Repository navigation
Replies: 2 comments
|
Reproduced on the current desktop release, with the same result as the original post. Version: desktop 0.1.1, macOS arm64. The blocking clause in In real use: a desktop update restarted the server while 8 runs were active. 6 were continued. The 2 that were waiting on Isolated repro (separate
Result: the request is Before the restart: After the restart: Not continuing matches #18461 and Investigated with Claude Opus 5.5 in Claude Code. |
|
Closing: current main already covers this as of 867112cda0. If you still see it on a build that includes that commit, reply and we'll reopen. |


Uh oh!
There was an error while loading. Please reload this page.
A pending native question disappears after the T3 server restarts. I reproduced this with a single server on
cfafb4bb54d244e4612f5681aad3832ea64cb0fb: Claude asks throughAskUserQuestion, the server is killed, startup marks the persisted requestexpired/not_resumable, and reopening the thread shows no answer card. The question text is already durable, but the native reply callback died with its process.Proposed scope: preserve unanswered native questions across graceful shutdown and crash recovery, and accept the user's eventual answer through the existing message-response path. That starts a new run in the same conversation instead of attempting to revive the old callback.
This changes the current recovery policy, so I am requesting explicit approval of this direction and scope rather than treating expiration of native callbacks as a defect. Does preserving the durable question through message responses fit the intended restart behavior?
Prepared and investigated with GPT-6.1-Sol and GPT-6-Astra through the Codex harness.
All reactions