collaboration

Collaboration protocol models for CrewMaster v2.0.0.

MODULE	DESCRIPTION
protocol	Collaboration protocol implementations for CrewMaster v2.0.0.
protocol_test	Tests for CollaborationProtocol implementations.

CLASS	DESCRIPTION
BlackboardConfig	Configuration for the blackboard collaboration protocol.
BlackboardProtocol	Blackboard protocol: all participants see all messages.
CollaborationProtocol	Base protocol for multi-agent collaboration strategies.
CollaborationResult	The result of executing a collaboration protocol.
ConsensusConfig	Configuration for the consensus collaboration protocol.
ConsensusItem	A single proposal or vote item in the consensus protocol.
ConsensusProtocol	Consensus protocol: participants propose and vote until consensus.
DebateConfig	Configuration for the debate collaboration protocol.
DebateProtocol	Debate protocol: two debaters argue, optional judge decides winner.
GrillMeConfig	Configuration for the grill_me collaboration protocol.
GrillMeProtocol	Grill-me protocol: critic and respondent alternate turns.
SupervisorConfig	Configuration for the supervisor collaboration protocol.
SupervisorProtocol	Supervisor protocol: supervisor assigns, workers execute, supervisor reviews.

__all__ module-attribute

__all__ = ['BlackboardConfig', 'BlackboardProtocol', 'CollaborationProtocol', 'CollaborationResult', 'ConsensusConfig', 'ConsensusItem', 'ConsensusProtocol', 'DebateConfig', 'DebateProtocol', 'GrillMeConfig', 'GrillMeProtocol', 'SupervisorConfig', 'SupervisorProtocol']

BlackboardConfig

Configuration for the blackboard collaboration protocol.

ATTRIBUTE	DESCRIPTION
participants	List of agents that contribute to the blackboard. TYPE: list[Any]
max_rounds	Maximum number of rounds of contributions. TYPE: int

participants class-attribute instance-attribute

participants: list[Any] = Field(description='List of AgentConfig instances')

max_rounds class-attribute instance-attribute

max_rounds: int = Field(default=3, ge=1)

model_config class-attribute instance-attribute

model_config = {'arbitrary_types_allowed': True}

BlackboardProtocol

BlackboardProtocol(config: BlackboardConfig)

Blackboard protocol: all participants see all messages.

Flow::

round 1: agent_1 -> agent_2 -> ... -> agent_N
round 2: agent_1 -> agent_2 -> ... -> agent_N
...

History is indexed by round and contributor. All participants see all messages from all prior contributions.

METHOD	DESCRIPTION
execute	Public entry point — delegates to _execute_impl.

ATTRIBUTE	DESCRIPTION
config

config instance-attribute

config = config

execute async

execute(operation: Any, runtime: Any, context_store: Any) -> CollaborationResult

Public entry point — delegates to _execute_impl.

CollaborationProtocol

Base protocol for multi-agent collaboration strategies.

Collaboration protocols orchestrate multiple agents through structured interaction patterns. Each protocol implementation manages turn state internally and produces a single typed output artifact.

Usage::

class GrillMeProtocol:
    async def execute(self, operation, runtime, context_store):
        # Run N rounds of critique -> revision
        ...
        return CollaborationResult(output=final_artifact)

METHOD	DESCRIPTION
execute	Execute the collaboration protocol.

execute async

execute(operation: Any, runtime: Any, context_store: Any) -> CollaborationResult

Execute the collaboration protocol.

PARAMETER	DESCRIPTION
operation	The Operation that triggered this collaboration. TYPE: Any
runtime	The RuntimeDriver to use for agent invocations. TYPE: Any
context_store	The ContextStore for resolving context projectors. TYPE: Any

RETURNS	DESCRIPTION
CollaborationResult	A CollaborationResult with the final typed artifact.

CollaborationResult

The result of executing a collaboration protocol.

ATTRIBUTE	DESCRIPTION
output	The final typed artifact produced by the collaboration. TYPE: Any
protocol_type	The name of the protocol that produced this result. TYPE: str
turn_count	Total number of turns taken during execution. TYPE: int
participant_outputs	Individual outputs from each participant indexed by agent name. TYPE: dict[str, list[Any]]

output class-attribute instance-attribute

output: Any = None

protocol_type class-attribute instance-attribute

protocol_type: str = ''

turn_count class-attribute instance-attribute

turn_count: int = 0

participant_outputs class-attribute instance-attribute

participant_outputs: dict[str, list[Any]] = Field(default_factory=dict)

model_config class-attribute instance-attribute

model_config = {'arbitrary_types_allowed': True}

ConsensusConfig

Configuration for the consensus collaboration protocol.

ATTRIBUTE	DESCRIPTION
participants	List of agents that vote/propose. TYPE: list[Any]
max_rounds	Maximum number of voting rounds before forced decision. TYPE: int
consensus_threshold	Fraction (0-1) of participants that must agree. TYPE: float

participants class-attribute instance-attribute

participants: list[Any] = Field(description='List of AgentConfig instances')

max_rounds class-attribute instance-attribute

max_rounds: int = Field(default=3, ge=1)

consensus_threshold class-attribute instance-attribute

consensus_threshold: float = Field(default=1.0, ge=0.0, le=1.0)

model_config class-attribute instance-attribute

model_config = {'arbitrary_types_allowed': True}

ConsensusItem

A single proposal or vote item in the consensus protocol.

ATTRIBUTE	DESCRIPTION
agent	TYPE: str
proposal	TYPE: str

agent class-attribute instance-attribute

agent: str = ''

proposal class-attribute instance-attribute

proposal: str = ''

ConsensusProtocol

ConsensusProtocol(config: ConsensusConfig)

Consensus protocol: participants propose and vote until consensus.

Flow::

round 1: each participant proposes -> vote on proposals -> check consensus
round 2: participants revise proposals -> vote -> check consensus
...

Consensus is reached when consensus_threshold fraction of participants agree on the same proposal. If max_rounds is reached without consensus, the majority position is returned.

METHOD	DESCRIPTION
execute	Public entry point — delegates to _execute_impl.

ATTRIBUTE	DESCRIPTION
config

config instance-attribute

config = config

execute async

execute(operation: Any, runtime: Any, context_store: Any) -> CollaborationResult

Public entry point — delegates to _execute_impl.

DebateConfig

Configuration for the debate collaboration protocol.

ATTRIBUTE	DESCRIPTION
debater_a	First debater agent. TYPE: Any
debater_b	Second debater agent. TYPE: Any
judge	Optional judge agent who determines the winner. TYPE: Any | None
max_turns	Maximum number of debate rounds. TYPE: int

debater_a class-attribute instance-attribute

debater_a: Any = Field(description='AgentConfig for debater A')

debater_b class-attribute instance-attribute

debater_b: Any = Field(description='AgentConfig for debater B')

judge class-attribute instance-attribute

judge: Any | None = Field(default=None, description='Optional judge AgentConfig')

max_turns class-attribute instance-attribute

max_turns: int = Field(default=3, ge=1)

model_config class-attribute instance-attribute

model_config = {'arbitrary_types_allowed': True}

DebateProtocol

DebateProtocol(config: DebateConfig)

Debate protocol: two debaters argue, optional judge decides winner.

Flow::

debater_a -> debater_b -> debater_a -> ... -> judge -> winner

If no judge is provided, the last debater's position is returned.

METHOD	DESCRIPTION
execute	Public entry point — delegates to _execute_impl.

ATTRIBUTE	DESCRIPTION
config

config instance-attribute

config = config

execute async

execute(operation: Any, runtime: Any, context_store: Any) -> CollaborationResult

Public entry point — delegates to _execute_impl.

GrillMeConfig

Configuration for the grill_me collaboration protocol.

ATTRIBUTE	DESCRIPTION
critic	The agent who critiques. TYPE: Any
respondent	The agent who produces and revises. TYPE: Any
max_turns	Maximum number of critique->revision cycles. TYPE: int
stalemate_threshold	Stop if respondent output doesn't change for this many consecutive turns. TYPE: int
finalize_step	Optional agent that does a final polish pass. TYPE: Any | None

critic class-attribute instance-attribute

critic: Any = Field(description='AgentConfig for the critic')

respondent class-attribute instance-attribute

respondent: Any = Field(description='AgentConfig for the respondent')

max_turns class-attribute instance-attribute

max_turns: int = Field(default=3, ge=1)

stalemate_threshold class-attribute instance-attribute

stalemate_threshold: int = Field(default=2, ge=1)

finalize_step class-attribute instance-attribute

finalize_step: Any | None = Field(default=None, description='Optional agent for final polish')

model_config class-attribute instance-attribute

model_config = {'arbitrary_types_allowed': True}

GrillMeProtocol

GrillMeProtocol(config: GrillMeConfig)

Grill-me protocol: critic and respondent alternate turns.

Flow::

respondent produces -> critic critiques -> respondent revises -> ...

Terminates when max_turns is reached, or when the respondent's output is unchanged for stalemate_threshold consecutive turns.

METHOD	DESCRIPTION
execute	Public entry point — delegates to _execute_impl.

ATTRIBUTE	DESCRIPTION
config

config instance-attribute

config = config

execute async

execute(operation: Any, runtime: Any, context_store: Any) -> CollaborationResult

Public entry point — delegates to _execute_impl.

SupervisorConfig

Configuration for the supervisor collaboration protocol.

ATTRIBUTE	DESCRIPTION
supervisor	The agent that assigns tasks and reviews results. TYPE: Any
workers	List of worker agents that execute assigned tasks. TYPE: list[Any]
max_iterations	Maximum number of assign->execute->review cycles. TYPE: int

supervisor class-attribute instance-attribute

supervisor: Any = Field(description='AgentConfig for the supervisor')

workers class-attribute instance-attribute

workers: list[Any] = Field(description='List of worker AgentConfig instances')

max_iterations class-attribute instance-attribute

max_iterations: int = Field(default=3, ge=1)

model_config class-attribute instance-attribute

model_config = {'arbitrary_types_allowed': True}

SupervisorProtocol

SupervisorProtocol(config: SupervisorConfig)

Supervisor protocol: supervisor assigns, workers execute, supervisor reviews.

Flow::

supervisor assigns task -> worker executes -> supervisor reviews -> iterate

The supervisor receives the worker's output and decides whether to accept it or request a revision. The final output is the consolidated result produced by the supervisor.

METHOD	DESCRIPTION
execute	Public entry point — delegates to _execute_impl.

ATTRIBUTE	DESCRIPTION
config

config instance-attribute

config = config

execute async

execute(operation: Any, runtime: Any, context_store: Any) -> CollaborationResult

Public entry point — delegates to _execute_impl.