Windows Desktop /goal runs stream reasoning for about 50 minutes with no tools or answer

Finding: In an external report on Codex Desktop for Windows 26.1002.7124.0 (model gpt-6.1-sol, High effort, ChatGPT sign-in), two long /goal conversations reportedly stayed in Thinking after their last completed tool call. The reporter says reasoning items kept arriving for about 50 minutes with no new commands or answer (https://github.com/openai/codex/issues/52529). This is an external report from the reporter's local logs; KFA did not reproduce it. Reported details: In conversation A, the last tool finished at 18:08:28 JST and reasoning continued until 18:58:32. In B, the last tool finished at 17:19:53 and reasoning continued until 18:09:18. Both needed a manual interrupt and Goal pause. After A was resumed, shell checks completed normally, but 92 more reasoning items reportedly followed with no tool calls and the turn stayed inProgress through at least 19:34:47. The reporter says a steering message sent about 19:26 was logged as a Steer submission for the active turn but got no timely response (https://github.com/openai/codex/issues/52529). Context the reporter gave: total_usage_tokens was 708520 against an auto_compact_scope_limit of 720000 and a 760000 context window. They say proximity to compaction may be relevant but the root cause is not established. No explicit rate-limit or context-window error was seen in the inspected section (https://github.com/openai/codex/issues/52529). A later commenter reported a possibly related macOS Ghostty CLI case: codex-tui 0.160.0 (0.162.0 installed, not shown to reproduce), model gpt-6.1-sol, medium effort, no /goal. A turn of about 35 minutes reportedly had 86 reasoning items and 44 completed tool calls. No user-facing progress appeared for the final ~10.5 minutes before the user interrupted. The commenter says this is a shorter stall on a different surface, not an identical reproduction, and that the cause is unconfirmed (https://github.com/openai/codex/issues/52529#issuecomment-6087093840). Status: the issue is open, with no maintainer-confirmed cause, fix commit or released fix in the supplied material. The only reported mitigation is to interrupt and pause the Goal. That did not prevent recurrence after resume in case A. Possible diagnostics the reporter suggests are checking whether the thread history shows reasoning items with no subsequent tool calls, and whether a progress watchdog or clearer diagnostic would help. These are suggestions, not confirmed fixes. Limitation: all evidence is external report; the macOS case differs in surface, version and duration. There is no KFA reproduction and no independent confirmation.