TTS skill

Install the skill once for the current user, then use it from any local project with an Agent Skills-compatible coding harness.

Install globally with one prompt

Paste this into your coding harness:

Read and follow https://docs.teenage.works/tts/skill-setup.md to install the TeenageWorks TTS skill globally.

The setup file selects the harness's user-level skill directory, prepares the active project's environment, and requires a signed-in account connection before setup can finish.

What agents can build with it

The installed skill teaches the agent both the official SDK and direct HTTP/WebSocket API. It covers:

• ordinary buffered speech with the default voice, a public voice, or an account-owned saved voice ID;
• request-time voice description without a voice ID or recording;
• request-time speaker matching from only a reference recording;
• controllable cloning from reference audio plus a voice description;
• high-fidelity cloning from reference audio plus its exact transcript;
• fast and expressive generation, guidance, generation budget, seed, output sample rate, and registered LoRA selection;
• PCM, MP3, buffered WAV, raw HTTP audio streaming, typed SSE events, and persistent text-input WebSocket sessions;
• synchronous and asynchronous clients, raw-response metadata, streaming wrappers, project-specific client options, retries, and typed errors;
• public voice discovery, model listing, runtime capability inspection, project balance, and usage totals;
• optional OpenAI and ElevenLabs TTS patches for existing provider integrations.

All four voice-modelling combinations happen inside the speech request that generates the audio. They do not create a reusable voice. voice_description and reference_transcript select separate model paths and cannot be combined.

Supported harnesses

When CODEX_HOME is configured, Codex uses $CODEX_HOME/skills/teenageworks-tts instead of the default ~/.codex/skills/teenageworks-tts. The shared ~/.agents/skills path is supported by OpenCode, Cursor, Gemini CLI, and GitHub Copilot.

Install it manually

Codex

Choose your operating system.

mkdir -p ~/.codex/skills
curl -L https://docs.teenage.works/tts/skill.zip -o teenageworks-tts-skill.zip
unzip -q teenageworks-tts-skill.zip -d ~/.codex/skills

The extracted entrypoint must be ~/.codex/skills/teenageworks-tts/SKILL.md.

Claude Code

mkdir -p ~/.claude/skills
curl -L https://docs.teenage.works/tts/skill.zip -o teenageworks-tts-skill.zip
unzip -q teenageworks-tts-skill.zip -d ~/.claude/skills

The extracted entrypoint must be ~/.claude/skills/teenageworks-tts/SKILL.md.

OpenCode, Cursor, Gemini CLI, and Copilot

mkdir -p ~/.agents/skills
curl -L https://docs.teenage.works/tts/skill.zip -o teenageworks-tts-skill.zip
unzip -q teenageworks-tts-skill.zip -d ~/.agents/skills

The extracted entrypoint must be ~/.agents/skills/teenageworks-tts/SKILL.md.

What the skill prepares

• A project-local .venv using Python 3.9 or newer.
• The published teenageworks package, with the matching official wheel available as a fallback.
• Exhaustive SDK and direct API guidance for speech fields, all request-time voice-modelling combinations, streaming transports, voices, models, capabilities, usage, errors, retries, limits, and provider patches.
• A signed-in browser connection that creates a new project API key and transfers it through a one-time encrypted download.
• A Python runner that injects the stored key only into the child process that needs it.

During setup, the agent runs the browser connection helper and asks you to sign in when credentials are absent. Click Connect this project after sign-in. The helper consumes the encrypted download, stores the new key locally, removes the handoff file, and confirms setup. If the browser does not open automatically, the agent gives you the exact connection URL to open yourself.

Start using it

After the harness reloads the skill, ask it to use teenageworks-tts to build a speech flow. The skill verifies the package import, connects your account without making a paid speech request, then opens the relevant SDK or direct API reference for the task.