agentleFS
abigaylehabigayleh/claude-workflow-plugin/skills/ship-feature/SKILL.md

ship-feature

Skill0 starsChanged 46 days agoMITView on GitHub

Run the full spec-to-ready-for-PR pipeline for a feature — worktree, detect the stack, implement, verify, review, capped fix loop, push the branch, re-review against the rebased branch, then hand the user a gh pr create command. Claude never merges to the base branch.

  • Deletes or force-pushes
  • Commits and pushes

15 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. 0 are common, 7 are distinctive, and 7 are very specific to this project.

  • 0

    Common

    Most of the 15 edited versions we read keep these.

  • 7

    Distinctive

    This author's own choices. Worth a look.

  • 7

    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. Distinctive:

    Follow the stages in order.

    line 13
  2. Very specific:

    Follow the git-workflow skill to create a feat/...

    line 19
  3. Very specific:

    If ecosystem is unknown, ask the user for their commands before continuing.

    line 33
  4. Very specific:

    If has_test_runner is false, skip test-writer and note that this feature is shipping without automated behavioral coverage.

    line 44
  5. Very specific:

    Run the commands Stage 1 actually found — test, then lint, typecheck, build.

    line 49
  6. Very specific:

    Report exactly what ran: Never say "tests pass" when has_test_runner is false.

    line 52
  7. Distinctive:

    Wait for both before continuing.

    line 64
  8. Distinctive:

    Keep a counter starting at 1.

    line 78
  9. Distinctive:

    If issues remain and the counter is under 3, increment and repeat.

    line 85
  10. Distinctive:

    Leave the work on the branch and hand it to a human.

    line 88
  11. Very specific:

    Commit following commit-message — one line, type: summary, no body.

    line 93
  12. Distinctive:

    If the repo has no remote, skip the push and say so — the rest of the pipeline still applies.

    line 101
  13. Distinctive:

    If anything blocking appears, return to Stage 6.

    line 116
  14. Very specific:

    If the feature has a visual surface, also hand over a dev-verify checklist so the user can look at it before opening the PR.

    line 135

“Common” is measured: kept by at least 6 in 10 of the 15 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

Ship a feature to ready-for-PR

You are orchestrating a fixed pipeline for the feature described in this skill's arguments (a spec file path, or a short description).

Follow the stages in order. Don't skip or reorder them. The pipeline ends at a pull request the user opens — you never merge to the base branch, and hooks will block you if you try.

Stage 0 — worktree

Follow the git-workflow skill to create a feat/... branch in its own worktree off the base branch. Every commit goes there.

Stage 1 — detect the stack

python3 "${CLAUDE_PLUGIN_ROOT}/scripts/detect-stack.py"

This returns the project's real test, lint, build, typecheck, and dev commands, or null where none exists. Use these throughout — never assume a command that wasn't reported. Read the notes array; it flags things like multiple lockfiles and missing runners.

If ecosystem is unknown, ask the user for their commands before continuing.

Stage 2 — implement

Launch the code-writer subagent with the spec.

If has_test_runner is true, launch test-writer in the same message so they run in parallel — and give it only the spec, never code-writer's output. Their independence is the entire value; a test written against the implementation just asserts the bugs back.

If has_test_runner is false, skip test-writer and note that this feature is shipping without automated behavioral coverage.

Stage 3 — verify

Run the commands Stage 1 actually found — test, then lint, typecheck, build. Skip any that are null and say which you skipped and why.

Report exactly what ran. Never say "tests pass" when has_test_runner is false. See ${CLAUDE_PLUGIN_ROOT}/conventions/testing.md.

Capture the failure output if anything fails.

Stage 4 — first review

In a single message, launch both:

  • code-quality-reviewer on the diff from Stage 2
  • ui-checker on the same diff — skip if there's no UI change

Wait for both before continuing.

Stage 5 — decide

Issues are "found" if any of these hold:

  • Stage 3 verification failed
  • code-quality-reviewer returned "Changes requested" or "Blocking issues"
  • ui-checker returned "Needs changes"

Issues → Stage 6. None → Stage 7.

Stage 6 — capped fix loop (max 3 attempts)

Keep a counter starting at 1.

  1. Launch debugger with the specific failure output — verification failures and/or blocking review findings. It diagnoses only.
  2. Launch code-writer with the debugger's root-cause report to implement the recommended fix.
  3. Return to Stage 3, then Stage 4.
  4. If issues remain and the counter is under 3, increment and repeat.
  5. After 3 attempts, stop and report the current state. Three failed attempts means the problem isn't what you think it is; a fourth won't fix it. Leave the work on the branch and hand it to a human.

Stage 7 — commit and push the branch

  1. Review the working diff with the diff-review skill.
  2. Commit following commit-message — one line, type: summary, no body.
  3. Rebase onto the current base per git-workflow.
  4. Push the feature branch:
    git push -u origin <branch>          # first push
    git push --force-with-lease --force-if-includes   # after a rebase

If the repo has no remote, skip the push and say so — the rest of the pipeline still applies.

Stage 8 — second review, against the rebased branch

The Stage 4 review saw the code as written. This one sees it as the pull request will: rebased onto the current base, with whatever landed there in the meantime. That's a different question, and it's where integration problems surface.

  1. Re-run the Stage 3 verification commands against the rebased branch. Code that passed before a rebase can fail after one.
  2. Launch code-quality-reviewer on the full branch diff against the base (git diff <base>...HEAD), not just the last commit — the PR reviewer will see all of it.
  3. If anything blocking appears, return to Stage 6.

Stage 9 — hand off the PR

Stop here. Opening the pull request is the user's decision.

Report:

  • What was implemented
  • What verification actually ran, and its result — name the commands
  • What each reviewer said, in both passes
  • How many fix attempts were needed
  • Anything you deliberately left alone, and why
  • The command to open the PR:
gh pr create --base <base> --head <branch> --title "<type: summary>" --body "..."

If the feature has a visual surface, also hand over a dev-verify checklist so the user can look at it before opening the PR.

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.