Topherhindman
fb7178eaaf
* Add Code of Conduct for clarity and updates Updated email contact and revised enforcement guidelines. * Add contributing guide for skill contributors Covers how to improve existing skills, add new ones, and file issues. Documents the freeze-forever principle, MCP-for-facts pattern, and file naming conventions so contributors understand the project's content philosophy before submitting changes. * Add issue templates for bug reports and skill requests Provides structured templates so contributors can report skill content that produces incorrect agent behavior, or request new skills. Includes contact links to LiveKit docs and the developer community for general questions. * Add pull request template with skill review checklist Guides contributors through the key quality checks: freeze-forever compliance, MCP delegation for API facts, size limits, trigger phrase quality, and agent-tested verification. * Add security policy for vulnerability reporting Documents how to report security issues via GitHub's private vulnerability reporting. Defines scope specific to this project — skill content that could cause agents to generate insecure code.
2.9 KiB
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.
- Fork and clone the repository
- Create a branch from
main - Make your changes
- 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.mdis 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:
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.