Class AgentSPEX

src/harness/agent.py:50–827

YAML-native Agent that executes YAML workflow definitions. This class is the main entry point for running YAML-defined workflows. It coordinates between the MCP client for tool execution, the LLM client for completions, and the Interpreter for step execution. Attributes: mc

class AgentSPEX:
    """YAML-native Agent that executes YAML workflow definitions.

    This class is the main entry point for running YAML-defined workflows.
    It coordinates between the MCP client for tool execution, the LLM client
    for completions, and the Interpreter for step execution.

    Attributes:
        mcp_url: URL of the MCP server.
        llm_client: LLMClient for LLM completions (supports multiple providers via LiteLLM).
        mcp_client: MCP client for tool invocation.
        tools: List of available tool definitions.
        submodule_tools: List of submodule tool definitions.
        submodule_registry: Registry of submodule specifications.
        agent_interpreter: Interpreter for executing workflow steps.
        usage_tracker: Dict tracking token usage across the workflow.

    Example:
        agent = AgentSPEX(
            mcp_url="http://localhost:8080",
            tool_config_fp="tools.json",
        )
        result = agent.run(args)

        # With custom LLM configuration (e.g., local vLLM server)
        llm_config = LLMConfig(api_base="http://localhost:8000")
        agent = AgentSPEX(
            mcp_url="http://localhost:8080",
            llm_config=llm_config,
        )
    """

    # Step execution methods delegated to agent_interpreter
    _DELEGATED_METHODS: FrozenSet[str] = frozenset(
        {
            "execute_basic_step",
            "execute_task_step",
            "execute_for_each_step",
            "execute_conditional_step",
            "execute_parallel_step",
            "execute_while_step",
            "execute_switch_step",
            "execute_increment_step",
            "execute_set_variable_step",
            "execute_return_step",
            "execute_input_step",
            "execute_call_step",
            "execute_gather_step",
            "execute_workflow_step",
        }
    )

    def __init__(
        self,
        mcp_url: Optional[str] = None,
        tool_config_fp: Optional[str] = None,
        *,
        session_id: Optional[str] = None,