Overview

Enhance your AI agents with custom skills.

What Are Skills?

Skills are instruction files that give AI agents specialized knowledge about your project, tools, or workflows. They are read by agents as part of their context, helping them understand conventions, patterns, and best practices specific to your codebase.


How Skills Work

Skills are simple markdown files placed in specific directories that AI agents scan. Each skill contains instructions, examples, and guidelines that the agent follows when executing tasks.

For example, a "backend expert" skill might include:

  • Your API conventions (REST patterns, error handling)

  • Database query patterns

  • Authentication middleware usage

  • Testing patterns specific to your backend


Setting Up Skills

Skills are configured per AI provider. Each provider looks for skills in its own directory:

Claude reads skills from the .claude/skills/ directory.

your-project/
└── .claude/
    └── skills/
        └── my-skill/
            └── SKILL.md

For setup instructions, see the Claude documentation on project-level instructionsarrow-up-right.


Writing a Skill

A skill is a markdown file (typically SKILL.md) that contains instructions for the agent. Keep these guidelines in mind:

  • Be specific — Include concrete examples from your codebase

  • Be concise — Agents have context limits; focus on what matters most

  • Be actionable — Write instructions the agent can follow, not general advice

  • Use examples — Show code patterns the agent should follow

chevron-rightExample skill filehashtag
circle-info

Skills are entirely optional. Agents work without them using the context from .locus/LOCUS.md, workspace documents, and the codebase index. Skills are an additional layer of specialization.


Multiple Agent Directories

If you use multiple AI providers, you can maintain skills in multiple directories:

Each provider will only read skills from its own directory.

Last updated