What problem does it solve? Building multi-agent AI systems fails most often at routing correctness, state integrity, and failure isolation. This Skill provides architectural patterns for agent registries, job queues, LLM fallback routing, tool authorization, and prompt injection defense so these problems are solved structurally rather than patched per feature. ## Core Features & Use Cases - Agent Registry and Contracts: Define each agent role with an explicit model, fallback model, tool allowlist, timeout, and versioned system prompt before writing any implementation code. - BullMQ Job Architecture: Structure agent work as stateless jobs with correlation IDs, retry policies, separate Redis connections, and per-role worker concurrency limits. - LLM Router with Fallback: Wrap every model call in a timeout, automatic fallback to a secondary model on rate limits, input sanitization against prompt injection, and token usage metrics. - Use Case: When building a system where an orchestrator agent decomposes a user request into sub-tasks executed by coder, researcher, and evaluator agents, use this Skill to design the pipeline state machine, tool dispatch authorization, and OpenTelemetry tracing before writing agent code. ## Quick Start Ask the AI to design the agent registry and BullMQ job architecture for a multi-agent pipeline with an orchestrator, coder, and evaluator role.
