The Agent Skills Directory

[PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection. It is designed to ingest and process untrusted data from multiple external sources, including YouTube transcripts, Reddit threads, and web articles. While the skill uses structured templates and sub-agents to process this data, a sophisticated attack embedded in the source material could potentially influence the agent's behavior during the note-generation phase.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute local utility scripts and common command-line tools such as curl, jq, pdftotext, and grep. These are used for legitimate metadata extraction and content validation, but represent a high-privilege capability that operates on data derived from external URLs.
[EXTERNAL_DOWNLOADS]: The skill performs numerous network requests to external services, including GitHub, YouTube, Reddit, Goodreads, and TMDB. These downloads are part of the core functionality for fetching metadata and content. The skill prioritizes well-known services and includes some safety-conscious instructions, such as avoiding direct fetching of IMDB or constructing unverified YouTube URLs.

adding-notes