Research Team Multi-Agent Blueprint

A framework-agnostic blueprint (blueprint.yaml, plus a README with a full worked example) defining a 5-agent research pipeline: a Planner that scopes the question into sub-questions and acceptance criteria, a Researcher that gathers sourced findings, a Fact-Checker that re-verifies every cited claim and rates it confirmed/probable/disputed/unverifiable, a Synthesizer that writes the answer with inline attribution and honest confidence levels, and a Critic that checks the draft against the original acceptance criteria and can send it back for another pass (bounded to 3 loop iterations by default, with an escalation path to a human reviewer). The YAML defines each agent's role, responsibilities, inputs, outputs, and handoff conditions; the README explains the design in plain English and maps it onto LangGraph, CrewAI, and a plain custom-orchestrator implementation, including sample routing code. The worked example walks one fictional research question through all five agents, including a full loop iteration. This is a design document, not a running system — no tools, models, or APIs are wired in; 'tools' entries are conceptual and meant to be bound to your own search/retrieval provider.