Conversation
Enhanced the message interface to include an "error" role. Updated the send function to filter out error messages from the history before sending. Improved error handling to replace the last assistant message with an error message if applicable, ensuring better user feedback during inference errors.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a8b91907ed
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| } | ||
| return [...prev, errorMsg]; | ||
| }); | ||
| track("studio_sse_error", { source: "inference" }); |
There was a problem hiding this comment.
track
The new call to track("studio_sse_error", …) references an undeclared identifier, so @arkor/studio-app no longer type-checks under strict TypeScript (TS2304: Cannot find name 'track'). This breaks CI/build flows that run tsc --noEmit and should be fixed by importing the intended tracker (or removing the call).
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
Pull request overview
This PR updates the Studio Playground chat UI to represent inference failures as dedicated "error" messages, avoiding duplicate/empty assistant bubbles and preventing prior error bubbles from being sent back to /api/inference/chat.
Changes:
- Add an
"error"role to the localMessagetype and emit error bubbles on inference failures. - On failure, replace the empty assistant placeholder when present; otherwise append an error message to preserve any partial streamed output.
- Exclude
"error"messages from the request history sent to/api/inference/chat, and attempt to emit a telemetry event on failure.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| } | ||
| return [...prev, errorMsg]; | ||
| }); | ||
| track("studio_sse_error", { source: "inference" }); |
There was a problem hiding this comment.
track(...) is referenced but not imported/defined anywhere in the Studio SPA, so this will fail TypeScript compilation (and would also throw at runtime if it slipped through). Import the correct telemetry helper (or use the existing telemetry API if one exists) and consider making the call resilient if telemetry is optional.
Summary
role: "assistant"for the error bubble while leaving the empty placeholder in place, so every failed send rendered as twoASSISTANTlabels (one empty + one[error] …). Repeated failures made the chat look like ASSISTANT spam."error"role toMessage. The catch path now (a) replaces the empty assistant placeholder when present, (b) otherwise appends the error so any partial streaming output from before the failure is preserved."error"messages out of the history forwarded to/api/inference/chat— cloud-api'sChatRequestBodyonly acceptssystem | user | assistant, and previous error bubbles shouldn't poison the next prompt's context.track("studio_sse_error", { source: "inference" })on the failure path so Playground inference failures land in the existing SPA telemetry channel.Test plan
pnpm --filter @arkor/studio-app exec tsc --noEmitpnpm --filter @arkor/studio-app test(46 passed)ERRORbubble per failed send, and that the next successful send does not include the prior error in the request body.