prompt_guard 1.0.0

Detect prompt injection attacks using ONNX models from Hugging Face Hub. Models are lazily downloaded and cached locally. Protects LLM applications from malicious prompts with fast local inference (~10-20ms after initial load).