RevFactory's Harness is a team-architecture factory for Claude Code. Describe your domain, and Harness generates a complete agent team: 4–5 specialist agents, an orchestrator skill, and domain-specific skills — all picked from six pre-defined patterns (Pipeline, Fan-out/Fan-in, Expert Pool, Producer-Reviewer, Supervisor, Hierarchical Delegation). The project reports +60% average quality improvement in author-measured A/B testing (49.5 → 79.3). Its companion project harness-100 ships 100 production-ready agent team harnesses across 10 domains (1,808 markdown files). Gaining +318 stars/day as of May 2026, reflecting strong demand for automated agent orchestration design.
RevFactory’s Harness operates one level above typical agent orchestration tools. Rather than orchestrating agents directly, it designs the orchestration — generating complete agent team architectures from a natural-language domain description. The output is a set of CLAUDE.md-compatible skill files that define specialist agents, their roles, and how they coordinate.
Six architecture patterns: Pipeline (sequential handoff), Fan-out/Fan-in (parallel with aggregation), Expert Pool (dynamic routing), Producer-Reviewer (generation + validation), Supervisor (delegator + workers), and Hierarchical Delegation (multi-level management).
Generated agent teams: Each generated team includes 4–5 specialist agents with defined roles, an orchestrator skill that coordinates them, and domain-specific skills tailored to the target use case.
Quality benchmarks: Author-measured A/B testing shows +60% average quality improvement (49.5 → 79.3 score) with 15/15 win-rate against non-harnessed agent workflows.
harness-100 companion: 100 production-ready agent team harnesses across 10 domains — content creation, software development, data/AI, business strategy, education, legal, health, and more. Available in English and Korean (200 packages total, 1,808 markdown files).
Teams that want to systematize how they structure multi-agent workflows. Product managers who need to spin up domain-specific agent teams quickly. Developers experimenting with different orchestration patterns for the same problem — Harness lets you try Pipeline vs. Expert Pool vs. Producer-Reviewer without manually rewriting the agent architecture each time.
Harness generates Claude Code skill files. The output is Claude Code-specific and doesn’t directly port to Codex or Gemini CLI without translation. The quality benchmarks are author-measured (n=15) with third-party replications pending — treat them as directional rather than definitive.
We analyzed Harness as part of the meta-skill layer of the agent orchestration stack in Harness Wars: Who Owns Your Coding Agent?.
AI agents that work well with Harness (RevFactory).