LiveKit Agent Skills

Skills that teach AI coding agents (Claude Code, Cursor, Codex, Gemini CLI, and anything else that reads the Agent Skills format) how to build, test, and ship voice agents with LiveKit and the LiveKit Agents SDK for Python and Node.js. Includes a Claude Code plugin that bundles the skills with the LiveKit Docs MCP server.

Without these skills, a coding agent working on a LiveKit project guesses at APIs from stale training data, ships prompt changes it never tried, and doesn't know the LiveKit tooling exists. With them, it looks facts up before writing code, live-tests changes in a conversation, writes turn-level tests, and runs simulations before a release.

What you get

Seven skills, one per job in the development loop. Each is small and hands off to the others by name, so a task typically loads two or three of them instead of one large document.

Stage	Skill	What it does
Know	reading-livekit-docs	Looks up current LiveKit facts (APIs, CLI flags, changelogs, pricing) through the Docs MCP server or lk docs instead of answering from memory. The other skills load it first.
Build	building-livekit-agents	Architecture for low-latency voice agents: keeping context small, splitting monoliths into handoffs and tasks, designing for users who listen rather than read.
Try	debugging-livekit-agents	Drives a multi-turn conversation with the agent running locally, in text or audio mode, via lk agent debugger, and reads the tool calls behind each reply. The default way to check a change.
Test	testing-livekit-agents	Turn-level tests in the project's own pytest or Vitest suite: assertions on messages, tool calls and handoffs, LLM judging of intent, tool mocking.
Simulate	writing-livekit-scenarios	Creates the scenarios a simulation runs (generate a baseline, refine it, cover the hard cases, organize into sets) and wires the agent to seed state from them and grade its final state.
Ship	running-livekit-simulations	Runs simulations in text or audio mode, against a local or deployed agent, in CI before a release, and triages the failures.
Operate	operating-livekit-agents	Deploys a version to LiveKit Cloud and rolls it back, and keeps a live agent healthy: the worker process model and prewarming, safe async in workers, provider timeouts, graceful shutdown, SDK upgrades, observability, and changing a codebase that's already in production.

You don't invoke skills by name. Your coding agent picks them up from what you say:

• "does livekit support gemini realtime?" · "what changed in agents 1.8?" → reading
• "add a tool that looks up an order" · "this agent feels sluggish, restructure it" → building
• "test my agent" · "try it as a confused first-time caller" · "why did it call that tool?" → debugging
• "write pytest tests for the greeting" · "pin that bug as a test" → testing
• "what should I test before we ship?" · "my simulations are flaky" → writing scenarios
• "run the scenarios and tell me what failed" · "test barge-in in audio mode" → running
• "deploy my agent" · "roll back to yesterday's version" · "first call after a restart is slow" → operating

Install

LiveKit CLI

lk skills install