As large language models (LLMs) become deeply embedded in developer workflows — from autocompleting code to generating Git commit messages and analyzing diffs — the security and reliability of these AI systems hinge on one critical detail: the prompt.

Prompts are not just inputs; they are the interface between your code and …