The Agent Skills Directory

[COMMAND_EXECUTION]: The script uses subprocess.run to execute yt-dlp in ingest.py. This is implemented safely using argument lists rather than shell strings, preventing shell injection. The path to the executable is resolved via shutil.which.
[COMMAND_EXECUTION]: Input URLs are rigorously validated in ingest.py using validate_youtube_url. The function enforces strict checks on schemes (http/https), hostnames, and uses a restrictive 11-character regex for video IDs to ensure no malicious shell meta-characters can be passed to the downloader.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted transcript data from YouTube. It mitigates potential injection risks by using the markdown_text function in ingest.py, which applies HTML escaping and markdown character escaping to all external content before writing it to the vault. The ingestion process is limited to data retrieval, and the skill does not grant the ingested content authority over agent instructions.
[SAFE]: The skill follows security best practices for local file operations. It sets os.umask(0o077) to ensure created files are restricted to the current user and uses the private_target function to explicitly verify that output paths do not involve symlinks, which prevents directory traversal and file hijacking attacks.
[SAFE]: The external dependency yt-dlp is a well-known and standard tool for the skill's stated purpose of transcript extraction.

ingest-youtube