The Agent Skills Directory

[Prompt Injection] (SAFE): No malicious instructions, jailbreak attempts, or override markers were detected in the skill description or body.
[Data Exposure & Exfiltration] (SAFE): The skill includes explicit security checklists advising against token exposure in logs and recommending environment variable protection. No hardcoded credentials or exfiltration patterns are present.
[Unverifiable Dependencies] (SAFE): References well-known and reputable libraries such as huggingface_hub and hf-transfer. No suspicious or unversioned remote package installations are suggested.
[Command Execution] (SAFE): The allowed tools are limited to Read, Grep, and Glob, which are standard for file system navigation and do not include high-risk shell execution capabilities.
[Indirect Prompt Injection] (LOW): While the described utilities process external data from the HuggingFace Hub, the skill explicitly provides remediation guidance including safe file path handling and error handling for network failures.

common-skills