[Bug]: t3_thread_send mode "restart" always fails on a running Claude thread ("cannot redirect an active run")#32890

Before submitting

• I searched existing issues and did not find a duplicate.
• I included enough detail to reproduce or investigate the problem.

Area

apps/server

Steps to reproduce

1. Start a Claude Agent thread and have it run a long foreground command, for example python3 -c 'import time; time.sleep(150)'.
2. From another thread, call t3_thread_send {threadId, mode: "restart", message: "…"} on it while that turn is running.
3. Repeat step 2 against a Codex thread whose turn is running.

Expected behavior

t3_thread_send documents restart as "interrupt and restart". The running Claude turn should be interrupted and the message should start as the next turn, as happens on a Codex thread.

Actual behavior

On the Claude thread, every restart fails:

The Claude driver reports these session capabilities: supportsInterrupt: true, supportsActiveSteering: true, supportsSteeringByInterruptRestart: false. So decideSteeringExecution with forceRestart rejects the call as interrupt_restart_steering. The same call on a Codex thread restarts the turn.

As a result, an agent that wants to interrupt another thread has to know which driver the target uses, or parse this message, before it can call t3_thread_interrupt and send again with queue. T3 can already interrupt a Claude turn. Could restart, on a driver without native interrupt-and-restart, do exactly that: interrupt the turn, then start the message as the next turn? If not, could the tool description say that restart works on Codex only, and say what to do instead?

Impact

Minor bug or occasional failure

Version or commit

T3 Code (Beta) Desktop 0.1.7

Environment

macOS; codex-cli 0.167.0; Claude Code 2.1.319

Workaround

t3_thread_interrupt on the Claude thread (task_cancel for a delegated child), then t3_thread_send with mode: "queue".