Check the setup hook in the cloud

The fresh-clone check runs the hook on the user's machine. This check runs it on the cloud machine before the change reaches the default branch. Each check is one cloud run that runs the script to prove it works before it merges, and it uses credits; say so in one line when you start it, without asking. Run it again after each fix to the hook.

The check runs a branch's hook, so it starts only after the user has agreed to push when going live: the setup branch, or, when the user pushes straight to the default branch, a temporary branch named tessl-check-hook, deleted once the check is done. Push to the default branch only once the hook passes.

1. Launch the check from the default branch, with the environment and the agent and model the actions use, naming the branch to check:

   tessl launch skill tessl-labs/factory-setup#check-cloud-setup --cloud \
     --repo <owner/repo> --base-branch <default branch> \
     --environment <environment> --agent <agent> --model <model> \
     --instructions "Branch: <branch>" \
     --label "Check the setup hook" --no-wait --json

   <environment> is the one the automations use: <repo>-factory, or, for automations set up in Tessl, a Loop's own, from attributes.environment.name in tessl schedule view <loop id> --json: the "Implement ticket" Loop's when there is one, otherwise the "Learn from agent runs" Loop's (named "Agent Retro" when set up earlier). The check only reads the branch, so read access is enough. When no Loop has an environment, create <repo>-factory with --github-access read for it. Take <agent> and <model> from a pair tessl launch models --json lists (it spells claude as claude-code), or leave --model out when the list has a default pair. The default branch has no hook yet on a first setup, so the run starts and the agent runs that branch's hook itself. A hook that fails while Tessl prepares the run leaves no output; one the agent runs keeps it. Keep launchRunId and give the user the run's link, data.webUrl from tessl launch view <id> --json.

2. Once a minute, for up to 10 minutes, read tessl launch view <id> --json.

3. When data.attributes.status is completed and data.attributes.outcome.status is succeeded, give the user data.attributes.result.summary: the hook passed twice, and what the machine has. Then go back to the step that asked for the check.

4. When data.attributes.outcome.status is failed, show data.attributes.result.summary and the hook's output in data.attributes.result.reason, fix the hook on the branch, push, and check again.

5. When the run itself fails, read tessl launch progress <id> --filename prepare-subphases.json: first_failed names the step, and cloud-launch-setup means the default branch's own hook failed.

6. For any other result (cancelled, no outcome, or still pending after 10 minutes), say so, give the run's link, and say the hook is unverified.