agentleFS
ZaxbyHubZaxbyHub/opencode-swarm/.agents/skills/commit-pr/SKILL.md

commit-pr

Skill517 starsChanged 3 days agoMITView on GitHub

Mandatory Codex/Copilot publication adapter for opencode-swarm. Use for every GitHub issue assignment that results in code changes, commits, pushes, draft PRs, PR body edits, PR readying, release notes, or CI closeout. Must be loaded before `git push`, `gh pr create`, `gh pr edit`, or `gh pr ready`. Routes to the single canonical source of truth at `.claude/skills/commit-pr/SKILL.md`.

  • Commits and pushes

We split this skill into the rules it gives an agent

31 rules. 0 are common, 18 are distinctive, and 13 are very specific to this project.

  • 0

    Common

    Most of the 6 edited versions we read keep these.

  • 18

    Distinctive

    This author's own choices. Worth a look.

  • 13

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Very specific:

    Use this skill for Codex-side publication work in opencode-swarm.

    line 15
  2. Very specific:

    Read and follow ../../../.claude/skills/commit-pr/SKILL.md in full before committing, pushing, opening, or updating a PR.

    line 19
  3. Very specific:

    Use the available shell execution tool for git, gh, build, test, and CI commands.

    line 32
  4. Very specific:

    Use apply_patch for manual file edits.

    line 33
  5. Distinctive:

    Use the runtime's parallel-execution capability for independent reads, status checks, and non-conflicting repo inspections.

    line 34
  6. Very specific:

    Do not stage .Codex/, IDE-local files, or unrelated worktree changes.

    line 35
  7. Distinctive:

    Do not open a second PR when one already exists for the branch.

    line 36
  8. Distinctive:

    If the branch already has a PR:

    line 50
  9. Distinctive:

    update the existing PR body instead of creating a new PR

    line 55
  10. Distinctive:

    verify the PR still points at the pushed branch head after any force-push

    line 57
  11. Very specific:

    use $swarm-pr-feedback when feedback needs fixes before closeout

    line 58
  12. Distinctive:

    If the user explicitly asks for a ready PR sooner, obey that request and document any remaining CI state clearly.

    line 67
  13. Distinctive:

    When the user asks to open, ship, ready, or close out a PR:

    line 71
  14. Very specific:

    verify check data belongs to the current headRefOid after every follow-up push

    line 74
  15. Very specific:

    if a required job is cancelled and downstream jobs are skipped, inspect the run and rerun failed/cancelled jobs

    line 75
  16. Distinctive:

    if an obsolete older-head run is already failed or still consuming concurrency, inspect it before cancellation and cancel only when it is …

    line 76
  17. Distinctive:

    wait for required checks to finish when practical

    line 77
  18. Very specific:

    after conflict resolution, verify GitHub reports both mergeable: MERGEABLE and mergeStateStatus: CLEAN.

    line 78
  19. Very specific:

    when gh pr checks looks stale, inspect the workflow run directly with gh run view <run-id> --json headSha,status,conclusion,jobs,url, and …

    line 79
  20. Distinctive:

    if you edit the PR body after checks are green, expect PR Standards/title checks to rerun and re-check before calling the PR green

    line 80
  21. Very specific:

    do not call the PR merge-ready while a required check is cancelled, skipped, in_progress, or otherwise non-green unless the user explicitly …

    line 81
  22. Very specific:

    After a forced install on Windows, EPERM while reading refreshed node_modules is usually host friction.

    line 94
  23. Distinctive:

    If the PR closes an issue:

    line 100
  24. Distinctive:

    if the PR merged before the comment was posted, add the missing issue comment immediately

    line 103
  25. Very specific:

    Read AGENTS.md, docs/engineering-invariants.md as needed, and the canonical Claude commit-pr skill.

    line 107
  26. Distinctive:

    Run the required local validation from the canonical skill.

    line 108
  27. Distinctive:

    Create or update the PR.

    line 110
  28. Distinctive:

    Verify issue comments, ready/draft state, and merge-readiness claims before finishing.

    line 113
  29. Distinctive:

    When local validation is noisy or host-limited:

    line 117
  30. Distinctive:

    document proven pre-existing failures in the PR body

    line 120
  31. Distinctive:

    use remote CI as the cross-platform source of truth for merge-readiness

    line 121

“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Commit & PR

Use this skill for Codex-side publication work in opencode-swarm.

Source Of Truth

Read and follow ../../../.claude/skills/commit-pr/SKILL.md in full before committing, pushing, opening, or updating a PR.

That Claude skill remains the repository's canonical publish protocol. This Codex-native skill is the adapter layer: it tells Codex which tools to use, how to handle existing PRs, and how to finish the post-open lifecycle without drifting from the repo contract.

If instructions conflict:

  1. AGENTS.md
  2. docs/engineering-invariants.md
  3. ../../../.claude/skills/commit-pr/SKILL.md
  4. this file

Codex Execution Rules

  • Use the available shell execution tool for git, gh, build, test, and CI commands.
  • Use apply_patch for manual file edits.
  • Use the runtime's parallel-execution capability for independent reads, status checks, and non-conflicting repo inspections.
  • Do not stage .Codex/, IDE-local files, or unrelated worktree changes.
  • Do not open a second PR when one already exists for the branch. Update the existing PR instead.

Publish Modes

Infer the mode from the user request and branch state:

  • new-pr: branch is ready to publish and no PR exists yet
  • pr-followup: a PR already exists and you are addressing review feedback or syncing validation
  • closeout: a PR already exists and you are updating body text, waiting on CI, marking ready, or confirming merge prerequisites

Required Follow-Through

Existing PR updates

If the branch already has a PR:

  • inspect unresolved PR feedback surfaces before updating or readying the PR: review threads/comments, requested-changes reviews, CI/check failures, mergeability/conflicts, and whether check data belongs to the current head SHA
  • update the existing PR body instead of creating a new PR
  • refresh ## Summary, ## Invariant audit, and ## Test plan when validation counts, caveats, or evidence changed
  • verify the PR still points at the pushed branch head after any force-push
  • use $swarm-pr-feedback when feedback needs fixes before closeout

Draft vs ready

Default behavior:

  • open or keep the PR as draft while follow-up edits are still expected, required checks are pending, or known caveats still need explanation
  • mark the PR ready only after the body is current and required remote checks are green

If the user explicitly asks for a ready PR sooner, obey that request and document any remaining CI state clearly.

Remote checks are authoritative

When the user asks to open, ship, ready, or close out a PR:

  1. inspect remote checks with gh pr view <n> --json statusCheckRollup,... or gh pr checks <n>
  2. verify check data belongs to the current headRefOid after every follow-up push
  3. if a required job is cancelled and downstream jobs are skipped, inspect the run and rerun failed/cancelled jobs
  4. if an obsolete older-head run is already failed or still consuming concurrency, inspect it before cancellation and cancel only when it is no longer the PR head under validation
  5. wait for required checks to finish when practical
  6. after conflict resolution, verify GitHub reports both mergeable: MERGEABLE and mergeStateStatus: CLEAN; local conflict-marker cleanup is not enough, and branch drift must appear in the closure ledger when it affected the PR
  7. when gh pr checks looks stale, inspect the workflow run directly with gh run view <run-id> --json headSha,status,conclusion,jobs,url, and keep watching downstream integration/smoke jobs after unit jobs pass
  8. if you edit the PR body after checks are green, expect PR Standards/title checks to rerun and re-check before calling the PR green
  9. do not call the PR merge-ready while a required check is cancelled, skipped, in_progress, or otherwise non-green unless the user explicitly accepts that state

Recommended commands:

gh pr view <number> --json body,headRefName,headRefOid,isDraft,mergeable,mergeStateStatus,state,statusCheckRollup,url
gh pr checks <number> --watch --fail-fast
gh run view <run-id> --json headSha,status,conclusion,jobs,url
gh run rerun <run-id> --failed

dist/ is generated output and is not committed (#1047); do not stage it and there is no dist-check drift gate. If package-check fails, inspect the CI log: it is a source / build / package.json#files problem (an incomplete or unbuildable tarball), not generated-file drift. Fix the source or manifest and rerun bun run package:smoke locally.

After a forced install on Windows, EPERM while reading refreshed node_modules is usually host friction; rerun the exact focused command with approved access before treating it as a code failure.

Issue comment requirement

If the PR closes an issue:

  • confirm the issue comment exists with the PR link, what changed, how to use it, and migration notes
  • if the PR merged before the comment was posted, add the missing issue comment immediately

Working Pattern

  1. Read AGENTS.md, docs/engineering-invariants.md as needed, and the canonical Claude commit-pr skill.
  2. Run the required local validation from the canonical skill.
  3. Stage only intended files.
  4. Create or update the PR.
  5. Refresh PR body text if the branch changed after review feedback.
  6. Watch remote checks to a useful conclusion when the user asked for publication or closeout.
  7. Verify issue comments, ready/draft state, and merge-readiness claims before finishing.

Clean-Main And CI Reality Checks

When local validation is noisy or host-limited:

  • compare suspicious failures against a clean origin/main worktree before blaming the branch
  • document proven pre-existing failures in the PR body
  • use remote CI as the cross-platform source of truth for merge-readiness

Final Sanity Check

  • one branch, one PR
  • PR head matches pushed HEAD
  • PR body matches current validation evidence
  • issue comment exists when the PR closes an issue
  • draft or ready state matches user intent
  • required remote checks are green, or any remaining non-green state is explicitly explained

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.