handoff
/handoff
Zamyka ukończoną pracę w julianpalomo.dev: uruchamia sprawdzenia, pyta o zmiany w dokumentach, commituje i pisze podsumowanie.
- Autor
- julian-palomo ↗
- Repozytorium
- julianpalomo.dev ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 11.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 3
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 11.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Zamyka ukończoną pracę w julianpalomo.dev: uruchamia sprawdzenia, pyta o zmiany w dokumentach, commituje i pisze podsumowanie.
Wyzwalacze z opisu
Oryginalny opis (EN)
Close out a finished unit of work in julianpalomo.dev. Runs the checks, applies the "is any doc now false?" test, stages by name, commits, runs the two reviewer subagents, and writes the change summary (complex parts in English and Spanish). Use when an implementation is done, before /ship-pr.
Jak działa
- 1. Checks
- 2. Leave the docs true
- 3. What to stage
- 4. Commit
- 5. Run the reviewers
- 6. Summary
Podgląd SKILL.md
---
name: handoff
description: Close out a finished unit of work in julianpalomo.dev. Runs the checks, applies the "is any doc now false?" test, stages by name, commits, runs the two reviewer subagents, and writes the change summary (complex parts in English and Spanish). Use when an implementation is done, before /ship-pr.
---
# Handoff
Run these in order. Don't skip a step silently; if you skip one, the summary
says so.
## 1. Checks
```bash
pnpm check
pnpm test
```
- Add `pnpm knip` when the change touches exports, `package.json` or `pnpm-lock.yaml`.
- Run `/run-evals` for one full `pnpm eval` run when the change touches the files its step 1 lists (the bot, the claims, `evals/` or the dependencies), and put its cost (bot, judge, calibration) in the summary. After changing the judge's prompt, model or effort, also run the calibration alone three times (`EVAL_CALIBRATION_ONLY=1 pnpm eval`) and say whether any item's verdict changed. Both call the real models.
- Add `pnpm test:hooks` when anything in `.claude/hooks/`, `biome.json`, `.claude/skills/implement-issue/wait-for-reply.sh` or `.claude/skills/mailbox/create-check.sh` changed. CI runs it too, but a local run catches a break before the push.
- `pnpm build` and `pnpm test:e2e` when the change touches routes, components, styles, the landscape, the build or CI: run them without asking, and say so in one line before starting. Otherwise rely on CI's e2e job and say so.
A failure stops the handoff: report it with its output.
## 2. Leave the docs true
For every doc the change could have made false or incomplete, read what it says
now and fix it. The test is "does an existing doc now say something untrue or
incomplete?", not "did I add a feature?". Triggers per doc:
[docs-table.md](docs-table.md). A change that makes nothing untrue needs no doc
edit; say so.
## 3. What to stage
```bash
git status --short
git diff origin/main --stat
```
List the files that belong to the task. Anything else (stray untracked files,
edits from another task, `CLAUDE.local.md`) is listed under "not staged" and
left untouched. Never `git add -A`, `--all` or `.`.
## 4. Commit
```bash
git branch --show-current
```
If that isn't the branch this issue's work belongs to, stop and say so: another
session may be using this folder (AGENTS.md §2, "One session, one folder").
```bash
git add <path> <path> …
git commit -m "<short lowercase imperative subject>"
```
Exemplar: the subjects in `git log --oneline origin/main` ("add ci workflow",
"update readme with status and commands"). No `Co-Authored-By`, no "Generated
with", no attribution of any kind. Several commits are fine when the work splits
into reviewable steps, as long as each one passes `pnpm check` on its own.
## 5. Run the reviewers
Both read-only subagents run on every handoff; this step is part of the
handoff, so don't ask first. They have no Bash, so write the branch diff for
them:
```bash
mkdir -p .diff-dump
git diff origin/main...HEAD --stat > .diff-dump/summary.txt
git diff origin/main...HEAD > .diff-dump/branch.diff
```
Then start both in one message, so they run in parallel, each in its own
context:
- `constraint-reviewer` checks the diff against AGENTS.md §1-§2 and `.claude/rules/`.
- `docs-truth-checker` finds doc sentences the diff made false.
Name any untracked task files in the prompt. `.diff-dump/` is gitignored.
For each finding, fix it in a new commit and re-run the checks from step 1, or
keep it and say why in the summary. One pass: run the reviewers again only when
the fixes are large enough to need their own review.
## 6. Summary
End with the change summary from [summary-template.md](summary-template.md).
Then `/ship-pr`.
If the exemplar's shape has changed, update this skill.
Pliki w paczce (3)
- SKILL.md 4 KB
- docs-table.md 3 KB
- summary-template.md 1 KB
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · julian-palomo/julianpalomo.dev | 100 | MIT |
| lumpen-mya/claude-handoff | 100 | MIT |
| generous-corp/pulp | 100 | MIT |
| Ovid/paad | 100 | MIT |
| Ovid/paad | 100 | MIT |
| Ovid/paad | 100 | MIT |
| webdevnerdstuff/claude-marketplace | 100 | MIT |
| terencebristol/terencebristol-toolkit | 100 | MIT |
| verkyyi/claude-fleet | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki f76c88217881 (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
- 100implement-issue
Realizuje zgłoszenie z GitHub od początku do końca: tworzy gałąź, publikuje plan do zatwierdzenia, implementuje i obsługuje recenzję PR. - 100ship
Zamyka sesję jednym ruchem: commituje, wysyła, zamyka beads i synchronizuje, gdy praca jest skończona. - 86close
Zamyka zakończoną pracę przed końcem sesji: dopisuje wpis do dziennika, aktualizuje punkt wznowienia i wypycha zmiany. - 100dev-ship
Wypycha gałąź, uruchamia wszystkie zestawy testów, otwiera lub aktualizuje pull request i obserwuje CI, ale nigdy nie scala. - 86end-session
Zamyka sesję pracy: sprawdza testy, aktualizuje ROADMAP i HANDOFF oraz wskazuje zmiany, które nie są w commicie. - 100ship-pr
Doprowadza pracę do PR gotowego dla człowieka: sprawdzenia, weryfikacja, samorecenzja i obsługa CI oraz botów.