Speek

Speek — Text-to-Speech via Kokoro TTS

Speak text aloud using the speek command (GPU-accelerated Kokoro TTS). In this environment, speek is approved to run outside the sandbox via the persisted command prefix ["speek"]; call it directly so it can access GPU and audio devices.

When to use

• Proactively at the conclusion of any task: summarize what you accomplished in one short sentence and speak it.
• When the user explicitly asks you to say or speak something.

Usage

speek "Done. The encoder now uses private Metal buffers with blit staging."

Keep the text short (one sentence, under 20 words). Speek loads a 134MB model on first invocation (~1s), so don't call it repeatedly in quick succession.

Guidelines

• Summarize the outcome, not the process: "Fixed the segfault in the decoder" not "I edited gpu_decoder_metal.rs lines 45-80"
• Use natural, conversational phrasing
• Do not speak errors or failures — only successful completions
• Do not speak if the task was trivial (e.g. answering a yes/no question)