Skip to main content
Open Skills
agent-harnessby Alireza RezvaniDevelopmentGitHub stars: 27.2k

Agent Harness plugin: what it installs and how to add it

Runs bounded agentic loops that turn goals into verified tasks and close only when checks pass or are explicitly waived.

From the Marketplace: claude-code-skills marketplace

Install

> /plugin marketplace add alirezarezvani/claude-skills> /plugin install agent-harness@claude-code-skills

Run both lines inside Claude Code: the first adds the marketplace, the second installs the plugin.

What is inside

  • 1skill
  • 1command
  • 1agent

About this plugin

Agent Harness turns a domain folder's skills, tools and checks into a controlled execution loop. It compiles a goal into a task plan, rejects vague goals with forcing questions, then drives execution, verification and retries with capped budgets. Verification runs through the controller, and exhausted retries escalate to a human instead of being treated as success.

Use it when an agent or subagent needs to take a goal through to a verified close. The state machine stores its status as JSON and refuses to close while any task remains unverified or unwaived. Its tools use only the Python standard library and emit JSON.

What's inside

  • 1 skill: agent-harness
  • 1 slash command: /cs-harness
  • 1 agent: harness-runner

More from claude-code-skills

Related plugins

Skills, not plugins

Prefer a single skill for development work? 471 skills in the same category, each installed on its own.

Development skills