THE FORGEKIT — three tools, one discipline.

1. promptforge.py — prompt version control + regression testing
2. grader.py — structural prompt analysis (7 failure-mode checks)
3. fuzz.py — adversarial mutation fuzzer

PLUG YOUR MODEL: replace runner_local() in promptforge.py and
runner in fuzz.py with your actual API call.
Wire: (prompt: str, input: str) -> str
