Auto Drive failing with OR

[elsbrock]

What version of Code is running?

code 0.4.3

Which model were you using?

openai/gpt-5-codex:nitro

What platform is your computer?

Darwin 25.0.0 arm64 arm

What steps can reproduce the bug?

I elaborated a plan and did /auto.

What is the expected behavior?

The plan gets carried out by Auto Drive.

What do you see instead?

Coordinator response invalid (attempt 1/3): response was not valid JSON
Coordinator response invalid (attempt 2/3): response was not valid JSON
Coordinator response invalid (attempt 3/3): response was not valid JSON

Afterwards, it showed "SUCCESS" (or something like that), but it actually did nothing.

I pressed some keys and it showed this:

   ╭─ Auto Drive · Stopped
   │
   │  Deriving goal from recent conversation
   │
   │  Actions
   │   00:00  Auto Drive started: Deriving goal from recent conversation
   │   00:52  Progress: Coordinator error: model response was not valid JSON
   │   00:52  Coordinator error: model response was not valid JSON
   │
   ╰─ [Ctrl+S] Settings · [Esc] Stop

  Auto Drive > Coordinator error: model response was not va ⣻ Stopped (52s, 1 turn)

Additional information

Using OpenRouter (not ChatGPT sign-in/API key)

[zemaj]

Heads-up: sanitizer for the Auto Drive tool-call JSON failure is staged in PR #355. Once that lands in a nightly build, could someone on OpenRouter reproduce /auto and confirm the coordinator no longer flags "response was not valid JSON"? A quick run on the same plan that previously failed would be perfect. Thanks for the original report—looking forward to your 👍 once the fix is available.

[elsbrock]

Sure thing, will have an eye on #355 and retry.