately keeps the worktree alive when something breaks.

Who it's for

• Team leads who want agent-written code to land as reviewable PRs, not ad-hoc commits
• Developers pushing an atomic-PR culture (a 200-line PR gets a real review; a 2000-line one gets rubber-stamped)
• Anyone tired of babysitting the red-CI → fix → re-push cycle
• Repos already on GitHub CLI, GitHub Actions, and an automated review bot

Example uses

1. "Implement this issue and open a PR" — the issue is split into three atomic PRs; two independent ones build in parallel worktrees, each running its own CI loop and auto-merging when green.
2. "CI is failing, fix it" — pulls failed logs with gh run view --log-failed, fixes only the identified cause as an atomic commit, re-QAs if behavior changed, pushes, and restarts verification from Gate A.
3. "Address the review bot's findings" — parses the Cubic review body, separates real issues from false positives, fixes the real ones, and waits for a fresh review until it reports "No issues found".

Caveat: the doc hard-codes a dev base branch, bun commands, the Cubic bot, .omo state and a companion ulw-loop skill. Adapt those to your own stack before relying on it.

Install the skill I found on Claude Skill Mart.
Copy the .agents/skills/work-with-pr folder from the GitHub repo code-yeongyu/oh-my-openagent into my ~/.claude/skills/work-with-pr/.
When it's done, tell me in one line what this skill can do.