Fix autosend chat hook logic #1900
Conversation
There was a problem hiding this comment.
Pull request overview
This PR fixes an infinite retry loop that occurred when backend/MCP tools returned data exceeding the LLM's token limit. The solution replaces the generic lastAssistantMessageIsCompleteWithToolCalls function with a custom hasCompletedUIToolCalls that only triggers auto-send for UI tools (prefixed with 'ui-'), preventing automatic retries for backend tool responses that may be too large.
Changes:
- Added
hasCompletedUIToolCallsfunction to filter tool calls by UI prefix before auto-sending - Replaced
lastAssistantMessageIsCompleteWithToolCallswithhasCompletedUIToolCallsin the chat hook - Added comprehensive test suite covering UI/backend tool scenarios and the specific "Prompt is too long" edge case
Reviewed changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.
| File | Description |
|---|---|
| packages/react-ui/src/app/features/ai/lib/tests/chat-utils.test.ts | New test file with 572 lines covering all scenarios for the UI tool detection logic |
| packages/react-ui/src/app/features/ai/lib/chat-utils.ts | Implements hasCompletedUIToolCalls function with UI tool prefix filtering |
| packages/react-ui/src/app/features/ai/lib/assistant-ui-chat-hook.ts | Updates sendAutomaticallyWhen to use the new hasCompletedUIToolCalls function |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| } | ||
| } | ||
| }, | ||
| sendAutomaticallyWhen: lastAssistantMessageIsCompleteWithToolCalls, |
There was a problem hiding this comment.
There are many references in AI SDK to use sendAutomaticallyWhen: lastAssistantMessageIsCompleteWithToolCalls
However, this function seems to have an issue:
vercel/ai#12099
https://ai-sdk.dev/docs/ai-sdk-ui/chatbot-tool-usage
For example, when the last message is this one, lastAssistantMessageIsCompleteWithToolCalls will return true.
{
"id": "PQ65e1VsSDZxYuS9",
"role": "assistant",
"parts": [
{
"type": "reasoning",
"text": "Searching OpenOps documentation for available AWS templates...",
"state": "done"
},
{
"type": "step-start"
},
{
"type": "tool-OpenOps_Documentation",
"toolCallId": "toolu_vrtx_014xu4P2LtCC84JMGmciw6X6",
"state": "output-available",
"input": {
"query": "AWS templates"
},
"output": {
"success": true,
"query": "AWS templates",
"queryResult": "... very large data truncated"
}
},
{
"type": "text",
"text": "Prompt is too long",
"state": "done"
}
]
}
| const UI_TOOL_PREFIX = 'ui-'; | ||
|
|
||
| export function hasCompletedUIToolCalls(messages: UIMessage[]): boolean { | ||
| const message = messages[messages.length - 1]; |
There was a problem hiding this comment.
[nitpick]
| const message = messages[messages.length - 1]; | |
| const message = messages.at(-1); |
|
Fixes OPS-3497