oak-runtime-verification
/oak-runtime-verification
Sprawdza w działającym systemie, że zmiana działa, w tym ścieżkę błędu, powtarzalnymi krokami i minimalnym dowodem po testach.
- Autor
- nasac0x ↗
- Repozytorium
- code-orchestrator ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 31.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 31.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Sprawdza w działającym systemie, że zmiana działa, w tym ścieżkę błędu, powtarzalnymi krokami i minimalnym dowodem po testach.
Oryginalny opis (EN)
Use after tests and validation pass and before the final review, whenever a change could show up at runtime - a screen, a command, an endpoint, a job or a data store - to prove the change works in the running system, including its failure path, with repeatable steps and minimal evidence.
Jak działa
- Overview
- Decide the surface
- Find how the project is exercised
- Exercise it
- Evidence
- Report
- When it cannot run
- Run event
Podgląd SKILL.md
---
name: oak-runtime-verification
description: Use after tests and validation pass and before the final review, whenever a change could show up at runtime - a screen, a command, an endpoint, a job or a data store - to prove the change works in the running system, including its failure path, with repeatable steps and minimal evidence.
phase: verify
domains:
- testing
- evaluation
stacks:
- any
allowed_agents:
- developer
skill_source: built-in
origin: null
status: active
---
# Runtime Verification
## Overview
Tests prove that the tests pass. This step proves that the change works in the
running system. It runs after validation is green and before the final review,
and it exercises the real surface the change touches: the happy path and the
failure path.
## Decide the surface
Pick exactly one surface per affected area:
| Surface | Use when | What to do |
| --- | --- | --- |
| `none` | refactor, types, build-only change | Skip, and state why in one line. |
| `command` | a CLI, script or job | Run it and inspect its effect and logs. |
| `endpoint` | an HTTP, RPC or message handler | Call it and inspect the response and effect. |
| `store` | a database, cache, file or queue | Query it before and after. |
| `screen` | a user interface | Drive the real application. |
Never open a browser or a simulator for a change that has no screen.
## Find how the project is exercised
Look in this order and stop at the first answer:
1. The project's own mandatory steps, or its `AGENTS.md`.
2. The documents those link to.
3. The repository itself: manifest scripts, container files, an end-to-end
suite, a seed command.
A project with no way to run the surface is a finding to report. It is not a
reason to skip the step.
## Exercise it
- Run the happy path, then at least one failure path.
- Check the real effect, not just the exit code: the row written, the message
sent, the file created, the log line, the cache entry.
- Write the steps down so someone else can repeat them exactly.
- For a migration, use the real engine and check three things:
1. applying it a second time changes nothing;
2. it migrates a store that already holds data;
3. the engine and its version are named in the report.
## Evidence
Collect the cheapest evidence that proves the point:
- `screen`: one still image per state;
- `command`: the verbatim output and exit code, as text;
- `endpoint`: the request sent, and the status and body received;
- `store`: the query, and the rows before and after.
Do not record video. Keep evidence in a temporary directory outside the
repository, and never commit it.
## Report
The report states:
- what was run;
- the evidence, by file name only;
- what was observed, including the failure path;
- a **Not covered** list;
- a verdict: `works_as_specified`, `does_not`, or `could_not_be_verified`.
`could_not_be_verified` names exactly what was missing.
Record the result in the developer's Verification Envelope as
`runtime_verification: { surface, verdict, report }`.
## When it cannot run
If the step needs something you do not have (a credential, a paid service, a
device, an unavailable dependency), say so and return `could_not_be_verified`.
Under `/autonomous` this is a stop reason. Never report the step as done on the
strength of unit tests.
## Run event
When a run is open, emit one event per surface, with counts and enums only:
```bash
oak run event --type runtime_verification surface=<none|command|endpoint|store|screen> verdict=<works_as_specified|does_not|could_not_be_verified> evidence_count=<n>
```
Wersje
W bazie nie ma innych skilli o nazwie oak-runtime-verification.
Migawka bazy z 13.11.2026 · wersja paczki 581c8b7904c8 (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
- 95runtime-verification
Weryfikuje, czy zmiana w kodzie działa, uruchamiając aplikację i obserwując zachowanie, przed wypchnięciem. - 90verification
Sprawdza, czy zmiana działa: dobiera i uruchamia testy, kontroluje interfejs w przeglądarce i zgłasza niewykonane kontrole. - 100runtime-validation
Dowodzi, że zmiana działa przez prawdziwe interfejsy: E2E w przeglądarce, smoke testy API, CLI, dostępność, wydajność i regresje wizualne. - 95oak-backend
Pomaga budować backendy w Oak, frameworku HTTP dla Deno, z middleware, kontekstem i bezpiecznym routingiem. - 90test-and-verify
Testuje jedno zachowanie krótką pętlą dowodów: punkt wyjścia, ewentualny dowód błędu, minimalna zmiana i końcowy dowód działania. - 100oak-api
Buduje bezpieczne ścieżki nauki z API Oak Curriculum, gdy potrzebne są endpointy Oak, mapy programowe lub wskazówki Apps SDK dla dzieci.