feat(skill): add /add-discord-voice-transcription

[mtichikawa]

Opt-in voice transcription for Discord and any other Chat SDK-bridged channel (Slack, Teams, Webex, Google Chat, etc.) via local whisper.cpp on the host. No cloud API, no OPENAI_API_KEY — fully on-device.

Pairs with @ira-at-work's #2317 (add-voice-transcription-free-whisper) which patches Signal/Telegram/WhatsApp adapters directly. This skill is the bridge-side complement — one shared hook in chat-sdk-bridge.ts covers every Chat SDK-bridged channel. Together the two skills close the voice gap on every channel NanoClaw supports.

Addresses the Discord side of @b1ek's #2426 (LLM cant see the image in discord — same shape applies to voice today).

Type of Change

• Feature skill - adds a channel or integration (source code changes + SKILL.md)
• Utility skill - adds a standalone tool (code files in .claude/skills/<name>/, no source changes)
• Operational/container skill - adds a workflow or agent skill (SKILL.md only, no source changes)
• Fix - bug fix or security fix to source code
• Simplification - reduces or simplifies source code
• Documentation - docs, README, or CONTRIBUTING changes only

Description

The Chat SDK bridge is shared by every chat-sdk channel (Discord, Slack, Teams, Webex, Google Chat). Hooking transcription here means all bridge-based channels gain voice support from a single integration point, gated on `process.env.WHISPER_BIN` so it's a no-op for installs that don't opt in.

Files:

• `.claude/skills/add-discord-voice-transcription/SKILL.md` — install instructions following the @ddaniels Signal v2 template (feat(skill): Add Signal channel adapter (V2) #1953). Pre-flight, prerequisites, `git fetch upstream skill/discord-voice-transcription && git merge`, env vars, restart, troubleshooting.
• `.claude/skills/add-discord-voice-transcription/REMOVE.md` — lightweight unset-env path + full code revert.
• `.claude/skills/add-discord-voice-transcription/VERIFY.md` — send a voice memo, check inline transcript.
• `src/transcription.ts` — `transcribeAudioBuffer(Buffer): Promise<string | null>` + `isAudioAttachment(att)` helper. Channel-agnostic. Shells out to `ffmpeg` for normalization (any container → 16 kHz mono WAV) and `whisper-cli` for the transcript. Returns null on any failure or empty output; the bridge only injects `[Voice: …]` when the transcript is non-empty.
• `src/channels/chat-sdk-bridge.ts` — +15 lines inside `messageToInbound()`. Transcript is stored on the attachment entry (for any downstream consumer) and appended to `serialized.content` so the agent sees it inline alongside the audio file placeholder.
• `src/transcription.test.ts` — 8 unit tests covering `isAudioAttachment`, env-gate, transcript trim, empty-output handling, and execFile-failure paths.

For Skills

• SKILL.md contains instructions, not inline code (code goes in separate files)
• SKILL.md is under 500 lines (~160 lines)
• I tested this skill on a fresh clone — pending live end-to-end test on Discord; build + unit tests green (see test plan)

Test plan

• `pnpm run build` clean on main
• `npx vitest run src/transcription.test.ts` — 8/8 passing
• `npx vitest run src/channels/chat-sdk-bridge.test.ts` — 12/12 passing (no regression)
• Live: send a Discord voice memo with `WHISPER_BIN=whisper-cli` set, confirm `[Voice: ]` reaches the agent
• Live: same with `WHISPER_BIN` unset, confirm no behavior change

🤖 Generated with Claude Code