release-dispatch
/release-dispatch
Uruchamia wydanie Legend w GitHub Actions po kontroli wstępnej i potwierdzeniu, monitoruje je do końca i analizuje błędy.
- Autor
- finos ↗
- Repozytorium
- legend-shared ↗
- Licencja
- Apache-2.0
- Ostatnia zmiana repozytorium
- 27.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 27.10.2026, bez odwołań do wycofanych modeli i API.
Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.
Co robi i kiedy
Uruchamia wydanie Legend w GitHub Actions po kontroli wstępnej i potwierdzeniu, monitoruje je do końca i analizuje błędy.
Wyzwalacze z opisu
Oryginalny opis (EN)
Dispatch a Legend release workflow and babysit it to completion — picks the right workflow, runs preflight, confirms, triggers, then watches and triages any failure. Use for "release 4.143.1", "cut a release off this branch", "run the release action", "kick off the release".
Jak działa
- 1. Establish the target
- 2. Preflight
- 3. Confirm before dispatching
- 4. Dispatch
- 5. Babysit
- 6. On failure
- 7. On success
Podgląd SKILL.md
--- name: release-dispatch description: Dispatch a Legend release workflow and babysit it to completion — picks the right workflow, runs preflight, confirms, triggers, then watches and triages any failure. Use for "release 4.143.1", "cut a release off this branch", "run the release action", "kick off the release". --- # Release dispatch Dispatches a release and stays with it. The point is that the user does not have to watch a 1–3 hour pipeline or discover a failure an hour after it happened. Read `../release-preflight/references/repo-matrix.md` (relative to this skill's directory) before anything. ## 1. Establish the target Repo from `git remote get-url origin`, branch and version from the user. If they gave a version but no branch, use the current branch and say which one you are using. **Pick the workflow from the matrix, not from the file listing.** `release.yml` is correct for all five repos. Do not use `release-large-runner.yml` (dead since February 2026) or `legend-stack-release.yml` (stale, `repository_dispatch`-driven). Verify with actual history rather than assumption: ```bash gh run list --workflow=release.yml -L 5 ``` ## 2. Preflight Run the `release-preflight` skill's checks in full. If anything blocks, stop and report — do not dispatch "to see what happens", because a failed release after `release:prepare` leaves a pushed tag and version commits to clean up. ## 3. Confirm before dispatching Dispatching publishes to Maven Central, which is **immutable**. Always confirm, stating plainly: - repo, branch, version, workflow - whether it is a real publish or a dry run - the expected duration from recent runs of that workflow **The dry-run trap** (`legend-pure` and `legend-engine` only): `dryRun: true` skips only the *publish* steps. `release:prepare` still runs, so it still pushes the tag and the version-bump commits. A dry run followed by a real run therefore needs a full `release-cleanup` in between. Say this whenever the user asks for a dry run — most people expect it to be side-effect free, and it is not. ## 4. Dispatch ```bash gh workflow run release.yml --ref <branch> -f releaseVersion=<version> ``` Add `-f dryRun=true` only where the input exists and the user asked for it. `gh workflow run` prints nothing on success, so confirm it actually started: ```bash gh run list --workflow=release.yml -L 3 ``` Report the run ID and URL. ## 5. Babysit Poll rather than blocking on `gh run watch`, which holds the terminal for hours: ```bash gh run view <run-id> ``` Sensible cadence is every few minutes early on, then spaced out — the build job alone is over an hour on legend-engine. For a long unattended watch, the `loop` skill fits well. Report meaningful transitions only: the build job finishing (this is the point the tag has been pushed and the branch bumped), the test matrix starting, the publish job starting. Do not narrate every poll. ## 6. On failure Hand straight to `release-triage`. Do not guess at a cause from the job name — a single failed job routinely mixes a real failure with an environmental one. Report the verdict and the recommended action; do not rerun or "fix" anything without asking. ## 7. On success Hand to `post-release-verify`.
Wersje
W bazie nie ma innych skilli o nazwie release-dispatch.
Migawka bazy z 13.11.2026 · wersja paczki 0428907bc789 (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
- 100release-triage
Diagnozuje nieudane wydanie lub przebieg CI Legend: pobiera nieudane zadania, odróżnia prawdziwe błędy kodu od znanych problemów środowiskowych i rekomenduje działanie. - 100release-cleanup
Cofa nieudane wydanie Legend: usuwa tag i commity podbicia wersji, by można było wydać ją ponownie. - 100release
Samodzielnie prowadzi wydanie Rove na npm: wykrywa podbicie z changesetów, monitoruje publikację w GitHub Actions i diagnozuje błędy CI. - 100cut-release
Przeprowadza wydanie agnostic-ai od początku do końca: tag i wydanie na GitHubie, gdy użytkownik chce wypuścić nową wersję. - 100ci-troubleshoot
Diagnozuje nieudane przebiegi GitHub Actions dla pi-agent-dashboard: taksonomię 11 workflowów, dobór testów, pipeline wydań, znane awarie oraz czytanie logów gh run i ponawianie zadań. - 100cut-new-release-candidate
Wyzwala kandydata do wydania dla gałęzi release, nie dla pełnych wydań, hotfixów ani innych przepływów dispatch.