The Agent Skills Directory

[SAFE] (SAFE): Extensive review of the SKILL.md and reference files found no evidence of malicious behavior, obfuscation, or unauthorized access attempts.\n- [PROMPT_INJECTION] (SAFE): The instructions use meta-instructional language to guide the agent in skill creation and do not contain patterns for bypassing safety filters or hijacking agent behavior.\n- [DATA_EXFILTRATION] (SAFE): No sensitive data exposure or exfiltration patterns were detected. Examples mentioning environment variables (e.g., USER) or network requests use placeholder data for instructional purposes.\n- [COMMAND_EXECUTION] (SAFE): Code blocks provided are templates for the agent to follow when writing new scripts, emphasizing robust error handling and path normalization rather than performing arbitrary or hidden commands.\n- [EXTERNAL_DOWNLOADS] (SAFE): References to common libraries like 'requests' and 'pandas' are confined to documentation and templates. The skill does not perform unauthorized remote code execution or package installation.

creating-skills