The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses yt-dlp to fetch metadata and transcripts from YouTube Shorts. This is a well-known service used for research purposes within the skill's defined scope.
[COMMAND_EXECUTION]: The skill executes shell commands using yt-dlp for data collection and runs a local Python script (swipe.py) to process the results. These operations are necessary for the skill's functionality.
[DYNAMIC_EXECUTION]: The swipe.py script dynamically modifies the Python path to import a local scoring module from a relative directory. This is a standard method for organizing multi-script tools within a larger project structure.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from social media (reel hooks, transcripts). While this creates a potential attack surface, the risk is mitigated by the skill's focus on statistical analysis and structural classification rather than direct execution of the ingested text.
Ingestion points: Reads data from captured.tsv and YouTube transcripts via yt-dlp.
Boundary markers: None explicitly defined in the data ingestion phase, as the data is processed by local scripts.
Capability inventory: File writing to ~/.claude/instagram/swipe.md, shell execution of yt-dlp and swipe.py.
Sanitization: The script performs regex-based classification and statistical median calculations rather than executing the raw content.

ig-viral