Skip to main content

Harness Skills Overview

Last updated on

Harness Skills are specialized prompt templates that teach AI coding assistants how to interact with the Harness platform. Each skill encapsulates the domain knowledge needed to accomplish a specific task, such as generating pipeline YAML, creating services, debugging executions, or analyzing costs. You invoke skills using natural language in your editor, and the AI assistant handles the underlying interactions with Harness.

What will you learn in this topic?​

By the end of this topic, you will be able to:

Before you begin​

Before you set up Harness Skills, ensure you have the following:

  • AI coding assistant: Claude Code, Cursor, GitHub Copilot, OpenAI Codex, or Windsurf.
  • Harness MCP Server: A configured MCP server for tool execution.
  • Harness API key: An API key to authenticate with the Harness platform.

How skills work​

Skills are Markdown files with structured instructions that AI editors load as context. Understanding this flow helps you troubleshoot skill behavior and extend skills with your own workflows.

The repository is designed as a workflow system, not a folder of prompts. Top level instructions (CLAUDE.md, AGENTS.md, .github/copilot-instructions.md) establish shared behavior, while individual skills specialize in creation, debugging, governance, and reporting tasks.

When you invoke a skill (for example, /create-pipeline), the AI reads the skill's instructions and uses the Harness MCP Server tools to execute actions against the Harness platform:

Natural language prompt
    → AI editor loads skill instructions
        → Skill orchestrates MCP tool calls (harness_list, harness_create, etc.)
            → Harness MCP Server
                → Harness Platform APIs

Skills do not embed API schemas directly. Instead, they use the harness_describe MCP tool to discover resource schemas at runtime, keeping skills lightweight and always up to date.

Set up skills​

Configure your AI coding assistant to load skill instructions and connect to the Harness MCP Server. Setup differs by editor.

Claude Code​

Clone the skills repository and start Claude Code from it:

git clone https://github.com/harness/harness-skills.git
cd harness-skills
claude

Configure the Harness MCP server in ~/.claude/settings.json:

{
  "mcpServers": {
    "harness-mcp-v2": {
      "command": "npx",
      "args": ["-y", "harness-mcp-v2"],
      "env": {
        "HARNESS_API_KEY": "<your-api-key>"
      }
    }
  }
}

Skills are auto-discovered from the CLAUDE.md file and the skills/ directory. Invoke a skill by name:

/create-pipeline
Create a CI pipeline for a Node.js app that builds, tests, and pushes a Docker image to ECR

Cursor​

Cursor auto-loads the project rules from .cursor/rules/harness.mdc.

  1. Open the harness-skills folder in Cursor.
  2. Configure the MCP server in ~/.cursor/mcp.json:
{
  "mcpServers": {
    "harness-mcp-v2": {
      "command": "npx",
      "args": ["-y", "harness-mcp-v2"],
      "env": {
        "HARNESS_API_KEY": "<your-api-key>"
      }
    }
  }
}
  1. Reference skills using @file:
@harness-skills/skills/create-pipeline/SKILL.md
Create a CI pipeline for my Go microservice

GitHub Copilot​

GitHub Copilot auto-loads instructions from .github/copilot-instructions.md.

  1. Open the harness-skills folder in VS Code.
  2. Configure the MCP server in .vscode/mcp.json:
{
  "servers": {
    "harness-mcp-v2": {
      "command": "npx",
      "args": ["-y", "harness-mcp-v2"],
      "env": {
        "HARNESS_API_KEY": "<your-api-key>"
      }
    }
  }
}
  1. Reference skills using #file:
#file:harness-skills/skills/create-pipeline/SKILL.md
Create a CI/CD pipeline for my Python app

For GitHub Copilot on GitHub.com, attach skill files as context in Copilot Chat, or add them as knowledge base references in your Copilot organization settings.

OpenAI Codex​

Codex auto-loads the AGENTS.md file as system instructions.

  1. Clone the repository into your working directory:
git clone https://github.com/harness/harness-skills.git
  1. Configure the MCP server in your Codex MCP configuration:
{
  "mcpServers": {
    "harness-mcp-v2": {
      "command": "npx",
      "args": ["-y", "harness-mcp-v2"],
      "env": {
        "HARNESS_API_KEY": "<your-api-key>"
      }
    }
  }
}
  1. Reference skill files as context when prompting:
Using the instructions in harness-skills/skills/debug-pipeline/SKILL.md,
diagnose why my deploy pipeline failed

Windsurf and other AI editors​

The skills are plain Markdown files with YAML frontmatter. They work with any AI coding tool that supports:

  • System instructions: Use CLAUDE.md as project level context.
  • MCP servers: Connect the Harness MCP Server for API access.
  • File context: Reference individual skills/*/SKILL.md files in prompts.

Next steps​