Coordination Patterns ===================== ``RoundBasedCoordinationPattern`` and ``BlackboardPattern`` provide peer-only coordination without a central orchestrator agent. Core behavior ------------- - Round-based execution with deterministic peer order (sorted peer ids). - No manager/supervisor role required. - Explicit termination reasons: - ``converged`` - ``max_rounds_reached`` - ``explicit_stop`` - ``peer_failure`` Blackboard specialization ------------------------- ``BlackboardPattern`` keeps an explicit shared board payload: .. code-block:: python { "task": str, "round": int, "messages": list, "proposals": dict, "decisions": dict, "history": list, } The reducer merges per-peer contributions by channel, appends immutable message history entries, computes ``state_hash``, and can declare convergence when the hash stays unchanged for configured stability rounds. Background references --------------------- - `AutoGen `_ - `CAMEL `_ - `MetaGPT `_ - `ChatDev `_ These are examples of LLM multi-agent coordination; these patterns are framework-native and intentionally minimal. Example ------- See ``examples/patterns/coordination_patterns.py``.