cronward
/cronward
Ustawia workflow GitHub Actions w cronward, aby uruchamiał go z własnej maszyny według harmonogramu, i sprawdza, czy przebieg się udał.
- Autor
- Frestina ↗
- Repozytorium
- cronward ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 07.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 07.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Ustawia workflow GitHub Actions w cronward, aby uruchamiał go z własnej maszyny według harmonogramu, i sprawdza, czy przebieg się udał.
Wyzwalacze z opisu
Oryginalny opis (EN)
Put a GitHub Actions workflow on cronward, which starts it on a schedule from the user's own Linux machine through workflow_dispatch and says whether each run passed; or test, check or remove such a job. Use when asked to schedule a workflow with cronward, to make a workflow's schedule reliable because GitHub's `on: schedule` runs late or not at all, or whether a cronward job passed.
Jak działa
- Rules
- Find the machine cronward runs on
- Put a workflow on cronward
- Other tasks
- Exit codes
Podgląd SKILL.md
---
name: cronward
description: Put a GitHub Actions workflow on cronward, which starts it on a schedule from the user's own Linux machine through workflow_dispatch and says whether each run passed; or test, check or remove such a job. Use when asked to schedule a workflow with cronward, to make a workflow's schedule reliable because GitHub's `on: schedule` runs late or not at all, or whether a cronward job passed.
---
# cronward
GitHub's `on: schedule` runs late, or not at all. cronward is a timer on the
user's always-on Linux machine that starts chosen workflows on a schedule
through the API (`workflow_dispatch`), follows each run to its end and says
whether it passed. The runs still happen on GitHub. Each job is one workflow
in one repo with one cron line, kept in the config on that machine.
These steps work for any agent: they are shell commands and file edits.
## Rules
- **Never handle a token.** Don't ask for one, read one, print one or store
one. The user stores tokens themselves, in their own terminal, with
`cronward token set <account>`. When a token is missing or lacks access,
say what to do and stop.
- **Use `cronward add` and `cronward remove`** rather than editing the
config by hand. They check the job with GitHub first, and keep the file's
comments.
- **A test run is a real run.** It runs the workflow on GitHub, with
whatever it does: deploys, releases, messages. Ask before starting a
workflow that has effects like that.
- Don't install cronward, its timers or lingering unless the user asks.
## Find the machine cronward runs on
Jobs live on the main machine, which holds the config and the tokens. Run
`add`, `remove`, `run`, `status`, `history` and `doctor` there.
1. If `systemctl --user cat cronward-tick.timer` succeeds, this is the main
machine: run `cronward` here (`~/.local/bin/cronward` if it is not on
`PATH`).
2. Otherwise, if `systemctl --user cat cronward-watch.service` succeeds, this
machine is a viewer, and its `ExecStart=` line ends with `watch <host>`.
Run each command there by putting `--on <host>` first:
`cronward --on <host> status`. It runs cronward over the user's ssh and
passes the exit code through; 2 with "ssh failed" means ssh could not
log in (no key, a passphrase), and "this key only reads the status"
means ssh logged in with the viewer's read-only key. Then ask the user
to run the commands, or to make ssh to `<host>` work for you.
3. Otherwise cronward is not set up for this user. Tell them, point them to
the install steps in cronward's README, and stop.
## Put a workflow on cronward
1. **Pick the workflow and the schedule.**
- The repo is `owner/name`, from `git remote get-url origin`.
- The workflow is its file name under `.github/workflows/`, such as
`check.yml`.
- The schedule is a 5-field cron line: the `cron:` under `on: schedule:`
if the workflow has one, or what the user asked for. It may fire at
most every 5 minutes.
- GitHub's cron lines are UTC, and so is cronward by default. If the user
gives a time in their own zone, pass `--timezone` with an IANA name
such as `Europe/Oslo`.
- One job has one cron line. For a workflow with several, add one job
per line.
2. **Make the workflow startable.** Its `on:` needs `workflow_dispatch:`.
If `on:` already has `workflow_dispatch:` and no `schedule:`, the repo
needs no change: skip this step and step 5, and say in step 6 that
nothing was committed. Otherwise add it beside the other triggers,
turning a short form such as `on: push` or `on: [push]` into a mapping.
Leave `schedule:` in place for now.
- If `workflow_dispatch` has required inputs without a default, plan to
give each with `--input KEY=VALUE`.
- The change must be on the default branch (or the branch given with
`--ref`) before cronward can start it: GitHub only knows workflow files
on the default branch. Commit and merge it the way this repo does that,
and ask before pushing if you are unsure.
3. **Add the job.**
```sh
cronward add <job> --repo <owner/name> --workflow <file> --schedule "<cron>"
```
Name the job after the repo and what it does, such as `widget-nightly`:
letters, digits, `.`, `_` and `-`. Other options: `--account`, `--ref`,
`--timezone`, `--finish-within` (such as `2h`), `--input KEY=VALUE`.
| Result | What to do |
| --- | --- |
| Exit 0 | Added. It prints the next slot. |
| Exit 1, "no account's token reaches" | Stop. Ask the user to add the repo to their token's repository access at the page it prints (or, for a repo of another owner, to store a token with `cronward token set <account>`, which also adds the account to the config), then run `add` again. You cannot do this for them. |
| Exit 1, "cannot read the workflows" | Stop. The token reaches the repo but lacks Actions: read and write. Ask the user to give it that at the page it prints, then run `add` again. |
| Exit 1, "not a workflow on ... default branch" | Step 2 is not merged yet. |
| Exit 1, "turned off in GitHub" | The user turned the workflow off. Ask before anyone turns it on. |
| Exit 2 | A wrong value, or several accounts reach the repo (then pass `--account`). The message says which. |
4. **Test it**, after asking if the workflow has effects (see Rules):
```sh
cronward run <job> --wait 8m
```
Keep the time under your own command time limit. It prints the run's URL
and exits 0 only if the run passed. On exit 1 it prints why:
- "still running": the timer follows it. Look again later with
`cronward status --json`.
- "add `workflow_dispatch:`": step 2 is not on the branch it ran.
- "may not do this": the token lacks Actions: read and write. The user
changes the token's permissions on GitHub.
- failed: the workflow itself failed. That is the repo's to fix, not
cronward's.
5. **Take out `schedule:`** once a run has passed, or GitHub's own
scheduled runs come on top of cronward's on the days they do fire. Keep
`workflow_dispatch:`. Commit it the same way as step 2.
6. **Tell the user** the job's name, its schedule and zone, the next slot
in their own local time, and the test run's URL. The next slot `add`
prints is in the main machine's local time, not UTC or the job's zone;
`next` in `status --json` is a Unix time. They get a notice when a job's
result changes to failed or didn't run, and when it passes again.
## Other tasks
- **Check jobs:** `cronward status --json`. For each of `jobs[]`: `name`,
`latest` (the newest final result: `kind` is `passed`, `failed` or
`didnt_run`, with `reason` and `url`), `running` (a run still going),
and `next` (a Unix time). `last_tick` and `tick_error` say whether the
timer works. `cronward doctor` checks everything and marks problems
`FAIL`. On a viewer, `cronward status --host <host> --json` shows the
status it last fetched, with no ssh: its `viewer` object says when
(`fetched`), and whether the main machine could not be reached
(`unreachable_since`, `error`) or had stopped ticking. `cronward watch
<host>` fetches it again.
- **Past results:** `cronward history <job> --json`, the job's last 50
results newest first, each with `slot`, `kind`, `reason` and `url`, for
when a job has failed before or passes only sometimes.
- **Take a job off:** `cronward remove <job>`. Put `schedule:` back in the
workflow if GitHub should run it on its own again.
- **Change a job:** `cronward remove <job>`, then `cronward add` it again
with the new values.
- **Pause a job:** that is `enabled = false` in its table in the config.
Ask the user, or take it off.
## Exit codes
0 fine; 1 a result that is not passed (`run --wait`, `status`), a check
GitHub refused (`add`) or a failed check (`doctor`); 2 an error, with the
reason on stderr.
Wersje
W bazie nie ma innych skilli o nazwie cronward.
Migawka bazy z 13.11.2026 · wersja paczki 4569cf287be8 (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
- 90triaging-ci-failures
Wyjaśnia, dlaczego przebieg GitHub Actions się nie udał: pokazuje zadania, kroki, istotne linie logu i czy awaria jest losowa. - 100ci
Uruchamia lokalnie sprawdzenia z workflow GitHub Actions przed pushem lub pull requestem, aby odtworzyć wyniki CI na własnej maszynie. - 100github-actions-debugger
Debuguje i monitoruje przebiegi workflowów GitHub Actions: sprawdza status, pokazuje logi nieudanych zadań i wskazuje przyczynę awarii CI. - 93schedule
Tworzy, wypisuje i usuwa zaplanowane zadania agenta w GitHub Actions, generując pliki YAML dla cyklicznych workflow. - 100fix-failing-actions
Diagnozuje i naprawia nieudane workflowy GitHub Actions w repozytorium drift: błędy CI, kontroli higieny bezpieczeństwa i workflowu wydań. - 100analyze-github-action-logs
Analizuje ostatnie uruchomienia workflow GitHub Actions, aby wykryć wzorce, błędy i możliwe usprawnienia.