Skip to content

Conversation

@majiayu000
Copy link

Summary

Test plan

  • Added 10 unit tests for _parse_thoughts_and_action method
  • All existing tests pass
  • poe fmt - formatting passes
  • poe lint - linting passes
  • poe pyright - type checking passes

Fixes microsoft#453

- Add validation for empty model responses with helpful error message
- Add check for missing <tool_call> tag with debugging info
- Handle case where <tool_call> is not followed by newline
- Add check for empty tool call content
- Improve error messages to help users diagnose model compatibility issues
- Add comprehensive unit tests for _parse_thoughts_and_action method

Signed-off-by: majiayu000 <1835304752@qq.com>
levidehaan pushed a commit to levidehaan/magentic-ui that referenced this pull request Dec 31, 2025
- Add Anthropic Claude as a model provider option in the UI
  - New AnthropicModelConfigForm component for configuring Claude models
  - Support for claude-sonnet-4, claude-3-5-sonnet, claude-3-5-haiku, claude-3-opus
  - Backend validation for Anthropic provider aliases
  - Updated autogen-ext dependency to include anthropic extra

- Improve Fara web surfer error handling (PR microsoft#484)
  - Add validation for empty/whitespace-only model responses
  - Check for required <tool_call> tag before parsing
  - Handle tool calls with or without newline delimiters
  - Provide descriptive error messages with response previews
  - Fall back gracefully from JSON to ast.literal_eval parsing
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Fara integration crashes when using LM Studio instead of vLLM (empty response triggers IndexError in _parse_thoughts_and_action)

1 participant