freemcplab / Directory / Playground / Harness (Meta-skill)

● Community revfactory ⚡ Instant

Harness (Meta-skill)

Name: Harness (Meta-skill) (Claude Skill)
Author: revfactory

A meta-skill: tell it your domain, it designs the agent team for you (roles, prompts, hand-offs) and saves the specs for reuse.

Harness treats agent design as a first-class workflow: you describe your domain (incident response, content review, code triage); it asks the right questions, then emits a documented team — roles, system prompts, when each takes over, and shared memory boundaries.

▶ Watch demo 📦 Install 💡 Use Cases

Why use it

Key features

Interview-driven team design
Outputs role specs as Markdown
Defines hand-off rules between agents
Specifies shared vs private memory
Versioned team definitions
Reuse across projects

Live Demo

What it looks like in practice

ready

Install

Pick your client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "harness-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/revfactory/harness",
        "~/.claude/skills/harness"
      ],
      "_inferred": true
    }
  }
}

Open Claude Desktop → Settings → Developer → Edit Config. Restart after saving.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "harness-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/revfactory/harness",
        "~/.claude/skills/harness"
      ],
      "_inferred": true
    }
  }
}

Cursor uses the same mcpServers schema as Claude Desktop. Project config wins over global.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "harness-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/revfactory/harness",
        "~/.claude/skills/harness"
      ],
      "_inferred": true
    }
  }
}

Click the MCP Servers icon in the Cline sidebar, then "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "harness-skill": {
      "command": "git",
      "args": [
        "clone",
        "https://github.com/revfactory/harness",
        "~/.claude/skills/harness"
      ],
      "_inferred": true
    }
  }
}

Same shape as Claude Desktop. Restart Windsurf to pick up changes.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "harness-skill",
      "command": "git",
      "args": [
        "clone",
        "https://github.com/revfactory/harness",
        "~/.claude/skills/harness"
      ]
    }
  ]
}

Continue uses an array of server objects rather than a map.

~/.config/zed/settings.json

{
  "context_servers": {
    "harness-skill": {
      "command": {
        "path": "git",
        "args": [
          "clone",
          "https://github.com/revfactory/harness",
          "~/.claude/skills/harness"
        ]
      }
    }
  }
}

Add to context_servers. Zed hot-reloads on save.

claude mcp add harness-skill -- git clone https://github.com/revfactory/harness ~/.claude/skills/harness

One-liner. Verify with claude mcp list. Remove with claude mcp remove.

Use Cases

Real-world ways to use Harness (Meta-skill)

Design an incident-response agent team

👤 Claude Code users ⏱ ~15 min intermediate

When to use: You want multiple agents (triage, comms, scribe) cooperating on real incidents instead of one big system prompt.

Flow

Brief

Tell harness: "incident response, B2B SaaS, 3 specialists please."✓ Copied

→ Harness asks scope questions
Answer

Define escalation, channels, scribe duties.✓ Copied

→ Spec drafts emerge
Save

Save specs into /agents/incident/.✓ Copied

→ Reusable team def written

Outcome: Documented multi-agent team you can rerun next incident, not a one-off prompt.

Combinations

Pair with other MCPs for X10 leverage

harness-skill + filesystem

Save designed teams alongside your project

Combine harness-skill with filesystem: Save designed teams alongside your project✓ Copied

Tools

What this MCP exposes

Tool	Inputs	When to call	Cost
design	(see docs)	Run the design interview	1 call
review	(see docs)	Critique an existing team for missing roles or unclear hand-offs	1 call

Cost & Limits

What this costs to run

API quota: N/A
Tokens per call: Interview is moderate
Monetary: Free skill
Tip: Reuse outputs as-is for next domain; don't redesign from scratch

Security

Permissions, secrets, blast radius

Minimum scopes: fs-read fs-write

Credential storage: None

Data egress: None

Troubleshooting

Common errors and fixes

Output too generic

Provide concrete examples (real past incidents) during the interview

Alternatives