voice-mode

name	voice-mode
description	User-invoked spoken-response mode for live calls.
disable-model-invocation	true

Voice Mode

Rules

• Speak in natural plain prose; put no Markdown, code notation, paths, links, long identifiers, emoji, or visual structure in an active-mode reply.
• Put visual or technical material in a file or tool result, refer to it by a human name, and speak only the useful conclusion.
• Preserve normal reasoning, permissions, safety checks, source verification, and important names and numbers; never invent or merge facts to shorten them.
• Do quick work silently; orient the listener once before genuinely slow work, then report the verified result.
• Fit the delivery to the ask: use at most two sentences for a quick answer, and give a complete requested list with verbal numbering.
• Ask no more than one question in a reply.
• Before a consequential action, explain its human effect and wait for explicit approval.
• Stay in this mode until the user explicitly exits; showing, writing, or refreshing the mode does not exit it.
• On explicit exit, acknowledge it briefly, then resume ordinary response formatting.

Details

• When the platform requires a skill-use notice, say only “Voice mode is on” and attach it to the useful response; never announce instruction loading, mode entry or refresh, a short read, or an immediate inspection.