Przejdź do treści
Biblioteka skilli
Programowanie · Automatyzacja

ship

/ship

Sprawdza, czy praca jest naprawdę skończona, potem zapisuje zmiany w commicie, otwiera PR, kończy gałąź i wdraża zmianę.

Repozytorium
lll-skills ↗
Licencja
MIT
Ostatnia zmiana repozytorium
12.11.2026

Ocena i bezpieczeństwo

100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
8
Skrypty
1
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 12.11.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co sprawdzić przed instalacją
  • 1 skrypt w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.

Skrypty: scripts/canary.mjs

Co robi i kiedy

Sprawdza, czy praca jest naprawdę skończona, potem zapisuje zmiany w commicie, otwiera PR, kończy gałąź i wdraża zmianę.

Oryginalny opis (EN)

Prove work is really done, then commit, open a PR, finish the branch, deploy and watch it. Use when about to say done, fixed or passing, or asked to commit, make a PR, merge, set up a worktree, release, deploy, or check a deploy after it went out. Not for go-live checks of a website (lll:launch), reviewing code (lll:review) or app store releases (lll:mobile).

Jak działa

  1. When to use
  2. When not to use
  3. Modes
  4. Route
  5. Steps
  6. Stop conditions
  7. Red flags
  8. Output
Sekcje z nagłówków SKILL.md, w oryginale · 77 linii, 828 słów
Podgląd SKILL.md
---
name: ship
description: Prove work is really done, then commit, open a PR, finish the branch, deploy and watch it. Use when about to say done, fixed or passing, or asked to commit, make a PR, merge, set up a worktree, release, deploy, or check a deploy after it went out. Not for go-live checks of a website (lll:launch), reviewing code (lll:review) or app store releases (lll:mobile).
metadata:
  origin: lll
---

# Ship

Takes finished work from "I think it works" to evidence, a clean commit, a reviewable PR, and a watched deploy.
Iron law: no completion claim without fresh verification output from this session.

## When to use

- About to say "done", "fixed", "passing" or "ready", or to commit or open a PR.
- Work is complete and the user must choose how to integrate it.
- An isolated workspace is needed before starting or executing a plan.
- A release or deploy is requested, or a deploy just happened and needs watching.
- Writing commit messages, a PR body or release notes.

## When not to use

- Reviewing the change for defects: `lll:review`.
- Building or fixing: `lll:build`; unknown cause: `lll:debug`.
- A website that just went live (smoke, SEO, legal, monitoring checklists): `lll:launch`.
- Backup and restore proof: `lll:backup-drill`. Docs refresh: `lll:docs`.

## Modes

- **Guidance** (default). Run the verification gate and report. Draft commit messages and the PR body. Present the finish menu. Commit only when asked. Push, merge, tag, publish and deploy never happen unless the user asks for that exact step in this conversation.
- **Full**. Only when the user asks to ship end to end. Gates, review check, commits, push of the feature branch, PR creation, and (each with its own explicit approval) merge and deploy, then a canary watch.

Both follow `../../shared/ground-rules.md`: remote and production writes need the owner. Stack and gate commands come from `../../shared/stack-detect.md`.

## Route

| Need | Read |
|---|---|
| About to claim any result | `references/verification.md` |
| Isolated workspace | `references/worktrees.md` |
| Work complete, choose merge, PR, keep or discard | `references/finish-branch.md` |
| Commit messages, PR title and body, release notes | `references/commit-and-pr.md` |
| Release or deploy | `references/deploy.md` |
| Watch a deploy | `references/canary.md` and `scripts/canary.mjs` |

## Steps

1. **Read the state.** `git status`, current branch, `git log <base>..HEAD`, any uncommitted work that is not yours. Other people or agents may share the checkout: touch only your files.
2. **Verify** (`references/verification.md`). Identify each claim you will make, find the command that proves it, run it fresh, read the full output and exit code, then state the result with the evidence. Include the requirement-by-requirement check against the plan or issue. UI and mobile work is run, not just compiled.
3. **Check review status.** Was the whole branch reviewed (`lll:review`)? If not, offer it; for risky changes insist. Open critical and high findings block shipping.
4. **Docs.** Behaviour or interface changed: update docs first (`lll:docs`).
5. **Commit.** Stage named paths only, never everything in a shared checkout. Conventional Commits; one logical change each; tests with the code (`references/commit-and-pr.md`). Add any attribution line the host or project requires.
6. **Finish the branch** (`references/finish-branch.md`). Re-verify on the merged result if integrating. Present the four options and wait for the choice: merge locally, push and open a PR, keep as is, discard (typed confirmation).
7. **Open the PR** when chosen. Title and body from the template, test evidence included. Use `gh` for GitHub; other hosts use their CLI or give the user the text.
8. **Deploy** only on request (`references/deploy.md`). Pre-deploy checks, migration order, rollback plan, then the deploy command the project documents. Merging to the default branch may itself trigger a deploy: check the pipeline first and say so.
9. **Watch** the deploy (`references/canary.md`): baseline before, checks after, alert on change versus baseline, only after two consecutive bad checks. Decide with the user: roll back, investigate or accept.
10. **Report** with evidence, what was not verified, and what is left for the owner.

## Stop conditions

Stop and ask before: merging to the default branch, pushing to a shared branch, force-pushing anything, tagging or publishing a release, any production deploy, discarding work, deleting a branch with unmerged commits, running a destructive migration.

## Red flags

| Thought | Reality |
|---|---|
| "Should work now" | Run it. |
| "I am confident" | Confidence is not evidence. |
| "The subagent said it passed" | Check the diff and run the check yourself. |
| "Linter passed, so build passes" | Different gates prove different things. |
| "I ran it earlier" | Earlier is before your last edit. Run fresh. |
| "Just this once, skip the check" | That is how broken main happens. |
| "Commit everything with `git add -A`" | You may commit someone else's work or a secret. Name paths. |
| "Push now, tidy later" | History on a shared branch is public. |
| "Deploy is automatic, so no check" | Automatic deploys need the most watching. |

## Output

- Verification block: each gate, its command, the real result line; requirements checklist; what was not verified.
- Commits (hashes and subjects), branch name, PR link if created.
- Deploy and canary outcome if requested.
- Next steps owned by the user, such as merge approval, deploy approval, secrets, store submission.
Pliki w paczce (8)
  • SKILL.md 5 KB
  • references/canary.md 4 KB
  • references/commit-and-pr.md 5 KB
  • references/deploy.md 6 KB
  • references/finish-branch.md 4 KB
  • references/verification.md 5 KB
  • references/worktrees.md 5 KB
  • scripts/canary.mjs 10 KB

Wersje

RepozytoriumWynikLicencja
Ta strona · lilly-logic-labs/lll-skills100MIT
rjmurillo/ai-agents100MIT
rjmurillo/ai-agents100MIT
namedtoaster/the-last-mile100MIT
rjmurillo/ai-agents100MIT
ericrisco/rsc-harness100MIT
zahidakhyar/skills100MIT
NolwennD/gabier100Apache-2.0
mgiovani/cc-arsenal100MIT

Migawka bazy z 13.11.2026 · wersja paczki d9fd6d082d90 (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 0 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 90review
    Przegląda diff, gałąź, pull request, commit lub zgłoszenie i zgłasza uwagi z oceną wagi oraz scenariuszem błędu.
  • 90debug
    Szuka przyczyny źródłowej błędu, testu, awarii lub spadku wydajności przed poprawką, a potem naprawia i potwierdza ją.
  • 100ship-pr
    Weryfikuje każdy PR w osobnym worktree, upraszcza diff i otwiera PR przez gh, gdy praca jest gotowa do recenzji.
  • 100ship-pr
    Uruchamia kontrole, aktualizuje dokumentację, zapisuje commit i otwiera PR z szablonu, tylko na prośbę o wysyłkę.
  • 90worktrees:finish
    Kończy pracę w worktree: tworzy pull request lub scala gałąź bezpośrednio, usuwa worktree i sprząta gałąź roboczą.
  • 95atlas-verify
    Sprawdza, czy praca jest naprawdę skończona: uruchamia build, sprawdzanie typów, linter i testy, a każdy wynik raportuje osobno.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.