elevenlabs/skills

ElevenLabs Skills – Plug‑and‑play AI voice capabilities for any coding‑assistant

What it is – A collection of agent skills that wrap ElevenLabs’ speech AI (text‑to‑speech, speech‑to‑text, voice‑changing, dubbing, music generation, etc.). The skills follow the open Agent Skills specification and can be invoked by any compatible AI coding assistant (e.g., Cursor, Claude‑based agents).

How to get it – Install the skill pack with a single npx command:

npx skills add elevenlabs/skills

All the individual skills are then available to the assistant.

Key capabilities

Skill	What it does
text-to-speech	Turn arbitrary text into lifelike speech using ElevenLabs’ AI voices
speech-to-text	Transcribe audio files, returning timestamps
speech-engine	Hook real‑time voice input/output into a custom LLM or chat agent
agents	Build full‑featured conversational voice agents
sound-effects	Generate short sound‑effects from textual prompts
music	Create AI‑composed music tracks
voice-changer	Convert a recorded voice into another target voice (speech‑to‑speech)
voice-isolator	Strip background noise and isolate the spoken component
dubbing	Translate and dub audio/video while preserving the original speaker’s voice
setup-api-key	Interactive guide to obtain and configure an ElevenLabs API key

Configuration – All skills need an ElevenLabs API key. Set it once in your environment:

export ELEVENLABS_API_KEY="your-api-key"

You can obtain the key via the setup-api-key skill or the ElevenLabs dashboard.

SDK / runtime support

• Python – pip install elevenlabs
• JavaScript/TypeScript – npm install @elevenlabs/elevenlabs-js (the recommended package; the older elevenlabs v1.x is deprecated)
• CLI – npm install -g @elevenlabs/cli or brew install elevenlabs/tap/elevenlabs. The CLI reads ELEVENLABS_API_KEY automatically. Each skill ships a references/ folder with detailed setup instructions for the language you choose.

Testing / evaluation – The repo includes an evals/ suite:

• Trigger tests – Verify that a skill fires for the correct user query (≈3 min).
• Functional tests – Check that the skill returns correct output (≈15 min). Run them with the provided Python script, optionally selecting specific skills or a custom model (e.g., gpt-5.4-high). Results are written to evals/results/<timestamp>/ as a markdown report and JSON file. The evaluations rely on the Cursor Agent CLI (cursor-agent) and require you to be logged in (cursor-agent login or CURSOR_API_KEY).

License – MIT, so you can freely use, modify, and redistribute the skills.

---

All information above is taken directly from the repository’s README.