Harness builds domain-specific agent teams and skills

This is less a single tool than a system for manufacturing AI workflows, which is where agentic tooling is likely headed if teams want anything beyond prompt spaghetti.

• –The 6-pattern team model gives the project real structure instead of hand-wavy “multi-agent” branding
• –Auto-generating `.claude/agents/` and `.claude/skills/` pushes knowledge into repo state, which is the right place for reusable agent behavior
• –The validation loop matters more than the orchestration: dry runs and with/without-skill comparisons are what make these systems trustworthy
• –The strongest use case is domain specialization, where one generic agent is weaker than a coordinated team with explicit roles
• –It still depends on Claude Code’s ecosystem, so the long-term value is tied to how portable those patterns are outside this runtime