The voice protocol your agent can follow

The skill is one Markdown file, SKILL.md, with YAML frontmatter, in the same format as other agent skills. Point your agent at it and it will run conversations the way the console does.

What is in it

1. Listen and caption. Show interim words as they arrive; set them once they are final.
2. Wait for the end of the utterance. Do not begin speaking until at least 900 ms of silence follow the last final segment (configurable).
3. Mute while speaking, stop when asked. The microphone is muted while the agent talks so it never hears itself. Escape, Space or the stop button cut the reply at once and mark it interrupted.
4. Speak in the active persona. Voice provider, style prompt and browser fallback hint per persona.
5. Write memory after every turn. Append a JSONL entry and refresh a rolling MEMORY.md of the last 50 turns.
6. Persona registry. The JSON format and how to add a character voice, including cloning.
7. License check. GET /api/license?key= before the skill runs.

Download

The file is served by /api/skill and only to a valid license. Without one the endpoint answers 402 license_required.