# Your own check Step

Run your Project's lint, typecheck and tests in the check Step, every one even when an earlier one fails, for the review Step to read.

The starter's check Step runs `bun run check`. In a Project that checks itself some other way, point the Step at a script that runs your checks:

```toml
# In a copy of .vago/workflows/small.toml
[[steps]]
id      = "check"
name    = "Checks"
type    = "shell"
run     = "sh .vago/scripts/check.sh"
on_fail = "continue"
```

`.vago/scripts/check.sh` runs each check even when one before it failed, so the log holds every failure, and exits non-zero if any did:

```sh
set -u

failed=0
npm run lint || failed=1
npx tsc --noEmit || failed=1
npm test || failed=1
exit "$failed"
```

With [`on_fail = "continue"`](/docs/workflows/#on_fail) a failed check doesn't stop the Ticket: the review Step reads the whole log through [`{{steps.check.output}}`](/docs/workflows/#step-outputs), which starts with the exit code, and fixes what failed. With `on_fail = "stop"`, a failed check leaves the Ticket stuck at this Step for you to look at.

Keep the id `check`. The starter's review prompt reads `{{steps.check.output}}`, and Vago refuses a Workflow whose prompt reads a Step it doesn't have.

## Caveats

- The starter's prompts tell the agents to run `bun run check` before they finish. Copy `small-implement.md`, `small-review.md` and `goal-review.md` under new names, change `bun run check` to `sh .vago/scripts/check.sh` in each, and point your Workflow's Steps at the copies. A `check` script in `package.json` that runs all three also works, since `bun run check` runs it.
- A check that runs for long without printing anything is fine: [`idle_timeout`](/docs/workflows/#idle_timeout) applies to agent Steps, not shell Steps.