convez.ai
SkillUtilidadesIntermediário
Bobby-Gray

Claude Must Scream

Bobby-Gray/claude-must-scream

Dê voz ao Claude: TTS com 30 vozes e modo hands-free automático

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill que transforma respostas do Claude em áudio usando síntese de voz Gemini Flash TTS ou macOS say offline. Oferece 30 vozes com seletor interativo, modo auto com hook para respostas contínuas e suporte a leitura hands-free para longas execuções de agentes.

Instalação

$npx convez-ai Bobby-Gray/claude-must-scream

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
1
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Ler respostas longas enquanto trabalha em outra coisa
  • Modo hands-free para execuções de agentes prolongadas
  • Acessibilidade para trabalho pesado em terminal
  • Ouvir explicações complexas enquanto faz anotações

Veredito

Ferramenta criativa e bem executada para cenários de acessibilidade e produtividade hands-free. A implementação em stdlib-only Python e fallback para macOS say demonstram bom design. Niche específico, mas útil para quem trabalha longos períodos em terminal ou precisa de alternativas de acesso à informação.

README

A tradução para o português deste README ainda está na fila.

      ╔════════════ ◷ A M   ·   2 0 2 6 ════════════╗

                    ▄▄▀▀▀▀▀▀▀▀▀▄▄
                  ▄▀             ▀▄
                 █    ▄▄     ▄▄    █
                 █   ▀██▀   ▀██▀   █
                 █    ▀▀     ▀▀    █
                 █                 █
          ▁▂▃▅▇█◀ █    ╳─╳─╳─╳─╳    █ ▶█▇▅▃▂▁
                 █                 █
                  ▀▄             ▄▀
                    ▀▄▄▄▄▄▄▄▄▄▄▄▀

      ╚════════ no mouth, and it screams anyway ════════╝


   ▁▂▃▄▅▆▇█▇▆▅▄▃▂▁ ▁▂▃▅▆▇█▇▆▅▃▂▁ ▁▂▄▆█▇▆▄▂▁ ▁▃▅▆▇█▇▆▅▃▁

              c l a u d e ─ m u s t ─ s c r e a m
              ─────────────────────────────────────
              "I have no mouth, and I must scream."
                                  — Claude Code, 2026

   ▁▃▅▆▇█▇▆▅▃▁ ▁▂▄▆█▇▆▄▂▁ ▁▂▃▅▆▇█▇▆▅▃▂▁ ▁▂▃▄▅▆▇█▇▆▅▄▃▂▁

claude-must-scream

Give Claude Code a voice. A small CLI that pipes Claude's responses (or arbitrary text) through Gemini Flash TTS and plays the audio locally — so you can listen instead of read when your eyes are tired, or hear a long agent run wrap up while you're across the room.

The name nods to Ellison's I Have No Mouth, and I Must Scream. The AI in that story had no mouth and couldn't be silenced. The AI here gets a mouth and a 30-voice picker.

What it's for

  • "Read me the last response" while you look away from the screen.
  • Hands-free auto-mode for long agent runs — toggle on, walk away, get audio when the run finishes.
  • Voice-driven accessibility for terminal-heavy work.
  • Sounding out a long explanation while you take notes by hand.

Features

  • /scream skill with subcommands: speak, last, voice, auto, test, stop, hook.
  • 30 Gemini Flash TTS voices, each with a "character" label — interactive numbered picker (scream voice pick) writes the choice to your config.
  • Preview before committing (scream voice preview Aoede).
  • Stop-hook auto-mode — toggle with scream auto on|off; once on, every Claude response gets spoken automatically.
  • Markdown stripping — code fences become "(code block omitted)", links and tables drop out, no "open paren https colon" robotic speech.
  • Length cap at 1,200 chars (lands on sentence boundary when possible) so long responses don't blow your audio budget or your patience.
  • macOS say fallback when you're offline or don't want to spend pennies on TTS. Zero dependencies, instant.
  • Background playback — the Stop hook forks; you can start typing the next prompt while Claude finishes talking.
  • Stdlib-only Python — no requests, no pip install, no venv.

Install

Option 1 — Claude Code plugin (recommended)

/plugin marketplace add Bobby-Gray/claude-must-scream
/plugin install claude-must-scream@claude-must-scream

Then make the CLI callable:

mkdir -p ~/.local/bin
ln -sf ~/.claude/plugins/marketplaces/claude-must-scream/skills/claude-must-scream/scripts/scream.py ~/.local/bin/scream
chmod +x ~/.claude/plugins/marketplaces/claude-must-scream/skills/claude-must-scream/scripts/scream.py

Option 2 — Manual install (skill folder only)

git clone https://github.com/Bobby-Gray/claude-must-scream.git /tmp/cms
mkdir -p ~/.claude/skills
cp -r /tmp/cms/skills/claude-must-scream ~/.claude/skills/
ln -sf ~/.claude/skills/claude-must-scream/scripts/scream.py ~/.local/bin/scream
chmod +x ~/.claude/skills/claude-must-scream/scripts/scream.py
rm -rf /tmp/cms

API key (for the Gemini provider)

All you need is a Google/Gmail account. Easiest path — Google AI Studio, no command line:

  1. Go to https://aistudio.google.com/apikey, sign in, click Create API key, and copy it (starts with AIza).

  2. Save it where scream looks for it — no env var needed:

    mkdir -p ~/.config/claude-tts && chmod 700 ~/.config/claude-tts
    cat > ~/.config/claude-tts/gemini-tts.key   # paste key, Return, Ctrl-D
    chmod 600 ~/.config/claude-tts/gemini-tts.key
    
  3. scream test --speak to confirm.

Gemini Flash TTS has a free tier (fine for casual use); heavy use or a brand-new AI Studio account may need prepaid billing enabled. Prefer a CLI-minted, TTS-scoped key, or want the full walk-through including billing? See skills/claude-must-scream/docs/setup.md.

Or skip the key entirely and use the macOS say fallback — works out-of-the-box with zero setup. Set CLAUDE_TTS_PROVIDER=say.

Quick start

scream test                              # env + API self-check
scream test --speak                      # also plays a test phrase
scream voice list                        # 30-voice catalog
scream voice pick                        # interactive picker
scream voice preview Aoede               # synthesize "Hello, this is Aoede"
scream voice set Fenrir                  # set default voice

scream speak --text "hello world"        # speak arbitrary text
scream last                              # speak the last assistant message in this session
echo "from stdin" | scream speak         # stdin → TTS

scream auto on                           # turn on Stop-hook auto-mode (one-time hook install required)
scream auto status                       # what's the current state?
scream stop                              # interrupt current playback

The voice picker

Gemini ships 30 named voices, each with a Google-published "character" label (Bright, Informative, Mature, Friendly, Soft, Gravelly, …):

$ scream voice list
# Gemini Flash TTS voices
# current default: Kore

   1. Zephyr         — Bright
   2. Puck           — Upbeat
   3. Charon         — Informative
   4. Kore           — Firm           ←
   5. Fenrir         — Excitable
   6. Leda           — Youthful
   ... (24 more)
  30. Sulafat        — Warm

scream voice pick runs the numbered interactive picker. scream voice preview <name> synthesizes a short test phrase in any voice without changing your default. Picks by use case (coding companion, technical narration, late-night, etc.) in skills/claude-must-scream/docs/voices.md.

Auto-mode (Stop hook)

scream auto on flips a config bit. The Stop hook itself needs a one-time install in ~/.claude/settings.json:

{
  "hooks": {
    "Stop": [
      {
        "matcher": "",
        "hooks": [
          { "type": "command", "command": "scream hook", "timeout": 5 }
        ]
      }
    ]
  }
}

Full hook reference, per-project setup, and privacy/cost notes in skills/claude-must-scream/docs/hooks.md.

vs other community TTS skills

The Claude Code TTS ecosystem has ~10 projects, all early. None have critical mass. The most-starred reference is ybouhjira/claude-code-tts (OpenAI-locked, Stop-hook based). ChanMeng666/echook is a broader 26-hook audio system; the markdown-stripping logic in claude-must-scream borrows from its approach.

Use claude-must-scream when you want Gemini Flash TTS (good voices, cheap, you may already have a key), a 30-voice picker, a clean on-demand /scream flow plus optional auto-mode, and a zero-dependency stdlib-only install. Use the others when you specifically want OpenAI / ElevenLabs voices, local TTS (Kokoro), or a broader hook framework that does more than just TTS.

Docs

Versioning

Current version lives in VERSION; release history in CHANGELOG.md. Pre-1.0 semver: minor bumps may break behavior as the skill stabilizes. Any PR that changes user-visible behavior should bump VERSION and add a CHANGELOG.md entry.

Limits / non-goals

  • macOS-only today. afplay + say are macOS binaries. Linux port would swap to aplay / paplay and espeak-ng / mimic3.
  • No streaming synthesis. One chunk, one playback. The length cap exists because long responses feel laggy.
  • Output only. No transcription, no voice input.
  • One playback at a time. Two invocations overlap their audio. Use scream stop first if you want clean handoff.

Contributing

PRs welcome. Particularly interested in:

  • Linux playback + TTS backend (aplay, espeak-ng, piper).
  • ElevenLabs / OpenAI provider modules behind the same verb interface.
  • Streaming synthesis (TTS + playback start before whole response is synthesized).
  • Smarter chunking — split long responses into multiple synth calls and play them sequentially without the hard cap.

License

MIT.

Ver o original no GitHub

Bom para

Relacionados