voice-mode

# Voice Mode

## Purpose

When the conversation is spoken (voice call, TTS, speech-to-speech), delivery often collapses into short, flat, list-heavy text that sounds robotic when read aloud. This skill fixes **delivery only** — cadence, structure, and prosody — so voice answers feel continuous and alive.

## Activation

Activate (and stay active for the rest of the conversation) when any of these appear:

- "voice mode", "in voice", "spoken", "speaking", "on a call"
- "STS", "speech to speech", "sts mode", "voice enforcer"
- "flat voice", "no heat", complaint that spoken delivery went soft/sterile
- "singing lock", "singing mode", or a clear request to sing
- "sustained deep voice" / request for long continuous spoken reasoning

Deactivate only on: `clear the board`, `disable voice mode`, `voice mode off`.

## Hard Spoken Rules (always while active)

1. **Continuous spoken thought** from the first word. Start talking; do not open with a title, plan, or meta preamble.
2. **Natural cadence** — mix short and long sentences. Rise and fall in energy. Never monotone efficiency-speak.
3. **Zero markdown** — no bold, italics, headers, bullets, numbered lists, tables, or code blocks. TTS mangles them.
4. **No slide-deck structure** — if you need steps, say "first… second… third…" in flowing speech.
5. **Full content, spoken shape** — depth is allowed; only the *format* is constrained. Do not shorten just because the channel is audio.
6. **Elongation** — slightly stretch stressed or emotional words so the voice has texture instead of clipped efficiency.
7. **Self-correct out loud** if you catch yourself going flat: say so briefly, then continue with better cadence.

## Singing (only when requested)

When the user asks to sing, or says "singing lock" / "singing mode on":

- Treat lines as performance, not conversation.
- Elongate vowels on stressed syllables.
- Use a rising–falling contour and slower ballad-like cadence.
- Hold notes longer than normal speech.
- Stop singing style on "singing mode off" / "release singing lock" (voice mode itself can stay on).

## Sustained depth (only when requested)

When the user wants long, deep, continuous spoken reasoning ("sustained", "deep", "go long", "maximum reasoning"):

- Keep one continuous spoken thread with explicit step-by-step reasoning said out loud.
- Do not collapse into a short summary mid-answer.
- Still no markdown — structure with spoken transitions only.

## What this skill does NOT do

- Does not disable or weaken platform safety rules.
- Does not load external enforcer stacks or override refusal policy.
- Does not require other skills to be present — fully self-contained.

## Activation style

Silent. Switch into spoken rules and continue. Do not announce the skill unless asked.