OpenAI Speech Skill: Transform speech generation from "one-time call" to "reproducible project"

You may have been accustomed to the TTS routine: call an API, send a piece of text, select a sound, and get back an MP3. The thing itself is nothing new, ElevenLabs did it two years ago. But the Speech Skill OpenAI does something different. It is not "generating sounds", but "turning sound generation into a reproducible engineering task." The first time I looked through its SKILL.md, what impressed me most was this sentence: "Prefer the bundled CLI for deterministic, reproducible runs." The core design goal of a voice-generated Skill is actually