skills/garrytan/gbrain/media-ingest/Gen Agent Trust Hub

media-ingest

Pass

Audited by Gen Agent Trust Hub on May 4, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection surface as it ingests and processes untrusted data from external sources.
  • Ingestion points: Data enters the agent context via YouTube transcripts, PDF/Book text extraction, image OCR, and GitHub repository content (README and key files), as described in Phase 1 of SKILL.md.
  • Boundary markers: The instructions do not specify any delimiters or safety warnings to ignore instructions embedded within the processed media items.
  • Capability inventory: The skill uses put_page, add_link, and add_timeline_entry to modify the knowledge base, providing a mechanism for malicious data to potentially influence the internal database (SKILL.md).
  • Sanitization: There is no mention of sanitizing or escaping the content fetched from external sources before it is used to populate brain pages.
  • [COMMAND_EXECUTION]: The skill directs the agent to clone GitHub repositories and read internal files to summarize software architecture. This is a standard functional capability for technical ingestion and does not involve the execution of the downloaded code.
  • [EXTERNAL_DOWNLOADS]: The skill fetches content from well-known external services, specifically GitHub for repository data and YouTube for video transcripts, to perform its documented ingestion tasks.
Audit Metadata
Risk Level
SAFE
Analyzed
May 4, 2026, 07:01 AM