The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill ingests untrusted data from external YouTube transcripts and interpolates it into prompts for AI analysis. Ingestion points: scripts/analyze_transcript.py and scripts/extract_transcript.py (fetching via YouTubeTranscriptApi). Boundary markers: Absent. Subtitle text is directly inserted into prompt templates without explicit delimiters or instructions to ignore embedded commands. Capability inventory: The skill performs filesystem writes and network reads but does not include capabilities such as code execution (eval/exec) or privilege escalation that could be exploited via malicious transcript content. Sanitization: Absent; the script only limits the processed text to the first 8,000 characters.

youtube-transcript