feat: implement native tool calling in LiteAgent#4659
Closed
joaomdmoura wants to merge 5 commits intomainfrom
Closed
feat: implement native tool calling in LiteAgent#4659joaomdmoura wants to merge 5 commits intomainfrom
joaomdmoura wants to merge 5 commits intomainfrom
Conversation
- Added support for native function calling in the LiteAgent class, allowing it to utilize LLM's built-in capabilities for structured tool calls. - Introduced a new execution mode that determines whether to use native tools or fallback to the ReAct text pattern based on LLM capabilities. - Updated system prompts to accommodate the new native tools functionality. - Enhanced the agent's invocation loop to handle native tool calls effectively, improving overall performance and response accuracy.
![github-code-quality[bot]](https://avatars.githubusercontent.com/u/9919?s=60&v=4){kind=small}
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
…ll hooks - Simplified the conversion of tools to OpenAI schema by removing redundant mapping. - Updated the after-LLM call hooks to support list-type answers, ensuring native tool calls are returned unchanged. - Added tests to verify correct usage count for native tools and ensure proper handling of duplicate tool names. - Enhanced existing tests to confirm functionality with after-LLM hooks active, addressing previous issues with tool call processing.
Co-authored-by: Copilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com>
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 2 potential issues.
Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
| "result": result, | ||
| "tool_name": func_name, | ||
| "tool_args": args_dict, | ||
| }) |
There was a problem hiding this comment.
Shared mutable state accessed from thread pool workers
Medium Severity
_execute_native_tool_call appends to self.tools_results (a plain list) at the end of the method, but in the parallel path this method is submitted to a ThreadPoolExecutor. list.append is only incidentally atomic under CPython's GIL and isn't safe under free-threaded Python (PEP 703). The result ordering in tools_results is also non-deterministic across runs.
Additional Locations (1)
| "conversation_history_instruction": "You are a member of a crew collaborating to achieve a common goal. Your task is a specific action that contributes to this larger objective. For additional context, please review the conversation history between you and the user that led to the initiation of this crew. Use any relevant information or feedback from the conversation to inform your task execution and ensure your response aligns with both the immediate task and the crew's overall goals.", | ||
| "feedback_instructions": "User feedback: {feedback}\nInstructions: Use this feedback to enhance the next output iteration.\nNote: Do not respond or add commentary.", | ||
| "lite_agent_system_prompt_with_tools": "You are {role}. {backstory}\nYour personal goal is: {goal}\n\nYou ONLY have access to the following tools, and should NEVER make up tools that are not listed here:\n\n{tools}\n\nIMPORTANT: Use the following format in your response:\n\n```\nThought: you should always think about what to do\nAction: the action to take, only one name of [{tool_names}], just the name, exactly as it's written.\nAction Input: the input to the action, just a simple JSON object, enclosed in curly braces, using \" to wrap keys and values.\nObservation: the result of the action\n```\n\nOnce all necessary information is gathered, return the following format:\n\n```\nThought: I now know the final answer\nFinal Answer: the final answer to the original input question\n```", | ||
| "lite_agent_system_prompt_native_tools": "You are {role}. {backstory}\nYour personal goal is: {goal}", |
There was a problem hiding this comment.
Native tools prompt references ReAct "Final Answer" concept
Low Severity
The post_tool_reasoning prompt used after native tool calls tells the LLM to "provide the Final Answer," which is a ReAct-specific text pattern. In native tool calling mode, there's no parser looking for "Final Answer:" — the LLM may emit it literally as part of its text response, producing user-visible artifacts like "Final Answer: 42" instead of just "42."
Additional Locations (1)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Note
Medium Risk
Adds a new LiteAgent execution path that bypasses ReAct parsing and executes provider-native tool calls (including optional parallelism), which can change tool invocation order and message history shape. Also adjusts LLM hook handling to preserve tool-call lists, affecting hook/response behavior across executors.
Overview
LiteAgent can now run in a native function-calling mode when the underlying LLM reports
supports_function_calling()and tools are present, falling back to the existing ReAct text loop otherwise.In native mode, LiteAgent sends OpenAI-style tool schemas to the LLM, detects tool-call responses across multiple provider formats, executes tool calls (optionally in parallel when safe), appends
assistant.tool_calls+toolmessages, and continues with a post-tool reasoning prompt; it also supports early return forresult_as_answertools and respects tool usage limits/caching.System prompting is updated to use a new
lite_agent_system_prompt_native_toolsslice (no ReAct formatting instructions), and_setup_after_llm_call_hooksis updated to accept list responses and not stringify tool-call lists so native tool handling continues to work with hooks enabled. Extensive unit tests cover mode detection, prompt selection, tool execution/parallel batches, usage counters/limits, hook regression, and duplicate tool-name deduping.Written by Cursor Bugbot for commit d2a5256. This will update automatically on new commits. Configure here.