[Windows Desktop] Long local task can remain stuck at Thinking after reconnect/resume

Open
#35,203 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
42/100
Issue type
Bug
Clarity
Needs clarification
Activity status
Active
Tech stack
rust
Domain
desktop

Research direction

No source files or tests are named. Start by tracing Windows Desktop reconnect/resume handling and the app-server turn lifecycle, focusing on the logged turn/interrupt and duplicate turn/start sequence; use the reproduction pattern and sanitized logs to follow state transitions. Done means one active turn survives reconnect without prompt duplication or interruption, or a recoverable error is shown.

Written by the indexing model from the issue text.

Description

app app-server bug windows-os
Environment
  • Surface: Codex Desktop for Windows (Microsoft Store package 26.721.3996.0)
  • Bundled Codex runtime observed by the app: 26.721.31836
  • Model: GPT-5.6 Terra, medium reasoning effort
  • Platform: Windows 11 x64
Summary

Fresh local tasks respond normally, but an existing long local task can become permanently stuck at “Thinking” after a reconnect/resume. The affected task cannot accept a reliable follow-up; the desktop client may duplicate the prompt/turn or interrupt the active task instead of resuming it.

Reproduction pattern
  1. Start a local Codex Desktop task and use it over a long project workflow.
  2. A reconnect/resume occurs while the task has substantial history.
  3. Send a normal follow-up, or reopen/resume the task.
  4. The UI remains at “Thinking” indefinitely. In the affected case, the same diagnostic prompt appeared twice and two turns were left in progress.
Observed evidence
  • Desktop/app-server logs show client-issued turn/interrupt events for the active task, followed after resume by a duplicate turn/start.
  • The resulting stream retried after a WebSocket close without reaching response.completed.
  • Fresh tasks and a direct ephemeral Codex CLI request against the same local project, account, model family, and network completed successfully. This points to desktop task recovery/state handling rather than the workspace sandbox or general connectivity.
  • No project files need to be read or included to reproduce the symptom.
Expected behavior

A reconnect should preserve one active turn, avoid replaying or duplicating prompts, and either continue the task or expose a recoverable error. Long-running project work must retain a safe continuation path without relying on an unrecoverable old chat.

User impact

This blocks long-running project work and risks loss of unrecorded decisions. A new task is an imperfect workaround because it cannot restore the old task's in-flight reasoning or task-local state.

Request

Please investigate Windows Desktop task resume/reconnect idempotency, especially duplicate turn/start / interrupt transitions for older long tasks. A feedback session and sanitized log timestamps can be supplied through the in-app feedback flow if needed.

Dominant language
Rust
Stars
125k
Forks
19.5k
Avg merge
1m
Merged PRs (30d)
1k

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from openai/codex

All issues in openai/codex

Similar issues

More Rust issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.