The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted transcript data from YouTube which could contain malicious instructions for the agent. Ingestion points: The file scripts/get_transcript.py fetches VTT transcript files from external URLs. Boundary markers: No delimiters or isolation instructions are present in the skill's usage guide or configuration. Capability inventory: The skill has the capability to execute subprocesses via the yt-dlp tool. Sanitization: The clean_vtt function in scripts/get_transcript.py removes metadata and HTML-like tags but does not filter for natural language instructions in the text.- [COMMAND_EXECUTION]: The script scripts/get_transcript.py executes the yt-dlp command-line utility. Evidence: It uses subprocess.run with a command list to fetch subtitles. This usage is safe from shell injection as it does not use shell=True and passes arguments as a list.- [EXTERNAL_DOWNLOADS]: The skill connects to YouTube to retrieve subtitle and metadata files as part of its core functionality. Evidence: Uses yt-dlp in scripts/get_transcript.py to fetch data from user-provided URLs.

youtube-watcher