-
Notifications
You must be signed in to change notification settings - Fork 201
Open
Description
Description
When composing an issue and sending a prompt, aborting mid-generation appears to stop the stream, but after a few seconds the full AI response still renders. This defeats the abort action and leads to confusing UX and wasted tokens/compute.
- Impact: Users cannot reliably stop responses; unexpected content appears post-abort.
Steps to reproduce
- Open the app and start composing a new issue.
- Type any prompt and click Send.
- Before the response finishes streaming, click Abort.
- Observe that streaming stops initially, then the complete response appears a few seconds later.
Expected behavior
- Aborting cancels the active request both client- and server-side.
- No additional tokens or final message should render after abort.
- UI may show “Generation cancelled,” optionally preserving partial output per product decision.
Actual behavior
- UI stops initially, but after a short delay the full model response renders as if abort never happened.
Frequency
Consistently reproducible.
Relevant files
src/server.ts(usescreateUIMessageStream)src/client.tsx(stream handling / Abort UI)src/tools.ts(if tools trigger server streams)
Metadata
Metadata
Assignees
Labels
No labels