The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection by processing and displaying text data from the YouTube API (e.g., video titles and descriptions).
Ingestion points: External metadata is fetched from the YouTube API in list-recent.ts and list-uploads.ts and provided by users in youtube-upload.ts and youtube-live.ts via description files.
Boundary markers: There are no explicit delimiters used to separate user-controlled content from agent instructions when displaying metadata.
Capability inventory: The skill possesses capabilities for network access to the YouTube API and local file system operations (writing tokens and reading configuration).
Sanitization: The skill performs metadata sanitization to replace characters like < and > to satisfy YouTube API requirements, but this does not specifically protect against prompt injection.
[COMMAND_EXECUTION]: The skill uses ts-node to execute local TypeScript scripts for YouTube management tasks. These operations are essential to the skill's purpose and are documented as expected behavior.
[CREDENTIALS_UNSAFE]: The skill handles sensitive YouTube OAuth2 credentials. It mitigates risk by advising users to store secrets in a .env file and by setting restricted file permissions (0600) on the generated token file (.youtube-token.json).

youtube-publisher