The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external information when defining learning tasks and gathering evidence. It includes specific mitigation instructions to treat all retrieved content as evidence rather than instructions and to explicitly ignore any directives found within source pages.- [DATA_EXFILTRATION]: The skill incorporates strong data privacy guidelines, such as prohibiting the inclusion of remote network requests, CDNs, or analytics in generated HTML files. It also instructs the agent to keep personal paths and credentials out of shareable artifacts.- [COMMAND_EXECUTION]: To ensure the quality of generated media, the skill employs system tools like ffmpeg and ffprobe for artifact verification. These tools are used for technical integrity checks on the agent's own output rather than executing external code.

visual-explanation