The Agent Skills Directory

[SAFE]: The skill emphasizes security best practices for model publishing, including warnings against using pickle-only PyTorch weights due to security scan triggers and recommending the safer 'safetensors' format.
[SAFE]: It explicitly warns against the use of 'trust_remote_code' in code snippets, noting that many enterprise environments block remote code for security reasons.
[SAFE]: The skill provides guidance on protecting sensitive information, warning that flipping a repository from private to public reveals the entire Git history, including old tokens and checkpoints. It advises secret scanning and the use of environment secrets instead of hardcoding keys.
[SAFE]: It addresses data privacy by warning about the implications of gated access, specifically regarding the collection and responsibility for Personal Identifiable Information (PII) under privacy policies.
[SAFE]: It provides detailed guidance on complying with the EU AI Act, including technical documentation duties and copyright policy requirements for model providers.

huggingface-playbook