The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from YouTube that could contain hidden instructions designed to manipulate the agent's behavior during content generation and account updates.
Ingestion points: Data enters the context through search_youtube_videos (which retrieves titles and metadata from existing YouTube videos) and list_channel_posts (which reads the current titles and descriptions of the user's videos).
Boundary markers: The instructions lack the use of delimiters or explicit "ignore embedded instructions" warnings when handling retrieved search results or existing post content.
Capability inventory: The skill has write access to the user's YouTube account via the update_post tool, allowing it to apply generated descriptions and tags that may have been influenced by malicious external input.
Sanitization: There are no documented procedures for sanitizing, escaping, or filtering the external content before it is used to formulate the final metadata for the video.

yt-seo