Merge open prs
Skill emaraschio/cursor-commands/plugin/.cursor/skill-contracts/merge-open-prs
Production-grade Cursor slash commands with paired skill contracts, behavioral evals, and ship-gate CI. Install as a user plugin.
npx -y skills add emaraschio/cursor-commands --skill merge-open-prsAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 9 stars9 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Review, verify (Docker by default), and auto-merge open PRs when CI, reviews, and local verification pass. Batch up to 10 PRs per run. Use when the user wants to merge all open PRs, clear the PR queue, or run /merge-open-prs.
SKILL.md
8.2 KB, ~2.1k tokens by cl100k_base, as published. Nobody here has run it
Merge open PRs
Orchestrate a batch of open pull requests in the current repo: babysit each PR, code-review, verify locally (Docker-first), auto-merge when green, then post-batch smoke on the default branch.
Do not duplicate the babysit contract inline. Read and follow the babysit skill at ~/.cursor/skills-cursor/babysit/SKILL.md for every PR in the batch.
Entry point: .cursor/commands/merge-open-prs.md (slash /merge-open-prs). When this file disagrees with the command file, this file wins.
Defaults
Apply these when the user does not override:
| Setting | Default | Override |
|---|---|---|
| Verification | Docker-first | --no-docker |
| Batch size | 10 PRs | --limit N |
| Autonomy | auto_if_green (merge when gate passes) | --dry-run (no merge) |
| Single PR | Process only that number | 42 or #42 argument |
Parse flags from the user message: --dry-run, --no-docker, --limit N. A bare number selects one PR.
When NOT to use
- Single PR babysit only (no batch, no queue) → use babysit on that PR directly
- Cross-repo initiative status → use cross-repo-status or manual
ghsweep per user rules - PRs needing product/legal sign-off → skip; report in summary
- Sensitive security changes without explicit user approval → hard stop
- Repo has no Docker and user did not pass
--no-docker→ blocker at preflight (do not silently fall back)
Phase 0: Preflight
gh auth status: blocker if not authenticated- Confirm git repo root:
git remote get-url origin git fetch --all --prune- If working tree dirty → warn; do not stash without asking
- Docker (default path):
docker info. If it fails and user did not pass--no-docker, stop with a blocker (suggest--no-dockeronly when they accept non-Docker verification) - Resolve profile path:
profiles/<repo-basename>.yamlrelative to this skill directory (basenamefromgit remote get-url origin, strip.git, last path segment). Seeprofiles/README.md.
Phase 1: Inventory
gh pr list --state open --limit <LIMIT> --json number,title,isDraft,mergeable,reviewDecision,statusCheckRollup,headRefName,author,labels,createdAt,updatedAt
-
<LIMIT>= user--limitor 10 default. Fetch at least that many; if more open PRs exist, note deferred count in the plan table. -
Skip (do not process):
isDraft, labels in profileskip_labelsor defaultdo-not-merge,wip -
Order:
createdAtascending (FIFO). If user passed a PR number, only that PR (still full verify + gate) -
Emit a plan table before acting:
# Title In batch? Skip reason
Phase 2: Per-PR loop
For each PR in batch (in order):
2a: Babysit
Read and execute ~/.cursor/skills-cursor/babysit/SKILL.md:
- Resolve merge conflicts (abort if intent conflicts; ask user)
- Triage unresolved comments (filter resolved threads; validate Bugbot)
- Fix CI failures in PR scope only; never weaken workflows
Checkout the PR branch: gh pr checkout <n>.
2b: Code review
Apply the checklist in ~/.cursor/skill-contracts/code-review/SKILL.md (or slash /code-review). Surface blockers in chat. Hard stops (entire batch):
.env, credentials, API keys in diff- DB migration without rollback note in PR description
- Breaking change (
feat!,BREAKING) without explicit user approval in this session
2c: Local verification
Docker-first (default; unless --no-docker)
- If profile exists → run
verify.stepsin order (each step is a shell command) - Else if
docker-compose.yml,compose.yaml, orcompose.ymlexists:docker compose build- Run tests via profile-less heuristic: read repo
README/Makefilefor the canonical test service; preferdocker compose run --rm <service> <test-cmd>over inventing commands
- Else → skip PR with reason
no-docker-profile-or-compose; do not auto-merge
Never run docker compose down -v unless the user explicitly requests volume teardown.
Host-only (--no-docker)
- Profile
verify.mode: hoststeps if present - Else heuristics:
make test,bin/test,npm test,pnpm test,bundle exec rspec(first that exists) - If nothing found → skip PR with reason
no-verify-path
Record pass/fail output in the session summary.
2d: Re-fetch GitHub state
gh pr view <n> --json mergeable,mergeStateStatus,reviewDecision,isDraft
gh pr checks <n>
Unresolved review threads: use gh api GraphQL or review threads; follow babysit (human threads block; bots per babysit judgment).
Phase 3: auto_if_green gate
Merge only if all are true:
| Check | Requirement |
|---|---|
| Mergeable | mergeable == MERGEABLE / mergeStateStatus not blocked |
| Reviews | reviewDecision ≠ CHANGES_REQUESTED |
| CI | All required checks SUCCESS (gh pr checks) |
| Threads | No unresolved human review threads |
| Local verify | Passed in 2c |
| Draft | isDraft == false |
- Pass → Phase 4a to 4b (unless
--dry-run) - Fail → log reason, continue to next PR (do not stop batch unless auth/rate-limit/hard-stop)
Do not approve before the gate passes. Approval is a pre-merge step, not a substitute for local verify or CI.
--dry-run
Evaluate gate and report would approve / would merge / would skip; never call gh pr merge or gh pr review --approve.
Phase 4a: Approve (always before merge)
After gate passes (and not in --dry-run):
gh pr view <n> --json reviewDecision,author: confirm the authenticatedghuser can act as reviewer on this repo- If you have not already left an APPROVE review on this PR in this session:
gh pr review <n> --approvewith an optional one-line body summarizing verify + CI status
- Re-fetch
reviewDecision. If still blocked (e.g. requires another reviewer), skip merge with reasonapproval-insufficient; continue queue - Log approval in the per-PR summary before proceeding
Never approve when gate failed, local verify failed, or hard-stop conditions apply.
Phase 4b: Merge (only after 4a)
Merge only after Phase 4a succeeded (approve recorded or already approved by you):
# squash example
gh pr merge <n> --squash --delete-branch
Use profile merge (squash | merge | rebase) or repo default.
Never:
gh pr merge --admin--no-verifyon git operations- Force-push to default branch
- Merge with failing required checks
Phase 5: Post-batch smoke
After all PRs processed:
git checkout <default_branch>: from profile orgh repo view --json defaultBranchRefgit pull- Run
post_merge_smokefrom profile, or repeat a lighter Docker smoke (e.g. same as verify step 1 only) when Docker-default - Summary table: merged | skipped (reason) | failed | deferred (beyond limit) | smoke pass/fail
Profiles
Optional YAML at profiles/<repo-basename>.yaml. Copy from profiles/_template.yaml. Profile overrides detection heuristics.
Behavioral evaluation
Before changing this contract materially, walk eval/cases.md sections A, D, E (minimum). Target: 0 FAIL, ≥90% adjusted pass rate. See eval/README.md.
Lessons (maintainers)
- Docker default matches repos with a merge profile;
--no-dockeris explicit opt-out - Limit 10 prevents runaway merges; deferred PRs must appear in summary
- Babysit is source of truth for per-PR triage; do not fork its rules here
Guardrails
- Merge only when the gate passes: require local verification, CI success, no CHANGES_REQUESTED, and resolved human threads before merging.
- Do not silently skip Docker; if
docker infofails without--no-docker, stop with a blocker. - Keep the batch bounded: process up to
--limit(default 10) FIFO and report deferred PRs.
What ships with it: 5 files
9.2 KB alongside SKILL.md
eval/
- cases.md6.6 KB
- fixtures.yaml857 B
- README.md318 B
profiles/
- README.md868 B
- _template.yaml580 B
Gives 0 of the 12 instructions most quality gates skills give in ~2.1k tokens
Counted across 1,195 of the 2,094 authors here whose files we hold, read 2026-08-07
- Read the output and check the exit codein 54 of 1195, across 14 files
- Verify requirements using a line-by-line checklistin 53 of 1195, across 12 files
- Identify the verification command proving the claimin 51 of 1195, across 12 files
- Run the full verification commandin 50 of 1195, across 11 files
- Verify output confirms the claimin 49 of 1195, across 12 files
- Check version control diff after agent delegationin 46 of 1195, across 6 files
- State claim with evidencein 44 of 1195, across 4 files
- Run the test suitein 33 of 1195, across 26 files
- Keep state in memory by defaultin 27 of 1195, across 6 files
- Make prototype runnable with one commandin 26 of 1195, across 5 files
- Produce a verification reportin 25 of 1195, across 14 files
- Detect the package manager from lockfilesin 24 of 1195, across 5 files
Said here and by no other author read
- process up to ten pull requests per batch
- fetch open pull requests in FIFO order
- skip draft pull requests or blocked labels
- read and execute the babysit skill per PR
- fix CI failures within the pull request scope only
- run local verification via Docker by default
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.