Skip to content
Search

Harness

lastmod 2026-03-30created 2026-03-302 edits

A Claude Code agent team & skill architect plugin

About

  • A meta-skill that, with a single phrase like "Set up a harness," designs a team of specialized agents suited to the domain and even auto-generates the skills the agents will use
  • Supports 6 architecture patterns and includes inter-agent orchestration and error-handling protocols
  • Architecture patterns
    • Pipeline: sequentially dependent tasks
    • Fan-out/fan-in: parallel independent tasks
    • Expert pool: selective invocation by situation
    • Generate-verify: quality review after generation
    • Supervisor: a central agent distributes work dynamically
    • Hierarchical delegation: recursive delegation from upper to lower levels
  • 6-step workflow: domain analysis → team architecture design (agent team vs subagents) → agent definition generation → skill generation → integration and orchestration → verification and testing
  • Two execution modes:
    • Agent team (default): TeamCreate + SendMessage + TaskCreate approach, recommended when 2 or more agents and collaboration are needed
    • Subagents: invokes the Agent tool directly, suited to one-off tasks that need no communication
  • Running the harness auto-generates agent definition files (e.g. analyst.md, builder.md, qa.md) in .claude/agents/ and skill files in .claude/skills/
  • Example team configurations it can generate
    • Deep research — Set up a research harness. I need an agent team that can investigate any topic from multiple angles — web search, academic sources, community reactions — a team that cross-verifies and writes a comprehensive report.
    • Website building — Set up a full-stack website development harness. A team that coordinates design, frontend (React/Next.js), backend (API), and QA testing as a pipeline from wireframe to deployment.
    • Webtoon production — Set up a webtoon episode production harness. It needs story writing, character design prompt, panel layout planning, and dialogue editing agents, and they should review each other's work for style consistency.
    • YouTube content planning — Set up a YouTube content production harness. A team where a supervisor agent coordinates trend research, script writing, title/tag SEO optimization, and thumbnail concept planning.
    • Code review — Set up a comprehensive code review harness. A team of agents that audit architecture, security vulnerabilities, performance bottlenecks, and code style in parallel and integrate the results into a single report.
    • Technical writing — Set up a harness that auto-generates API documentation from this codebase. A team that handles endpoint analysis, description writing, usage example generation, and completeness review as a pipeline.
    • Data pipeline design — Set up a data pipeline design harness. An agent team that hierarchically delegates schema design, ETL logic, data validation rules, and monitoring setup.
    • Marketing campaign — Set up a marketing campaign production harness. A team that carries out target market research, ad copywriting, visual concept design, and A/B test planning with iterative quality reviews.
  • revfactory/harness-100 — released 100 production-ready agent team harnesses across 10 domains (200 packages in Korean and English)
    • Each harness includes 4-5 specialized agents, an orchestrator skill, and domain-specific skills
    • Consists of 1,808 Markdown files covering content creation, software development, data/AI, business strategy, education, law, healthcare, and more
    • All generated with the Harness plugin
  • Requires enabling Claude Code's agent teams feature: CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1

See also

Favorite site

author Zer0 sitemap