The Agent Skills Directory

[COMMAND_EXECUTION]: The test file tests/test_swipe_paths.py uses subprocess.run to execute the skill's own script for functional validation. The implementation follows security best practices by passing arguments as a list rather than a shell string.
[DYNAMIC_EXECUTION]: The script scripts/swipe.py utilizes dynamic path modification to import classification logic from a sibling skill (ig-reel). This facilitates code reuse within the vendor's ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text data (Instagram reel hooks) provided by the user in a TSV format. This establishes an indirect prompt injection surface. Ingestion points: Data enters the system via scripts/swipe.py reading user-supplied files. Boundary markers: The generated output does not employ explicit delimiters to isolate user-provided data from agent instructions. Capability inventory: The skill is limited to local file system access. Sanitization: Input text is processed for analysis but not sanitized against potential injection patterns.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions for using yt-dlp to fetch publicly available metadata and subtitles from YouTube, which is a standard data collection workflow.

ig-viral