Przejdź do treści
Biblioteka skilli
Programowanie · Testowanie

show-your-work

/show-your-work

Sprawdza w uruchomionej aplikacji, czy zmiana działa, i zbiera dowody, które człowiek może zweryfikować przed scaleniem.

Repozytorium
evident ↗
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
15
Skrypty
5
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
  • 5 skryptów w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.

Skrypty: scripts/build-proof.mjs, scripts/ledger.mjs, scripts/questions.mjs, scripts/report.mjs, scripts/run-folder.mjs

Co robi i kiedy

Sprawdza w uruchomionej aplikacji, czy zmiana działa, i zbiera dowody, które człowiek może zweryfikować przed scaleniem.

Wyzwalacze z opisu

prove my changesshow it worksproof of workanswer Q-1a2b: ...
Oryginalny opis (EN)

Use when the user asks to prove, verify or QA a change, commit or pull request in the running app (web, API, or a target you define), wants evidence a person can check that it works before merging, says "prove my changes", "show it works" or "proof of work", asks to prove bugs an exploring run found (for example once they are fixed), asks to re-prove run <folder>, or answers a question a run asked ("answer Q-1a2b: ...").

Jak działa

  1. Terms
  2. 1. Set up
  3. 2. Claims
  4. 3. Approve, prove and report
  5. Answering a question
  6. Auto mode
Sekcje z nagłówków SKILL.md, w oryginale · 68 linii, 903 słów
Podgląd SKILL.md
---
name: show-your-work
description: 'Use when the user asks to prove, verify or QA a change, commit or pull request in the running app (web, API, or a target you define), wants evidence a person can check that it works before merging, says "prove my changes", "show it works" or "proof of work", asks to prove bugs an exploring run found (for example once they are fixed), asks to re-prove run <folder>, or answers a question a run asked ("answer Q-1a2b: ...").'
license: MIT
compatibility: Needs an agent that can start subagents, Node.js 18+, git, the driver your target's definition names, and the evidence skill installed next to this one.
---

# Show your work

Turn a change into claims of what is now true for users, have the user approve them, then have blind testers prove or disprove each claim with evidence.

**Core rule: testers get only the claims.** Subagents do all the work. Their prompts hold only the dispatch lines below, plus the two lines `PARALLEL.md` adds. Provers get only each claim's test fields (what `--session` writes): never its source, its code-check mark, the diff or the code. Never add your own notes, suspicions or implementation details to any prompt, to `brief.md` or to a claim. The writer writes every claim. You edit `claims.json` only as `PROOF.md` §1 step 3 says.

For "re-prove run <folder>", read `<skill directory>/REPROVE.md` and follow it instead of the steps below.

## Terms

- **Skill directory**: the absolute folder that holds this `SKILL.md`.
- Run terms are in `RUN.md`: read them before §1.
- **Auto mode**: on when the request says `--auto` or `auto mode`, or `<project folder>/environment.md` Preferences say `Mode: auto` (read that line before §1); `--ask` or `ask mode` turns it off. When on, read `<skill directory>/AUTO.md` and `## Auto mode` below first, and follow them wherever a step asks the user.

## 1. Set up

1. Input. By default: the app repo's changes against its merge base with the main branch (ask if unclear), plus untracked files. Diff command: `git diff <merge base> -- . ':(exclude,glob)**/.evident/**' && git ls-files --others --exclude-standard -- . ':(exclude,glob)**/.evident/**'`, or with no merge base (a new repo, no main branch) `git diff HEAD -- . ':(exclude,glob)**/.evident/**' && git ls-files --others --exclude-standard -- . ':(exclude,glob)**/.evident/**'` (evident's own folder is never part of the change). A named commit: `git diff <commit>~1 <commit>`. A range ("the last 3 commits"): `git diff <from> <to>`. What the user names to prove ("prove that checkout with a coupon works") is input too, with or without a diff; with only their words, the diff command is `none`. Bugs an `bug-bash` run filed are input too ("prove BUG-002 of <run folder>", "prove the bugs of <run folder>"): the diff command is `none` unless a change is named as well, and the writer's prompt gets a `Findings:` line with each bug folder. A claim from a bug says what a user should see, so proven means the bug no longer happens. With no changes and no commit, words or bugs named, ask which commit or behaviour to prove.
2. If the input is too thin to write claims, ask what the change is meant to do. "Prove my changes" with no request for the change in this conversation is too thin. Never fill a gap from the code or your own guesses.
3. Follow `<skill directory>/RUN.md` with kind `proof`.

## 2. Claims

Dispatch one subagent with exactly this prompt, filled in:

```
Read <skill directory>/CLAIMS.md and <skill directory>/<CLAIMS-findings.md when bugs are the input, else CLAIMS-change.md> and follow them.
Run folder: <run folder>
Project folder: <project folder>
Skill directory: <skill directory>
Code path: <app repo>
Diff command: <diff command>
Ledger dir: <ledger dir>
Findings: <bug folders, comma-separated; only when bugs are the input>
```

If it returns `too big: propose phases <list>`, show the phases and ask which to prove first. Add the user's answer to `## To prove` in `brief.md`, verbatim, and dispatch the writer again.

Then dispatch a second subagent with exactly this prompt, filled in:

```
Read <skill directory>/CODE-CHECK.md and follow it.
Run folder: <run folder>
Skill directory: <skill directory>
Code path: <app repo>
Diff command: <diff command>
```

Run `node <skill directory>/scripts/build-proof.mjs <run folder> --plan` and show its output to the user verbatim, `Problem:` lines included.

## 3. Approve, prove and report

Follow `<skill directory>/PROOF.md`.

## Answering a question

When the user answers a question a report or chat summary asked ("answer Q-1a2b: per org"), find the project folder as `RUN.md` step 1 does (`project.mjs` only), then run `node <skill directory>/scripts/questions.mjs <project folder> <Q-id> <their answer, verbatim>` and show its lines verbatim. A message pasted from the report's Copy answers holds one `answer <Q-id>: ...` line per question: run the command once per line, in one call. Its `decide <BUG-NNN or C-id> of ...` lines are fix-and-prove's: follow `<skill directory>/../fix-and-prove/SKILL.md` `## Decisions from a page` for them. Do nothing else: the feature's next explore and proof read the answer. Never answer a question yourself, in any mode. On `Problem: no question`, say so and ask for the id again.

## Auto mode

Only when the run is in auto mode. `AUTO.md`'s rows hold here too.

| Where | In auto mode | Kind |
|---|---|---|
| §1 step 1: merge base unclear | The merge base with `main`, else `master`. A repo with no branch but the current one: `git diff HEAD`, as §1 says. Other branches but neither `main` nor `master`: stop: `Stopped: auto mode found no main or master branch to diff against. Run again naming the commit or range to prove.` | rule |
| §1 step 1: no changes, and no commit or behaviour named | Stop: `Stopped: auto mode needs a change, a commit or the behaviour to prove. Run again naming one.` | hard stop |
| §1 step 2: input too thin to write claims | Stop: `Stopped: auto mode needs the change's intent in the request. Run again with what the change is meant to do.` Never fill the gap from the code. | hard stop |
Pliki w paczce (15)
  • AUTO.md 7 KB
  • CLAIMS-change.md 2 KB
  • CLAIMS-findings.md 4 KB
  • CLAIMS.md 13 KB
  • CODE-CHECK.md 1 KB
  • PROOF.md 8 KB
  • PROVER.md 8 KB
  • REPROVE.md 2 KB
  • RUN.md 7 KB
  • SKILL.md 6 KB
  • scripts/build-proof.mjs 74 KB
  • scripts/ledger.mjs 9 KB
  • scripts/questions.mjs 4 KB
  • scripts/report.mjs 97 KB
  • scripts/run-folder.mjs 8 KB

Wersje

RepozytoriumWynikLicencja
Ta strona · v1n33th/evident100MIT
natsummerance/agents90MIT
natsummerance/agents90MIT

Migawka bazy z 13.11.2026 · wersja paczki 0d6f4f4a992d (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

  • 100verify
    Sprawdza, czy zmiana robi to, o co chodziło, od najtańszej kontroli w górę, zanim przekaże się PR do przeglądu.
  • 90verify
    Dowodzi, że zmiana działa, uruchamiając sprawdzenia projektu i raportując dowody, zanim zadanie zostanie uznane za skończone.
  • 100verify-ui-change
    Sprawdza w działającej aplikacji, czy zmiana w aplikacji webowej naprawdę działa, i zwraca werdykt z wysłanym żądaniem, zmienionym stanem i wskazaniem file:line do poprawy.
  • 100verify
    Dowodzi, że zmiana działa przed zgłoszeniem jej jako gotowej, i pokazuje dowody od „powiedziane” do „uruchomione”.
  • 90truly-done
    Sprawdza w działającej aplikacji webowej, czy zmiana działa: klika i pisze w przeglądarce bez interfejsu i odpowiada zaliczone, niezaliczone lub brak pewności, z dowodem.
  • 100verify-with-playwright
    Steruje prawdziwą przeglądarką na działającym serwerze deweloperskim AI Atelie, by zweryfikować zmianę, i zapisuje zrzuty ekranu oraz wideo jako dowody do PR.

Uwagi i zgłoszenia

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