Unfortunately it looks like we may need another wait here, as agentapi will become stable before it starts processing the task. We probably need to wait until the prompt shows up in the messages endpoint or something.

This ended up looking very similar to what we'd wantcoder exp task status --watch to be. Will update in a separate PR.

I know we have the required envs check, but I'm wondering if it'd be better to define thesetraiage.sh envs in one place so they're applied to all action steps?

Mayhaps doing something like this early on:

echo "APP_SLUG=$APP_SLUG" >> "$GITHUB_ENV"

(There's probably a better way to do it.)

Or write atraiage.config.sh which can be sourced intraiage.sh.

Long-term I want to remove the need for this (e.g by usingcoder task send). Leaving it here, exposed and ugly, makes it easier to delete later IMO.

We could be fancy, look up the app URL and do assh bash 'while :; curl; if ...; fi; done'.

The only motivation here is to reduce coderd load, though.

As with app_slug above, I'm disinclined to do fanciness if it can later be incorporated into the CLI.