livekit / agent-skills

livekit-agents is the substantial one: architectural guidance for workflow design, task handoffs, and what the README calls mandatory testing practices when building a low-latency voice agent on the Agents SDK. livekit-simulations is narrower — it generates persona/goal/challenge test scenarios from an agent's own code and the behavior a reader wants stress-tested, then runs them, locally or from CI.

This is for a team already building on LiveKit Cloud and the Agents SDK who wants an agent that reasons about architecture correctly without also needing to be re-taught the current API on every release. If you are not building voice agents on LiveKit specifically, there is nothing here for you — two skills is the entire package, and both assume you already know what LiveKit is.

READ THE FULL ANALYSIS＋

The skills deliberately hold no facts. The README calls this the "freeze forever" principle: a skill should stay correct even if never updated, so both files teach behavior and approach and send every concrete API detail — signatures, config options — to a companion LiveKit Docs MCP server instead. That MCP server is a separate install; a reader who skips it gets the architectural opinions with no current API to apply them to.

Checked 19 September 2026 from the repository README and its two currently live skills.