The Agent Skills Directory

[SAFE]: The skill functions as a technical design manual for multi-agent architectures, providing patterns for supervisor/orchestrator, peer-to-peer/swarm, and hierarchical systems. It focuses on context isolation and token efficiency.
[SAFE]: The Python utility script (scripts/coordination.py) implements benign communication buses and coordination logic. It uses standard Python libraries (dataclasses, enum, uuid, time) to manage in-memory message passing and voting mechanisms without external network dependencies or sensitive file access.
[SAFE]: The architectural guidance includes security-positive design principles, such as explicit output validation between agents, monitoring for supervisor bottlenecks, and implementing circuit breakers to prevent error propagation cascades.
[SAFE]: The skill provides implementation examples for well-known and trusted agent frameworks, including LangGraph, AutoGen, and CrewAI. All external links point to official documentation or public repositories of established organizations.
[INDIRECT_PROMPT_INJECTION]: The skill defines architectures that ingest external data via message payloads (AgentMessage) and task specifications. However, it explicitly mitigates these risks by instructing the user to craft domain-focused prompts with strict deliverables and to implement validation checkpoints between agents.

multi-agent-patterns