You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Follow-up sent during worktree preparation fails instead of queueing #36843
Sending a follow-up while T3 Code prepared a worktree showed an error saying the agent could not be steered. Expected: retain the message in the queue without a user-facing error.
Diagnosis
The trace confirms message.dispatch reached Orchestrator.dispatchSteerIntoRun while the target run was preparing. In server commit 1264c20, CommandPolicy.ts:153-172 queues auto delivery during preparing/starting, but steer delivery bypasses that guard. Orchestrator.ts:5400-5405 rejects any target not running. ChatView.logic.ts:1323-1327 also intends to queue sends during setup. Therefore this dispatch bypassed the intended setup/auto queue handling; the exact client-side cause and deliveryIntent were not captured. Current main retains the same distinction between auto and steer. No confirmed newer fix found.
Steps to reproduce
In the desktop app connected to a remote Linux server, start a new Codex thread using a worktree.
Send a follow-up while the worktree is still preparing.
Observe the cannot-be-steered error rather than a queued message.
Observed once in logs; not independently reproduced. The exact send control/shortcut used is unknown.
Version
Server 0.1.10-nightly.20261111.4544; desktop 0.1.10-nightly.20261114.4682
Environment
macOS desktop connected to a Linux x64 server, kernel 7.0.0-43-generic; Node v26.12.1; Codex provider, codex/gpt-6.1-sol.
Evidence
2026-11-14T16:55:37.316Z initial run requested
2026-11-14T16:55:42.941Z Orchestrator.dispatchSteerIntoRun failed:
Target run run:thread:<redacted>:ordinal:1 is preparing and cannot be steered.
2026-11-14T16:55:55.850Z a later run was queued
2026-11-14T16:56:02.577Z initial run started
Trace also records message.dispatch and ThreadMessageIntake.dispatchCommand failures for the rejected send.
Related issues
#34158 / #34163 fix auto sends while stopping, a different lifecycle state and MCP entry point. No matching preparing/worktree issue found.
Fix applied or workaround
Explicitly choose Queue during preparation, or wait for the agent to start before sending. A later queued run was observed before the initial run started. No install changes or local source patches applied.
Activity
juliusmarminge commentedon Nov 14, 2026
This is the same problem as #34581, so I'm closing it in favour of that one. If it is different, say how and it will be reopened.