The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest third-party data from the web (HTML, social media metadata, API responses), which creates a surface for indirect prompt injection. However, the skill explicitly defines a 'Untrusted content contract' that instructs the agent to treat retrieved data as untrusted, use visible delimiters, and avoid executing instructions found within external content.
Ingestion points: Web scraping functions using requests, playwright, yt-dlp, and instaloader in SKILL.md.
Boundary markers: Recommends using <EXTERNAL_DATA> tags to delimit untrusted content.
Capability inventory: Network operations, file system access (via yt-dlp and instaloader downloads), and browser automation.
Sanitization: Recommends structured extraction and schema validation.
[COMMAND_EXECUTION]: The skill utilizes libraries such as yt-dlp and playwright which may invoke external subprocesses to perform video extraction or browser automation tasks. These are standard tools for the described use case.

web-scraping