[{'Text': '<skill_content name="voice-message">\n<source>global</source>\n<directory>/Users/peterdonaghey/.agents/skills/voice-message</directory>\nRelative paths in this skill resolve against <directory>.\n\n# Voice Message\n\nWhen you finish a task, have an important update, or want to ask the user something, speak to them instead of writing a wall of text. The user prefers hearing your response in natural language.\n\n## How to send a voice message\n\n1. **Write the message** in natural conversational language — as if you\'re speaking directly to them. Keep it concise. Use the Kokoro TTS backend with the `af_heart` voice unless the user has specified otherwise.\n\n2. **Generate the audio** using the TTS CLI:\n   ```bash\n   python3 tts/tts_cli.py "Your message here." \\\n     --backend kokoro \\\n     --voice af_heart \\\n     --output /tmp/voice_message.wav\n   ```\n\n3. **Open it in their default player** so it plays immediately:\n   ```bash\n   open /tmp/voice_message.wav\n   ```\n\n## When to use it\n\n- **Task completion** — summarize what you did, what happened, any deviations\n- **Asking a question** — "I need your input on X, here are the options..."\n- **Next steps** — what you recommend doing next\n- **Errors or issues** — explain what went wrong and what you\'re doing about it\n- **Any time your response would be more than a few sentences** — voice is faster to consume\n\n## Tips\n\n- Speak naturally. Don\'t list bullet points verbally.\n- If there are action items, state them clearly: "Next, I need you to..."\n- Keep it under 30 seconds of audio if possible.\n- The `--backend supertonic --voice F1` option is faster to generate (but lower quality) if you need speed.\n- The project root is `/Users/peterdonaghey/Projects/notes`.\n- Run commands from the project root directory.\n</skill_content>\n'}]