The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and processing untrusted data from external sources (YouTube transcripts and comments).
Ingestion points: Transcripts are downloaded via yt-dlp and comments are fetched via the YouTube Data API (as described in SKILL.md).
Boundary markers: The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when the agent subsequently processes or reads the stored transcript chunks and comments.
Capability inventory: The agent is granted powerful capabilities including Bash, Write, and Edit, which could be targeted if malicious instructions are embedded in video content or comments.
Sanitization: While the skill mentions stripping VTT headers and deduplicating lines, it does not include sanitization logic to detect or filter out prompt injection attempts within the content.
[COMMAND_EXECUTION]: The skill relies on shell command execution via the Bash tool to run scraping scripts and external utilities.
The $ARGUMENTS placeholder and VIDEO_ID variables are used to construct command lines for pnpm and yt-dlp. Maliciously crafted channel handles or video IDs could potentially lead to command injection if the agent or the platform does not properly sanitize these inputs before execution.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of yt-dlp, a third-party command-line utility.
The instructions recommend installing the tool via pip install yt-dlp. This uses the official Python Package Index (PyPI), which is a well-known and trusted package registry.

youtube-scrape