verify
/verify
Uruchamia lokalną pętlę weryfikacji gałęzi przed PR, dobierając kontrole na podstawie zmienionych plików.
- Autor
- shakacode ↗
- Repozytorium
- agent-workflows ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 28.09.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 5
- Skrypty
- 1
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 28.09.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- 1 skrypt w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.
Skrypty: bin/verification-evidence-reuse-test.rb
Co robi i kiedy
Uruchamia lokalną pętlę weryfikacji gałęzi przed PR, dobierając kontrole na podstawie zmienionych plików.
Oryginalny opis (EN)
Run a local verification loop for the current branch before creating or updating a PR, selecting checks from the repo's binstub contract and changed files. Use when asked to verify, test, or prepare PR changes.
Jak działa
- Instructions
- Default Verification Order
- Scope Guide
- Output Format
Podgląd SKILL.md
---
name: verify
description: Run a local verification loop for the current branch before creating or updating a PR, selecting checks from the repo's binstub contract and changed files. Use when asked to verify, test, or prepare PR changes.
---
# Verify Command
Run a local verification loop for the current branch before creating or updating a PR.
Use `/verify` for local pre-PR checks. Use `/run-ci` when you need `.agents/bin/ci-detect` or want to reproduce CI job selection locally.
## Instructions
1. Read `AGENTS.md` first. It is the canonical source for boundaries and repository safety rules. Read `.agents/bin/README.md` and `.agents/agent-workflow.yml` for workflow commands and policy.
2. Resolve `BASE_BRANCH` from `.agents/agent-workflow.yml` key `base_branch`, then inspect the current branch diff
with `git status --short`, `git diff --name-only "origin/${BASE_BRANCH}...HEAD"`, and
`git diff --stat "origin/${BASE_BRANCH}...HEAD"`.
3. Decide the required verification set that covers the changed surface area using the **Scope Guide** below. Always
include `.agents/bin/lint` when present, and always include `.agents/bin/validate` before
creating a commit, including documentation-only changes. The repository wrapper owns its required coverage;
execute selection through its trusted-base wrapper or protected entry point under the
[delivery coverage contract](references/verification-evidence.md#delivery-coverage).
Do not infer reduced coverage from file extensions, project labels, or a candidate's changed policy.
4. Run each command in order; on failure, pause the command sequence to diagnose and fix it. Record the failing command, relevant error output, and next fix to attempt.
5. For formatting failures (auto-fixable formatter or lint offenses), run the repo's documented autofix command or `.agents/bin/lint` mode when it supports fixes; do not manually edit formatting-only changes.
6. After one or more edits for a failure, restart at the failed command and continue forward. Track a loop counter per
command:
- Increment the counter when the same command fails on the same first item (test name, lint offense, or formatter
file) as the previous run.
- Reset the counter when the first failing item changes or when you advance to a different command.
- After three consecutive cycles on the same item, stop blind retries. Perform one bounded diagnosis pass:
reproduce/minimize the failure, compare a supported hypothesis with evidence, and try a scoped correction
only if justified. If it still fails or no safe correction is supported, return the evidence and smallest next
action to the coordinator; a solo agent acts as coordinator. Do not reset the counter to repeat this pass.
- If a later fix regresses a previously passing file, symbol, or test, pause edits and isolate the cause.
Preserve unrelated work; use a supported correction or independent review rather than a blind revert.
- The coordinator resolves decisions within existing authority. Ask the user only when a required scope,
permission, product, or consequential tradeoff decision cannot be resolved from available evidence.
A failed check stays failed until it passes; diagnosis or escalation never waives a required gate.
- Do not claim a failure is fixed until the command passes locally.
7. Report required commands, results, and any escalation reason. When the wrapper reports delivery
coverage, preserve its phase, candidate/base identity, selected or full coverage, omitted checks,
and selection reason. A selected pass does not qualify an omitted
promotion check. Once the required checks pass, continue
the authorized task; repeat or broaden verification only after relevant changes, failures, unresolved
concerns, or an explicit repository requirement.
For evidence passed between verification, review, and closeout, read
[Verification evidence reuse](references/verification-evidence.md). Reuse is conservative and
never overrides a required repeat, current-head CI, or independent review.
## Default Verification Order
Use this order unless the changed files make a narrower or broader set clearly appropriate:
1. Formatting and whitespace:
- `git diff --check "origin/${BASE_BRANCH}...HEAD"` for committed branch content before creating or updating a PR; detects trailing whitespace and conflict markers, not source formatting
- `.agents/bin/lint` when present, or the repo's documented formatter check
2. Mandatory pre-commit gate:
- `.agents/bin/validate` - **mandatory gate before every commit/PR update**, using the repository's documented coverage; promotion requires its complete invocation
3. Ruby (or the repo's equivalent backend language):
- the repo's type/signature validation command when signatures or public APIs changed
- the repo's targeted unit-test command for the changed backend behavior
4. JavaScript and TypeScript (or the repo's equivalent frontend/package language):
- the repo's package build command
- the repo's package lint command
- the repo's type-check command
- the repo's targeted package-test command for the changed package, or a targeted single-test-file run scoped to the
changed package
- the repo's full package-test command when broad package behavior changed or the touched files are not covered by a narrower package test
- the repo's end-to-end/browser test command when the branch changes performance- or framework-sensitive areas such
as SSR rendering, client hydration, or browser-visible integration behavior; on fresh Linux environments, install
the e2e browser dependencies first per the repo's e2e setup
5. Docs:
- the repo's docs-sidebar/coverage check when docs under the repo's documented docs directories changed
- the repo's link checker when Markdown URLs were added or edited; do not substitute an ad hoc link checker unless the
branch changes the link checker, its config, or the documented link-check workflow itself
6. CI workflows and YAML:
- `actionlint` when any `.github/workflows/` file changed
- `yamllint .github/` when any `.github/workflows/` file changed
- Do not run the repo's source linter on `.yml` files
7. Broad suite — pick the narrowest command that covers the change:
- the repo's broad-but-fast suite command for broad coverage without the slowest generated/example suites
- the repo's full suite command when shared runtime behavior, generators, cross-package contracts, or release-critical paths changed
- the repo's complete lint gate when a branch intentionally needs the full lint pass across all source languages and
formatting; otherwise keep using the narrower lint commands above
## Scope Guide
- Core library/backend changes: run the repo's targeted unit tests, the mandatory pre-commit lint gate, and type/signature validation when signatures or public APIs changed.
- Integration or test-app changes: run the repo's integration test command or a targeted integration spec scoped to the changed surface. For changes that affect performance- or framework-sensitive areas such as SSR rendering or client-side behavior, also run the repo's end-to-end/browser test command.
- Frontend/package changes: run the repo's package build, package tests, package lint, and type-check commands.
- Generated examples or scripts: run the relevant generator/script command plus formatting and linting.
- Documentation-only changes: run the repo's formatter check, the docs-sidebar/coverage check for the documented docs directories, and the link checker for new or changed URLs. If committing, still run the mandatory pre-commit lint gate; see Instructions step 3 for why this applies even to docs-only commits. The lint gate does not validate Markdown.
- Package-specific frontend changes (for example a separately-packaged area with its own scripts, per `AGENTS.md`): run that package's own local formatter check via its own scripts plus any focused tests for the changed surface.
- Package-specific backend changes: run that package's own lint command (with any package-scoped flags it documents) and any targeted unit tests.
- GitHub Actions workflow changes: run `actionlint` and `yamllint .github/`. Do not run the repo's source linter on `.yml` files.
- Anything not listed above (for example, build-script edits, generator templates, signature-only changes, or build scripts): apply the narrowest set of checks that covers the changed surface and explain the choice in the output.
## Output Format
Use this concise summary. Include delivery coverage and omissions only when
the wrapper reports them; always include required results and any escalation:
```text
Verification:
- Delivery coverage: <phase; candidate/base identities; selected|full>
- PASS git diff --check "origin/${BASE_BRANCH}...HEAD"
- FAIL <repo formatter check>
- Omissions: <checks or none>; reason: <trusted selection or escalation>
Next fix:
- Run the repo's format/autofix command to fix formatting, then rerun the formatter check.
```
If a command is intentionally skipped, explain why in one line. Prefer local verification over waiting for CI.
Pliki w paczce (5)
- SKILL.md 9 KB
- agents/openai.yaml 1 KB
- bin/verification-evidence-reuse 3 KB
- bin/verification-evidence-reuse-test.rb 4 KB
- references/verification-evidence.md 5 KB
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · shakacode/agent-workflows | 100 | MIT |
| yonatangross/orchestkit | 100 | MIT |
| yonatangross/orchestkit | 100 | MIT |
| NolwennD/gabier | 100 | Apache-2.0 |
| ericrisco/rsc-harness | 100 | MIT |
| sentimony/agentic-workflow | 100 | MIT |
| JoDi-27/JoDi-SDD | 100 | MIT |
| marcel-bich/marcel-bich-claude-marketplace | 100 | MIT |
| kdlbs/kandev | 100 | AGPL-3.0 |
Migawka bazy z 13.11.2026 · wersja paczki 320a80b564f5 (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
- 100verification-loop
Weryfikuje kod przed pull requestem: uruchamia kontrole jakości i walidację zmian. - 86lifecycle-loop
Zamienia prośbę w kartę celu, uruchamia jej kontrole i prowadzi pętlę pracy test-first, aż kontrole przejdą lub postęp stanie. - 100verification
Uruchamia pełny zestaw weryfikacji agenta: bezpieczeństwo, wzorce, jakość kodu i kontrole właściwe dla języka. - 100pr-summary
Streszcza bieżący pull request na podstawie diffu, komentarzy i zmienionych plików, przy przeglądzie PR i przed merge. - 100write-frontend-tests
Planuje i pisze testy integracyjne dla zmienionych stron i komponentów frontendu, na podstawie różnicy gałęzi względem dev. - 100create-changelog
Generuje wpis changelogu dla bieżącej gałęzi Git na podstawie zmienionych plików, klasyfikuje domenę i typ, pisze krótki komunikat i po potwierdzeniu uruchamia just changelog add.