Git for your LLM prompts.

Version every change, test prompts and catch regressions before they ship, and fetch the current prompt from the edge at runtime — no hardcoding.