harness

Harness is a plugin for Claude Code that acts as a team-architecture factory. Given a plain-language description of a domain or project, it generates agent definitions under `.claude/agents/` and skill files under `.claude/skills/`, so a project gains a coordinated set of specialized agents rather than a single general-purpose assistant. The plugin is triggered by prompts such as "build a harness for this project," "design an agent team for this domain," or "set up a harness." It then runs a six-phase workflow: domain analysis, team-architecture design, agent definition generation, skill generation, integration and orchestration, and finally validation and testing. Six architectural patterns are offered for shaping the team: Pipeline for sequential dependent tasks, Fan-out/Fan-in for parallel independent tasks, Expert Pool for context-dependent selective invocation, Producer-Reviewer for generation followed by quality review, Supervisor for central dynamic task distribution, and Hierarchical Delegation for top-down recursive delegation. Two execution modes are available: Agent Teams (the default, using TeamCreate, SendMessage and TaskCreate) for two or more agents that need to collaborate, and Subagents for one-off tasks without inter-agent communication. Generated skills use a Progressive Disclosure approach intended to keep context usage efficient, and the plugin includes orchestration support for inter-agent data passing, error handling and coordination protocols. Validation covers trigger verification, dry-run testing, and comparison tests with and without the generated skills. Installation is available through the Claude Code plugin marketplace (`/plugin marketplace add revfactory/harness` followed by `/plugin install harness@harness-marketplace`) or by copying the skills directory into `~/.claude/skills/harness`. Agent Teams must be enabled via the `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` environment variable. The README lists example prompts for deep research, full-stack website development, webtoon production, YouTube content planning, code review and refactoring, technical documentation, data pipeline design, and marketing campaigns. It also positions the project relative to neighboring tools such as Archon, meta-harness, ECC, wshobson/agents and LangGraph, and cites an author-run A/B study on a sister repository reporting improved output quality and reduced variance, with the caveat that third-party replications are pending. The project is licensed under Apache 2.0.