The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): Path Traversal in scripts/extract_transcript.py. The script accepts an optional custom_filename command-line argument which is joined with the target directory (~/Brains/brain/) using vault_path / custom_filename. There is no validation to prevent directory traversal sequences (e.g., ../), allowing a user or a malicious instruction to overwrite sensitive files outside the intended directory (e.g., ../../.bashrc).
[PROMPT_INJECTION] (HIGH): Indirect Prompt Injection surface (Category 8). The skill ingests data from YouTube, an untrusted external source, and saves it into an Obsidian vault intended for agent context. Ingestion points: Video title, description, and transcript fetched via yt-dlp in extract_transcript.py. Boundary markers: Absent; the content is placed directly into Markdown and YAML frontmatter without clear delimiters or instructions to ignore embedded commands. Capability inventory: File-write access to the local filesystem. Sanitization: Inadequate; while the script sanitizes titles for the default filename, it does not sanitize the contents of the transcript or description for malicious instructions.
[EXTERNAL_DOWNLOADS] (LOW): Requirement for yt-dlp. The skill relies on an external utility to fetch and parse content from third-party websites.

youtube-transcript