**Bug Description** 1. The safety classifier was intermittently rejecting tools—the same command would fail and then work seconds later without any changes on our end. 2. Five subagent freezes, exactly at 600 seconds, with two distinct causes—some were actual retry loops due to issue 1, but at least one was a false positive: the command completed successfully, yet the watchdog still terminated the stream while a 1.6-minute Playwright script was still running. 3. A separate crash due to a lost c

I need more information to generate a GitHub issue title. Could you please provide the bug report details? I need: - A description of the problem - Error messages (if any) - Steps to reproduce - Expected vs actual behavior - Any relevant context about Cla

Status Open

Reported on v2.1.221

Maintainer reply None cached

Activity 0 comments · opened Aug 26, 2026

Bug Description

The safety classifier was intermittently rejecting tools—the same command would fail and then work seconds later without any changes on our end.
Five subagent freezes, exactly at 600 seconds, with two distinct causes—some were actual retry loops due to issue 1, but at least one was a false positive: the command completed successfully, yet the watchdog still terminated the stream while a 1.6-minute Playwright script was still running.
A separate crash due to a lost connection, unrelated to the above.

Environment Info

Errors

[]