Readiness checks can run past the two-minute timeout

What happened

A readiness check can keep a preview or browser test waiting well past the two-minute deadline.

The polling loop checks the deadline between attempts, but each environment.ready command gets the usual 30-minute command timeout. It also accepts a successful result that arrives after the deadline. The first check when reopening a prepared workspace has the same problem.

I'd expect the deadline to stop a hanging check, including that first check.

How to reproduce

1. Set environment.ready to sleep 180.
2. Start the environment or reopen a prepared workspace.
3. Wait for the readiness check.

The command can keep running beyond the two-minute deadline and still be accepted when it exits successfully.

How are you running Facility?

Other. Checked from a source checkout in a local sandbox.

Version

Commit 6338c8c4259b63df50efb28b904907ecc3388e94.

Evidence

Readiness loop and generic command timeout, prepared-workspace probe. I traced this in the source. The regression test uses a shorter deadline and a real sleeping subprocess; I haven't completed the real Docker workspace check because the runner won't start in the ARM sandbox.

Start in services/api/src/workspaces/project-environment.ts, reading the readiness loop around lines 526-574 and the prepared-workspace probe around lines 285-294. Reproduce with environment.ready set to sleep 180, then inspect the existing regression test setup with its shorter deadline and sleeping subprocess. Done means both readiness paths stop at the two-minute deadline and do not accept a result that arrives afterward.