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

That installs all seven skills for the coding agents you pick, with the ones it detects preselected, and adds the LiveKit Docs MCP server to each one that supports it. lk skills update pulls the latest; lk skills --help has the rest. lk agent init offers to do this for new agent projects.

Claude Code

Install the LiveKit plugin. It bundles all seven skills and the LiveKit Docs MCP server.

Pick one: the plugin or lk skills install, not both. Claude Code doesn't deduplicate plugin skills against project skills, so with both installed every description loads twice and the copies compete with each other for routing.

/plugin marketplace add livekit/agent-skills
/plugin install livekit@livekit

Claude Code doesn't auto-update third-party marketplaces by default. To get skill updates as they land, turn on auto-update for the livekit marketplace in /plugin under Marketplaces, or run /plugin marketplace update livekit now and then.

Plugin skills are namespaced, so if you invoke one by hand it's /livekit:building-livekit-agents rather than /building-livekit-agents. You rarely need to: Claude picks them up from what you say.

npx skills

npx skills add livekit/agent-skills

That installs all seven for whichever coding agents you use.

You can also copy the folders you want from skills/ into your agent's skills directory.

You'll also want

• The LiveKit CLI (lk). Four skills drive it: lk docs, lk agent debugger, lk agent simulate, and the lk agent deploy commands. Install or update it from the CLI docs. The skills check for the commands they need and will tell you to update if one is missing.
• A LiveKit Cloud project, with lk authenticated to it, for simulations and deployment.
• The LiveKit Docs MCP server (optional, and already set up by lk skills install and the Claude Code plugin). It's the same source lk docs uses, with less friction. Setup for each coding agent is at docs.livekit.io/intro/mcp-server.

How the skills are built

The skills are conceptual. They teach how to approach a job (what to check, what goes wrong, which tool fits) and send the agent to --help and the live docs for flags, schemas, and API names. That keeps them correct as LiveKit changes. "Start the agent, send turns, read the tool calls, restart after an edit" stays the same even when the flag list changes.

So expect your agent to look things up while using them. That's intended.

The design rules, authoring conventions, and the eval harness that checks each skill triggers on the right requests are in AGENTS.md.

Upgrading from the old skills

livekit-agents and livekit-simulations were split into the seven skills above and removed. If your agent still has either installed, delete it and reinstall with one of the methods under Install.

A stale copy of livekit-simulations in particular will steer an agent wrong — it described a CLI surface that no longer exists.

Contributing

See CONTRIBUTING.md for the process and AGENTS.md for how skills here are written and evaluated. The most useful bug reports are about skill content: an agent followed a skill and did the wrong thing.

License

MIT