youtube-notes
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted transcription content from external YouTube videos which is then fed directly into the model context.
- Ingestion points: Transcripts and captions are pulled dynamically from external sources in
scripts/fetch_video.tsusingyoutube-transcript-apior local audio translation. - Boundary markers: There are no strict isolation delimiters or specific system level boundaries enforced around the retrieved transcript fragments to prevent the agent from executing instructions hidden inside captions.
- Capability inventory: The skill possesses file write capabilities to output JSON files or save generated digests directly to the user's local notes vault path.
- Sanitization: Content undergoes whitespace normalizations and structure alignment but lacks filtering for adversarial text patterns or natural language instructions.
Audit Metadata