TL;DR
Harness is an L3 meta-factory layer framework in the Claude Code environment that automatically generates specialized agent teams and skills from just a domain description.
Key features
6 team architecture patterns: Automatically selects the optimal pattern for the domain among Pipeline, Fan-out/Fan-in, Expert Pool, Producer-Reviewer, Supervisor, and Hierarchical Delegation
Automatic skill generation: Creates .claude/skills/ files for each agent based on domain analysis results
Automated agent definition: Generates role-specific agent definition files in the .claude/agents/ directory
Multilingual support: Can be commanded in English, Korean, and Japanese
When to use it
When multiple specialized agents need to collaborate on complex software projects
When you need to repeatedly configure agent teams in the Claude Code environment
When you want to automate optimal agent configuration based solely on domain knowledge
When you need to distribute large-scale tasks that are beyond the capabilities of a single-agent approach