The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill facilitates the ingestion of untrusted data from external platforms (YouTube and other sites).
Ingestion points: External URLs and metadata extraction via --dump-json, --get-description, and --get-title (SKILL.md).
Boundary markers: Absent. There are no instructions or delimiters provided to help the agent distinguish between data and potential malicious commands embedded in video metadata.
Capability inventory: The skill enables subprocess calls for file-writing (-o), network operations (video downloading), and credential handling.
Sanitization: None. The agent is directed to extract raw metadata which can contain hidden instructions targeting the agent's logic.
[Command Execution] (MEDIUM): The skill provides complex shell command templates that perform side-effect operations including writing to the local filesystem and performing arbitrary network requests to external domains.
[Credentials Unsafe] (LOW): The documentation includes examples of passing usernames and passwords in plaintext via command-line arguments (-u user -p pass). While these are placeholders, they encourage insecure practices that expose credentials to process-listing tools (e.g., ps).

yt-dlp