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
- babs/skillstheir own version, first seen Nov 8, 2026
- theexperiencecompany/gaiatheir own version, first seen Nov 3, 2026
- azents/azentstheir own version, first seen Oct 23, 2026
- mergewatch/mergewatch.aitheir own version, first seen Oct 18, 2026
- shubham303/meelu-analytics-mcptheir own version, first seen Oct 13, 2026
- franciscoghansen/claude-skillstheir own version, first seen Oct 13, 2026
Other versions15 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 15
- classroomio/classroomio★ 1.7ksame name
- hoangsonww/Claude-Code-Agent-Monitor★ 1.2ksame name
- ultimo-rs/ultimo★ 802same name
- AmineDjeghri/awesome-os-setup★ 673same name
- theexperiencecompany/gaia★ 325same name
- seldonframe/seldonframe★ 40same name
- mergewatch/mergewatch.ai★ 26same name
- azents/azents★ 16same name
- shubham303/meelu-analytics-mcp★ 9same name
- babs/skills★ 4same name
5 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository9 other files it gives its agents.
- brainstormingSkill
- commit-messageSkill
- dev-verifySkill
- diff-reviewSkill
- env-doctorSkill
- fix-bugSkill
- git-workflowSkill
- refactorSkill
- workSkill
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
- Distinctive: line 13
Follow the stages in order.
- Very specific: line 19
Follow the git-workflow skill to create a feat/...
- Very specific: line 33
If ecosystem is unknown, ask the user for their commands before continuing.
- Very specific: line 44
If has_test_runner is false, skip test-writer and note that this feature is shipping without automated behavioral coverage.
- Very specific: line 49
Run the commands Stage 1 actually found — test, then lint, typecheck, build.
- Very specific: line 52
Report exactly what ran: Never say "tests pass" when has_test_runner is false.
- Distinctive: line 64
Wait for both before continuing.
- Distinctive: line 78
Keep a counter starting at 1.
- Distinctive: line 85
If issues remain and the counter is under 3, increment and repeat.
- Distinctive: line 88
Leave the work on the branch and hand it to a human.
- Very specific: line 93
Commit following commit-message — one line, type: summary, no body.
- Distinctive: line 101
If the repo has no remote, skip the push and say so — the rest of the pipeline still applies.
- Distinctive: line 116
If anything blocking appears, return to Stage 6.
- Very specific: line 135
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.
“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.
Files
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-revieweron the diff from Stage 2ui-checkeron 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-reviewerreturned "Changes requested" or "Blocking issues"ui-checkerreturned "Needs changes"
Issues → Stage 6. None → Stage 7.
Stage 6 — capped fix loop (max 3 attempts)
Keep a counter starting at 1.
- Launch
debuggerwith the specific failure output — verification failures and/or blocking review findings. It diagnoses only. - Launch
code-writerwith the debugger's root-cause report to implement the recommended fix. - Return to Stage 3, then Stage 4.
- If issues remain and the counter is under 3, increment and repeat.
- 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
- Review the working diff with the
diff-reviewskill. - Commit following
commit-message— one line,type: summary, no body. - Rebase onto the current base per
git-workflow. - 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.
- Re-run the Stage 3 verification commands against the rebased branch. Code that passed before a rebase can fail after one.
- Launch
code-quality-revieweron the full branch diff against the base (git diff <base>...HEAD), not just the last commit — the PR reviewer will see all of it. - 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.
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
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.

