The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple shell commands to handle content processing.
Invokes yt-dlp for YouTube transcript extraction.
Uses curl for downloading PDFs and article headers.
Executes python3 snippets for text cleaning and HTML parsing.
Attempts to automatically install missing dependencies using brew install yt-dlp.
[EXTERNAL_DOWNLOADS]: The skill fetches data from remote servers to perform its primary function.
Downloads transcripts and metadata from YouTube via user-supplied URLs.
Downloads PDF files from remote servers.
Fetches web article content via curl and extraction tools.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it processes untrusted data from the web.
Ingestion points: User-provided URLs for YouTube videos, web articles, and PDF documents (SKILL.md).
Boundary markers: The instructions do not explicitly describe sanitizing the textual content for embedded instructions before passing it to the action-planning stage.
Capability inventory: File writing (.txt, .pdf), network requests (curl, yt-dlp), and subprocess execution (python3, pdftotext).
Sanitization: Implements filename sanitization to prevent directory traversal or shell escaping in filenames, but lacks semantic sanitization of the extracted content body.

tapestry