[bcherry]

What does this PR do?

Expands the debugging-livekit-agents skill documentation to cover audio mode, which speaks turns into the agent's microphone input for testing transcription, endpointing, and speech-to-speech models. Updates trigger queries and README to reflect that the debugger now supports both text and audio modes.

Changes

• SKILL.md:

  • Updated description to mention audio mode and new use cases ("it mishears my order number")
  • Added "Audio mode" section explaining when to use it, costs, how to compare transcripts, and limitations
  • Updated "When to use something else" table to distinguish transcription bugs (use audio mode) from interruptions (use simulations)
  • Clarified that text mode is the default but audio mode is available

• queries.json: Added two trigger eval queries for audio mode use cases (order number transcription, speech-to-speech models)

• README.md: Updated skill summary to note both text and audio modes

• building-livekit-agents/SKILL.md: Updated guidance to mention sending turns as text or speech

Checklist

• Content passes the "freeze forever" test: references --help and start --help instead of specific flags
• No time-relative phrasing
• Description is third person, includes user phrases ("it mishears", "speech-to-speech model"), names sibling skills
• If description changed: trigger eval queries added for audio mode scenarios
• No new eval misses expected (audio mode is a documented feature, not a new skill)

Eval results

n/a: trigger queries added for audio mode coverage; no description changes to existing queries

https://claude.ai/code/session_01GgJFuYzMp7KtqGhSQYD5Zr

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}