What it does
agentic-harness is a plugin for Claude Code. It analyzes an existing repository, designs a team of agents and the skills they use, generates them into the repository's .claude folder, and builds an orchestrator that becomes the entry point for every prompt. The plugin does not do domain work itself; it builds and maintains the agents and skills that do. Alongside the setup skill it ships a read-only review skill that inventories the harness, detects drift and proposes improvements, and a doctor skill that checks whether the plugins, skills, MCP servers and CLIs a harness expects are actually installed. Every change is approved before anything is written and recorded in a change history. When a project already uses a spec-driven system or an issue tracker, the generated orchestrator coordinates with it instead of replacing it. Advisor skills suggest a spec system or tracker when none is present and leave installation to that tool's own installer. A feedback skill can draft redacted feedback and, only with explicit consent, file it as a GitHub issue on the plugin's repository.
Capabilities
- Generates a repository-specific team of agents, skills and an orchestrator under .claude/
- Read-only review that detects drift and proposes harness improvements
- Coordinates with existing spec-driven systems and issue trackers
- Optional dual-tracker sync that keeps a SaaS tracker in step with a repo-native one
- Doctor check for installed plugins, skills, MCP servers and CLIs
- Every change approved first and recorded in a change history
When to use it
- You want Claude Code to work with agents and skills tailored to one repository
- You already use a spec system or issue tracker and want the agents to coordinate with it
- You want to audit an existing Claude Code setup for drift and unused pieces
