skills

Assesses the security and compliance posture of ML model files. Scans pickle, pytorch, safetensors, and other model formats for malicious code using multiple scanners, produces a risk score (0-100), maps findings to global compliance frameworks (ISM-2072, EU AI Act, OWASP LLM Top 10, MITRE ATLAS, NIST AI RMF), and provides prioritized remediation steps. Use this skill whenever the user's intent is to evaluate whether model files are safe, trustworthy, or compliant, not when they want to train, fine-tune, convert, benchmark, or deploy models. This includes checking if a downloaded HuggingFace model is safe to load, auditing a project's model files for supply chain risks, verifying compliance with AI security regulations, investigating suspicious pickle files, or assessing whether trust_remote_code configs are dangerous. The distinction is security analysis OF model files, not working WITH models.