voice-reply-mode

# Voice Reply Mode Use this skill when a user wants an agent to behave like this: - 用户发语音 → agent 回语音 - 用户发文字 → agent 回文字 ## What this skill includes - Workspace-level behavior rules for `SOUL.md` / `IDENTITY.md` / `TOOLS.md` - A reusable Edge TTS helper script - Gateway config snippets for `messages.tts` - Channel notes for Telegram / Feishu style deployments ## Important boundary This skill can package **rules, scripts, and config snippets**. It does **not** automatically change a user's global gateway config unless the user explicitly asks and authorizes it. ## Recommended workflow 1. Read `references/workspace-snippets.md` 2. Read `references/gateway-config.md` 3. Copy the relevant snippets into the target agent workspace 4. If the user explicitly asks, patch gateway config with the snippet from `references/gateway-config.md` 5. Validate with one text message and one voice message ## Minimal success criteria - Text message receives text reply - Voice message receives voice reply - Agent workspace documents the preferred voice ## Files in this skill - `references/workspace-snippets.md` — snippets for `IDENTITY.md`, `SOUL.md`, `TOOLS.md` - `references/gateway-config.md` — `messages.tts` examples and caveats - `references/channel-notes.md` — Telegram / Feishu notes - `scripts/edge-tts.sh` — helper script for local TTS generation ## Notes - In many deployments, the decisive switch is gateway-level `messages.tts.auto = "inbound"`. - Workspace files define behavior expectations, but gateway config determines whether automatic voice replies actually happen. - If schema rejects `identity.voice`, keep voice preference in workspace docs instead.