The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from YouTube transcripts and video metadata, which could contain malicious instructions designed to manipulate the agent's analysis.
Ingestion points: External data is fetched in scripts/fetch_transcript.py from YouTube captions and video descriptions.
Boundary markers: The instructions in SKILL.md and references/analysis-patterns.md do not specify the use of delimiters or 'ignore' instructions to separate the transcript text from the agent's core instructions.
Capability inventory: The skill has the ability to write analysis files to the local filesystem via scripts/analyze_video.py and performs network operations using yt-dlp and youtube-transcript-api.
Sanitization: While metadata (like filenames) is sanitized, the transcript text is processed as raw strings for analysis without filtering for embedded instructions.
[COMMAND_EXECUTION]: The skill executes the yt-dlp command-line utility via subprocess.run in scripts/fetch_transcript.py to fetch video metadata and subtitles.
Evidence: Commands are executed using a list of arguments rather than a shell string, and the target YouTube URL is validated and reduced to a specific 11-character video ID using regex in scripts/utils.py before being passed to the tool, mitigating injection risks.

youtube-analysis