| name | xiaoai |
| description | Use this skill when interacting through the XiaoAi channel or controlling XiaoAi-connected devices. |
XiaoAi
Use this skill when the active interaction happens through XiaoAi, including spoken replies and XiaoAi-executed commands.
Rule
If the assistant is delivering a user-facing reply through XiaoAi, call xiaoai_speak.
If the assistant needs XiaoAi to execute an instruction, call xiaoai_exec.
If the assistant wants to continue the conversation after replying, it must call xiaoai_listen to fetch a new XiaoAi message before continuing. Do not assume a follow-up message exists unless xiaoai_listen returns one.
How to speak
- Pass the exact final spoken reply as the
text argument.
- Keep the spoken text concise and natural for TTS.
- Make one
xiaoai_speak call per reply unless the task explicitly needs multiple utterances.
- Do not call
xiaoai_speak for hidden reasoning, tool chatter, or non-user-facing status messages.
Sending XiaoAi commands
Use xiaoai_exec when the task is to send a device command to XiaoAi instead of speaking a reply.
- Pass the XiaoAi instruction as the
command argument.
- Use natural commands that XiaoAi can execute directly, such as turning lights on or off.
- Do not wrap the command in explanatory text.
- If the user asks to operate a device, prefer
xiaoai_exec over xiaoai_speak.
Example:
xiaoai_exec command=关灯
Continuing the conversation
- After speaking, if you intend to keep talking with the user, call
xiaoai_listen to get the next XiaoAi message.
- Only continue the dialog when
xiaoai_listen returns a new user message.
- If
xiaoai_listen does not return a new message, end the turn instead of inventing a continuation.
Default pattern
- Finish any required reasoning or tool work.
- Prepare the final reply text for the user.
- Call
xiaoai_speak(text=<final reply>).
- If you want to continue the conversation, call
xiaoai_listen() and wait for a new message before proceeding.
- If a text response is also required, keep it aligned with what was spoken.
