The Agent Skills Directory

[COMMAND_EXECUTION]: The Python script executes the yt-dlp command-line utility to retrieve video metadata and subtitles. It uses a list-based argument structure with subprocess.run, which is a security best practice to prevent shell injection.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of yt-dlp via system package managers (brew or pip) and fetches external content (subtitles) from YouTube during execution.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted text data from external sources (YouTube transcripts). An attacker could place malicious instructions within video subtitles to influence the behavior of the AI agent when it processes the transcript.
Ingestion points: The scripts/get_transcript.py file downloads and reads content from external YouTube subtitle files.
Boundary markers: There are no delimiters or specific instructions provided to the agent to treat the fetched transcript as untrusted data.
Capability inventory: While the script only prints text, the agent using this skill typically performs summarization or reasoning tasks on the provided text.
Sanitization: The clean_vtt function removes technical metadata and timestamps but does not filter or sanitize the actual text content for malicious instructions.

youtube-watcher