agent.pythontoolwithout_approval

Flags supported Python tool declarations without a conservative approval marker.

Why It Matters

Agent-callable functions expose capability to the model or agent loop. Risky tools should require approval or narrower inputs before they can run.

Review

Bad:

@tool
def deploy(target: str) -> str:
    return target

Good:

@tool(require_human_approval=True)
def deploy(target: str) -> str:
    return target

Framework Note

Supported v0.2 scope includes LangChain, LangGraph, CrewAI, MCP decorators, OpenAI tool-calling dictionaries, Anthropic tool-use dictionaries, LlamaIndex FunctionTool, and Gemini function_declarations.