Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
89 lines (57 loc) · 2.93 KB

CONTRIBUTING.md

File metadata and controls

89 lines (57 loc) · 2.93 KB
Raw

Contributing to LiveKit Agent Skills

Thanks for your interest in contributing! This project provides reusable skills that help AI coding agents build voice AI applications with LiveKit.

How to Contribute

Improving Existing Skills

The most valuable contributions improve skill content — making guidance clearer, fixing incorrect behavioral advice, or adding missing patterns that agents commonly need.

  1. Fork and clone the repository
  2. Create a branch from main
  3. Make your changes
  4. Open a pull request

Adding a New Skill

New skills should follow the structure in skills/livekit-agents/ as a reference:

skills/
└── your-skill-name/
    ├── SKILL.md          # The skill content
    └── references/       # Supporting documentation

Every skill must include YAML frontmatter:

---
name: skill-name-kebab-case
description: >-
  Trigger phrases and brief description that help agents
  recognize when to activate this skill.
license: MIT
metadata:
  author: your-name
  version: "0.1.0"
---

Filing Issues

  • Bug reports: Skill content that causes agents to produce incorrect code or behavior
  • Skill requests: Ideas for new skills that would help agents build with LiveKit
  • Questions: General questions about usage or design

Skill Content Principles

All contributions must follow the "freeze forever" principle — content should remain correct indefinitely without updates.

Encode behavior, not knowledge

Skills teach how to approach problems, not API specifics. API signatures, configuration options, and method names change — behavioral guidance does not.

Good: "Always test agent implementations by verifying audio pipeline connectivity" Bad: session = AgentSession(llm=openai.LLM(model="gpt-4o"))

Direct to MCP for facts

All factual information must come from the LiveKit Docs MCP server. Skills should instruct agents to look up current API details rather than hardcoding them.

Require testing

Every skill that guides agent implementation must include testing expectations. Agents should never produce untested code.

Stay under 500 lines

Skills are loaded into agent context windows. Keep them concise — under 500 lines — so they don't crowd out the user's actual project context.

File Naming Conventions

  • Skill directories use kebab-case
  • SKILL.md is the only uppercase filename in a skill directory
  • Supporting documents go in references/

Development Setup

These skills are designed to work with the LiveKit Docs MCP server. To test MCP integration locally, install the server following the instructions at:

https://docs.livekit.io/mcp

Code of Conduct

This project follows the Contributor Covenant Code of Conduct. By participating, you agree to uphold this code.

License

By contributing, you agree that your contributions will be licensed under the MIT License.