ElevenLabs Agent Skills
A ready-to-use collection of ElevenLabs voice skills for any AI coding assistant that follows the Agent Skills specification, covering TTS, transcription, dubbing, SFX and music generation.
Evidence shows the API key is passed via an environment variable (ELEVENLABS_API_KEY); no hardcoded credentials or excessive permission claims were found, so sensitive_data_handling gets 2 and source_attribution gets 2 (MIT license with clear copyright). However, the individual skill directories were not provided, so actual API endpoints, data flows, or write/delete side effects cannot be verified: least_privilege, data_flow_transparency, and external_effects get 1, user_confirmation gets 1, rollback gets 0 with no evidence at all, and dependency_security gets 1 since SDK/CLI dependencies are only shown as install commands.
The README claims trigger and functional evals in evals/ with runnable commands, but this is an unverified assertion; self_consistency and failure_messages have only thin documentation-level support (1 each). dependency_availability depends on external SDK versions, and the warning about the wrong npm package name itself signals dependency risk (1).
Audience is developers building with ElevenLabs products, covering TTS/STT/agent scenarios with Python/JS/CLI support, so audience_and_scenarios gets 2; evals/results with trigger evaluations support trigger_precision (2). But capability_boundaries and environment_fit (per-skill prerequisites, runtime differences) get 1 because skill-internal files are missing.
Repository structure is clear (per-skill directories, references/ docs, evals/), naming is consistent, install instructions are complete: information_architecture, install_notes, and naming_stability get 2; license has full MIT text (3). But examples/FAQ are only claimed to exist in references/ without being shown (1); there is no version number or CHANGELOG (versioning_changelog 1); maintenance responsibility is only implied by an unverified publisher identity (1); only one known limitation is documented — the deprecated npm package warning (1).
One-line install command, ten skills covering the full voice product line: marginal_value 2; output forms (skill spec, eval reports) are usable for developers: output_usability 2; cost is mainly one API key: cost_benefit 2; but all value claims remain unverified by execution.
README claims (eval results, references docs, SDK support) cannot be cross-checked within the provided files: claim_traceability and cross_source_corroboration each get 1; the README separates factual description from warnings (e.g., the deprecated package notice), so fact_inference_separation gets 2.
- This is a static review only; all eval capability claims are unverified by independent execution.
- Skill-internal files were not provided, so actual network calls, side effects, and permission scope cannot be verified — manually review each skill's source before use.
- The README explicitly warns that @elevenlabs/elevenlabs-js is easily confused with the npm package 'elevenlabs'; verify package names on install.
- The API key is configured via an environment variable — avoid leaking it in logs or eval results; no rollback mechanism is documented.
- Publisher identity is unverified by the FollowAgents registry; do not trust this repository on brand alone.
What does this agent do, and when should you use it?
elevenlabs/skills is the official collection of agent skills for ElevenLabs developer products, built to the agentskills.io Agent Skills specification and usable with any compatible AI coding assistant. The repository ships ten skills: text-to-speech, speech-to-text, speech-engine, agents, sound-effects, music, voice-changer, voice-isolator, dubbing and setup-api-key. Every skill requires an ELEVENLABS_API_KEY environment variable and works through the Python SDK, the JavaScript/TypeScript SDK, or the ElevenLabs CLI, which wraps the REST API. Installation is a single command: npx skills add elevenlabs/skills. An evals/ directory provides trigger and functional evaluation scripts that write timestamped reports to disk. The package is a skill bundle loaded into coding agents, not a standalone long-running service.
The repository provides reusable instruction and example directories: text-to-speech converts text to lifelike speech with ElevenLabs AI voices; speech-to-text transcribes audio files to text with timestamps; speech-engine adds real-time voice conversations to a custom LLM or chat agent; agents builds conversational voice AI agents; sound-effects generates sound effects from text descriptions; music generates music tracks via AI composition; voice-changer transforms a recording's voice into a different target voice (speech-to-speech); voice-isolator removes background noise and isolates vocals/speech; dubbing dubs audio/video into other languages while preserving the original speakers' voices; setup-api-key guides through obtaining and configuring an API key. Each skill's references/ folder contains SDK setup instructions, and evals/run_all.py supports --trigger-only, --functional-only, --skills and --model flags.
- A developer building a voice app with an AI coding assistant who wants the agent to know how to call ElevenLabs APIs correctly
- An engineer adding real-time voice conversation to a custom LLM or chat agent using the speech-engine skill
- A content-creation agent that needs to generate voiceover, sound effects or background music via text-to-speech, sound-effects and music
- A team localizing video with the dubbing skill, preserving original speakers' voices in other languages
- Anyone cleaning up recordings with voice-isolator for noise removal and vocal isolation
- A team maintaining agent systems built on these skills who runs the trigger and functional evals to verify behavior
What are this agent's strengths and limitations?
- Built to the open Agent Skills specification, so it works with any compatible AI coding assistant rather than locking into one agent product
- Covers the full ElevenLabs voice surface (TTS, STT, dubbing, voice changing, isolation, SFX, music) in a single install
- Provides three call paths — Python SDK, JavaScript/TypeScript SDK and CLI — and explicitly warns against the outdated v1.x npm package
- Ships a built-in eval suite verifying both skill triggering and functional correctness
- Every skill depends on the ElevenLabs API and a paid API key; nothing works off the ElevenLabs platform
- The eval harness is tied to the Cursor Agent CLI and Cursor authentication, adding toolchain overhead to reproduce results
- No pricing or usage information is provided in the repository; API cost must be assessed separately before adopting
- The skills are instruction bundles for coding agents, not a standalone deployable service, and require a spec-compatible agent runtime
How do you install or deploy this agent?
- Install the skill bundle: npx skills add elevenlabs/skills
- Get an API key: use the setup-api-key skill or the ElevenLabs dashboard at https://elevenlabs.io/app/settings/api-keys
- Set the environment variable: export ELEVENLABS_API_KEY="your-api-key"
- Install SDKs as needed: Python with pip install elevenlabs; JavaScript/TypeScript with npm install @elevenlabs/elevenlabs-js; CLI with npm install -g @elevenlabs/cli or brew install elevenlabs/tap/elevenlabs
Warning: do not run npm install elevenlabs — that is an outdated v1.x package.
How do you use this agent?
Load the skills into any AI coding assistant compatible with the Agent Skills specification, then issue a related request (e.g. 'convert this text to speech'); the matching skill fires and guides the agent to call the SDK or CLI, which reads ELEVENLABS_API_KEY automatically. Each skill's references/ folder contains full setup guidance including migration from deprecated packages. Run evaluations with python3 evals/run_all.py -v (all), --trigger-only (~3 min), --functional-only (~15 min), --skills text-to-speech agents (specific skills), or --model gpt-5.4-high (custom model). Evals require the Cursor Agent CLI on PATH and Cursor authentication via cursor-agent login or CURSOR_API_KEY. Results land in evals/results/<timestamp>/ with a report.md and results..