@crewhaus/prompt-injection-detector 0.7.0

Detection regexes, not malicious behavior

“Cross-cutting safety classifier for tool outputs (regex + structural heuristics + optional LLM tier)”