Top AI Product

revfactory/harness is a meta-skill for Claude Code: instead of giving you an agent, it designs the agent team. Tell it “build a harness for this project,” describe the domain, and it picks a team architecture, defines specialised agents, and generates the skills they use.

## Six team patterns, picked for you

Harness ships with six battle-tested multi-agent patterns and chooses among them based on the problem shape: Pipeline, Fan-out/Fan-in, Expert Pool, Producer-Reviewer, Supervisor, and Hierarchical Delegation. Each comes with explicit message protocols and review gates. Skills auto-generate with Progressive Disclosure — context stays lean, details load only when an agent actually needs them. The author positions harness at the “L3 Meta-Factory” layer: it doesn’t run the agents, it builds the harness those agents will run inside.

## Why it matters

Multi-agent systems sound elegant until you wire one up. Deciding which agents exist, who talks to whom, who reviews whose work, and what the message format is — that’s the hidden hard part most teams reinvent badly. A library of named, tested patterns plus auto-generated agent and skill definitions turns that reinvention into a one-shot configuration. The paired research, “Harness: Structured Pre-Configuration for Enhancing LLM Code Agent Output Quality,” reports that the gains scale with task complexity: the harder the task, the more pre-configured team structure pays off. That tracks how multi-agent systems actually fail — on the seams, not on the individual agent’s reasoning.