Przejdź do treści
Biblioteka skilli
AI i agenci · Automatyzacja

engrave

/engrave

Zamienia potknięcie lub błąd w trwałą regułę, czyli sprawdzenie, hook albo linijkę skilla w repozytorium.

Repozytorium
vicky-skills ↗
Licencja
MIT
Ostatnia zmiana repozytorium
13.11.2026

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
1
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 13.11.2026, bez odwołań do wycofanych modeli i API.

Licencja

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

Co robi i kiedy

Zamienia potknięcie lub błąd w trwałą regułę, czyli sprawdzenie, hook albo linijkę skilla w repozytorium.

Wyzwalacze z opisu

codifylessonfor the futurefrom now onmake sure we knowprevent this
Oryginalny opis (EN)

Engrave a lesson: turn a snag, mistake or correction into a law (a check, hook or skill line) in its repo. Use when the user says "codify", "lesson", "for the future", "from now on", "make sure we know" or "prevent this", when a fixed snag cost a retry or its root cause is a platform limit, and for each finding settling or `captains-log` hands over. Any other correction waits for that settling.

Jak działa

  1. Steps
Sekcje z nagłówków SKILL.md, w oryginale · 36 linii, 966 słów
Podgląd SKILL.md
---
name: engrave
description: Engrave a lesson: turn a snag, mistake or correction into a law (a check, hook or skill line) in its repo. Use when the user says "codify", "lesson", "for the future", "from now on", "make sure we know" or "prevent this", when a fixed snag cost a retry or its root cause is a platform limit, and for each finding settling or `captains-log` hands over. Any other correction waits for that settling.
---

# Engrave

> _So let it be written, so let it be done. A lesson leaves as a law, never as a promise._

A **snag** is one thing that went wrong: a red check, a crash, a retry, a wrong claim, or a correction from the user. A correction that cost no retry is engraved in the batch `make-it-so`'s SETTLE.md runs, not on the spot. A **law** is what stops the snag next time. An apology or "I'll remember" is no law: the next session never reads it.

Build every law for the next agent: it sees only the files it opened, copies the nearest example, and takes the shortest path that passes.

## Steps

1. **Name the snag** in one line: what happened, and what should have happened. A law that states how something behaves (a file format, a render, an engine rule) waits for the check that proves it: a fix only a live client can confirm is engraved after the user reports it working, never on the theory. Done when the line names a visible event, not a feeling.
2. **Pick the law** at the highest rung that holds it. First find out whether the snag is a **repeat**: look it up in the ledger, `bash ${CLAUDE_PLUGIN_ROOT}/scripts/lesson-ledger.sh since 2000-01-01T00:00:00Z`, and grep the skills, agent files and tests where the lesson would land for a rule that already covers it. A match means the rule exists and nothing made the agent follow it. A first-time snag becomes one line in the skill or agent file it belongs to, placed in the step the agent is in when the snag happens. A repeat is promoted. A wrong claim made from memory or a glance, and a rule the user asks to enforce, are promoted the first time. A promoted snag climbs this ladder, top first:
   1. **Remove the way to make it.** One owner per piece of state, one supported way per task, the old way and its dead examples deleted, a hand-synced list replaced by its source.
   2. **Guard it.** Build a check or a hook where one can hold the rule, and a skill line only where neither can, with the reason (`robot-voice` step 1). A check's error names the fix: the file, command or skill to use instead. Where the bad pattern is already common, the check fails only on a change that adds more.
   3. **A skill line**, with an example of the snag beside it.

   Done when you can say "first time" or "repeat of <entry or `file:line`>" and the snag has its law.
3. **Route it.** A home the user names wins. Otherwise send each part to its own home, checked in the order plugin, user setup, repo:
   - **The plugin's repo**, only when the user can push to it: `gh repo view <repository in plugin.json> --json viewerPermission` says `ADMIN`, `MAINTAIN` or `WRITE`. When it can't answer, the repo is not pushable. It takes a lesson that would help an agent in any repo, such as one about the plugin's skills or hooks. Hand it to an idle session in the plugin's repo (`ListAgents`, then `SendMessage`), else build it on a fresh branch from its main in a worktree, without asking. Never switch a checkout another session uses.
   - **The user's own agent setup**: a pattern across tasks or repos. That is their `CLAUDE.md` (`~/.claude/CLAUDE.md`), or the repo that declares it when their setup lives in one.
   - **The repo you work in**: a lesson about that repo (its code, data, build or workflow), in a test, a standard or its agent docs, on the current effort's branch.

   Done when each part of the lesson has one destination.
4. **Retire the old rule.** When the law replaces or reverses a rule written anywhere (a doc, a skill, a decision record, a ticket), run `plot-twist` with the old rule as the decision it replaces. When a structural law (rung 1 or 2) now holds what a written line said, delete that line, or cut it to a pointer at the check. Done when `plot-twist` has reported and no line restates a rule a check now holds, or the law replaces no written rule.
5. **Build and prove it** this turn, under the destination's own rules: test first where it is code, its checks run. Replay the real snag against the law: the check or hook goes red on the snag's own input (the command, the diff, the text), and green once it is fixed. An exception the law must allow sits on its own line, with its reason. When you commit the rule, the commit message ends with `Engraved: <snag>`, and you run `bash ${CLAUDE_PLUGIN_ROOT}/scripts/lesson-ledger.sh add <plugin|setup|repo> "<snag>"`, so `sup` counts it from any repo. A law that lands in a skill, in a repo with an `evals/` suite, also gets an eval case in the same commit: `evals/<skill>/<case>/`, with a prompt that recreates the snag and a free grader (regex, tool_used, tool_order, file_exists) that fails against the skill text from before the law. Name the case at the end of the ledger snag: `"<snag> (eval case: <skill>/<case>)"`. A law that is a hook or a check with its own test needs no case. Done when the law is committed and went red on the snag, with its eval case when it lands in a skill, handed to a session that accepted it, or filed with what it waits on.
6. **Report** in one line per snag, in the reply that closes it:

   ```markdown
   📓 Engraved: <snag> → <law> in `<file>`
   ```

   Add the rung and, for a promoted snag below rung 1, the reason in a few words the rung above it couldn't hold it. Done when every snag of the turn has its line.

Charter: C5, C6 ([CHARTER.md](../../CHARTER.md)).

Wersje

W bazie nie ma innych skilli o nazwie engrave.

Migawka bazy z 13.11.2026 · wersja paczki 47bc7d981d51 (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

  • 86lesson
    Zamienia poprawkę lub powtarzający się błąd w zabezpieczenie: regułę, skrypt, test albo wpis w zasadach, jako ostatnią opcję.
  • 90learning
    Zamienia powtarzający się błąd w najmniejszą trwałą poprawkę, np. linię w CLAUDE.md, regułę lub hook, i proponuje usunięcie przestarzałych reguł.
  • 90correction-to-skill
    Zamienia poprawkę użytkownika w datowaną regułę w pliku skilla lub instrukcji, w tej samej sesji, i mówi, gdzie ją zapisał.
  • 100learning-to-skill
    Zapisuje lekcję z sesji do plików instrukcji agentów, np. AGENTS.md lub CLAUDE.md, wybierając plik i sekcję, bez duplikatów i tylko po akceptacji.
  • 100remember
    Zamienia trwałe instrukcje, preferencje i poprawki użytkownika w regułę projektu, a rzadziej w skill lub hook, tak by obowiązywały w kolejnych sesjach.
  • 100lesson
    Zapisuje korektę lub wypracowaną regułę jako lekcję ważoną pewnością, która wraca przed podobną pracą, na przykład po poleceniu „zawsze” albo „nigdy tak nie rób”.

Uwagi i zgłoszenia

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