# Run a checklist (/studio/run-checklists)

# Run a checklist [#run-a-checklist]

```text
/stuntdouble:run <checklist>
```

## What happens [#what-happens]

1. Studio checks the checklist file. If a field is wrong, it names the field
   so your agent can fix it and try again.
2. It prints a link to the [viewer](/studio/viewer), where you can watch the
   run and stop it.
3. An actor opens Chrome on your machine and goes through the flow. Page text
   and screenshots stay with the actor, so your main session stays short.
4. Studio reports the result, for example "failed. 3/4 checks passed",
   then each failed or unjudged check with its evidence, and the viewer link.
   It offers to fix what failed and run again.

If the checklist sets `local_base_url` in `stuntdouble.json`, a local run
swaps the checklist URL's origin for it. That's how one checklist tests
localhost on your machine and a preview URL in the cloud.

## Scores [#scores]

A checklist with only pass/fail checks has no score: it passes when every
pass/fail check passes. A checklist with one or more weighted checks gets a
score. It passes when the score meets `pass_threshold` (100 if you leave it
out) and no pass/fail check failed. Informational checks are recorded, never
scored.

## Verdicts are self-judged [#verdicts-are-self-judged]

In Studio, the agent that went through the flow also decides whether each
check passed. That's fast and free, and good for catching what you
broke. For a second opinion, a [cloud run](/studio/cloud-runs) adds an
independent judge.

## Resume a run [#resume-a-run]

If a run ends early, because you closed the session, the actor was
interrupted or you stopped it from the viewer, carry on where it left off:

```text
/stuntdouble:run <checklist> resume
```

Checks already judged keep their verdicts, and the actor judges only the ones
left.