Separate skill for contributing to existing LiveKit agent codebases #6
Description
What should the skill teach agents to do?
Add a separate skill (e.g. livekit-agents-contributing or a variant of the existing skill) tailored for developers working within an established LiveKit agent codebase. It would retain the timeless principles from the current skill (verify APIs via MCP, latency-first thinking, mandatory testing) while dropping the setup/bootstrapping content.
Why is this needed?
The current livekit-agents skill is focused on building voice AI agents from scratch — cloud setup, project connection, workflow architecture design (handoffs/tasks), and initial agent creation. This is great for greenfield projects.
However, teams that already have a LiveKit agent codebase in production need different guidance. The challenges shift from "how to build an agent" to "how to add features, debug issues, and maintain an existing agent."
Scope
- Extending an existing agent — adding modules, providers, tools, and integrations without breaking what's already there
- Debugging production issues — common failure patterns (event loop errors in forked processes, provider failures, latency regressions)
- Performance optimization — prewarm strategies, async patterns, context size management
- Testing within an established test infrastructure — working with existing fixtures and factories rather than creating a
tests/directory from scratch - Safe async patterns — especially relevant for LiveKit's fork-based worker model where
asyncio.run()in the wrong place breaks everything