The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text data extracted from external Douyin videos, including titles, descriptions, and subtitles, which represents an indirect prompt injection attack surface. An attacker could craft video metadata containing hidden instructions to influence the agent's subsequent analysis and content generation steps.
Ingestion points: Data enters the agent's context through the output of scripts/extract_douyin_video.py.
Boundary markers: The instructions do not define clear delimiters or warnings to ignore embedded instructions in the extracted content.
Capability inventory: The skill has the ability to execute local scripts and perform network operations.
Sanitization: There is no evidence of sanitization or filtering applied to the extracted subtitles or descriptions before they are interpreted by the model.
[COMMAND_EXECUTION]: The skill requires the agent to execute a Python script (extract_douyin_video.py) within the /workspace environment to process external URLs. While this is the intended functionality, it involves command-line execution of a script that handles user-supplied arguments.
[EXTERNAL_DOWNLOADS]: The Python script performs network requests to the douyin.com domain and potentially other content delivery networks via the yt-dlp and requests libraries to fetch video data and subtitles.

viral-video-copywriting