dragon.ai.agent.config.dispatch.DispatchHeader

class DispatchHeader[source] 

Typed header sent by the batch dispatcher to an agent’s input queue.

Parameters

task:: Human-readable task description for the agent.
serialized_ddict:: Serialized handle of the shared DDict for the current run.
completion_event:: A Event the agent sets when it finishes. The dispatcher blocks on this event instead of polling DDict. None when the agent is invoked outside of Dragon Batch (e.g. direct unit tests).
dispatch_id:: Universally unique identifier for this specific invocation, generated automatically via uuid.uuid4() . Scopes all per-invocation DDict keys (RESULT_KEY, STATUS_KEY, LLM_EVENT_KEY, TOOL_EVENT_KEY, HITL_EVENT_KEY) so that concurrent dispatches of the same (task_id, agent_id) pair — e.g. retries or parallel pipeline runs — never collide in the shared DDict. The dispatcher reads this value back after construction to format the same keys when checking status.
upstream_agent_ids:: List of direct-dependency agent_id values whose results this agent should read from DDict. Populated from PipelineNode.depends_on by the batch dispatcher. Empty list (default) = fall back to reading the full global state (backward-compatible).
tracing:: Pipeline-level tracing flag propagated from OrchestratorConfig. When True the agent creates a DictTracingProcessor and writes per-event DDict keys in real time.

__init__(task: str , serialized_ddict: str , completion_event: Any = None, dispatch_id: str = <factory>, upstream_agent_ids: list [str ] = <factory>, tracing: bool = False) → None 

Methods

__init__(task, serialized_ddict, ...)

Attributes

__init__(task: str , serialized_ddict: str , completion_event: Any = None, dispatch_id: str = <factory>, upstream_agent_ids: list [str ] = <factory>, tracing: bool = False) → None 