strands.vended_tools.stop.stop

Tool for gracefully ending the agent loop.

Provides :func:make_stop (a factory for customized stop tools) and :data:stop (the default instance). The tool calls agent.cancel(message=..., after_current_tools=True): any sibling tools in the same batch run to completion, then the loop exits with stop_reason == "cancelled" and the message as the final assistant message. The same message is returned as the tool result the model sees for its stop request.

DEFAULT_MAX_MESSAGE_LENGTH

Default cap on the stop message length. The cap exists so a runaway model can’t blow the conversation history in one shot; adjust via make_stop(max_message_length=...) when a longer summary is legitimate.

make_stop

def make_stop(
    *,
    name: str = "stop",
    description: str = DEFAULT_STOP_DESCRIPTION,
    max_message_length: int = DEFAULT_MAX_MESSAGE_LENGTH
) -> DecoratedFunctionTool

Defined in: src/strands/vended_tools/stop/stop.py:56

Create a stop tool that gracefully ends the agent loop.

The tool calls agent.cancel(message=..., after_current_tools=True) so the loop ends after the current tool batch without invoking the model again.

Arguments:

• name - Tool name. Defaults to "stop".
• description - Tool description shown to the model.
• max_message_length - Maximum accepted length for the model-supplied message argument, in characters. Must be a positive integer. Defaults to :data:DEFAULT_MAX_MESSAGE_LENGTH (4096).

Returns:

A decorated tool that signals the event loop to stop after the current tool batch completes.

Raises:

• ValueError - If max_message_length is not a positive integer.

stop

Default stop tool. Ends the agent loop when called by the model.