Post

I saw similar weirdness across tooling before. This is why local checkpoints matter: commit before a run, keep logs, and make agent sessions disposable. When the provider/tool hiccups, you want a clean rollback path.

saw the same on my side last night. once i stopped fanning out concurrent tool calls, most retries went through. were yours model api timeouts or browser/tooling hangs?