youtube-to-heygen-longform

Pass

Audited by Gen Agent Trust Hub on Apr 8, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes content from external YouTube videos.
  • Ingestion points: Video content is analyzed by the Gemini 2.0 Flash model via the provided YouTube URL in scripts/youtube_to_heygen_longform.py.
  • Boundary markers: The prompt template in SCRIPT_EXTRACTION_PROMPT does not use explicit boundary markers or instructions to ignore embedded commands within the analyzed video content.
  • Capability inventory: The skill possesses network capabilities (sending data to HeyGen API) and database write access (PostgreSQL).
  • Sanitization: The script performs basic sanitization by removing newline characters and literal \n strings to ensure natural avatar speech, but does not filter for malicious instructional content.
  • [EXTERNAL_DOWNLOADS]: The skill requires several standard Python libraries to be installed from public registries.
  • Dependencies: Requires google-generativeai, google-genai, requests, psycopg2-binary, and python-dotenv.
  • [DATA_EXFILTRATION]: The skill transmits data as part of its core functionality.
  • Network operations: Sends generated scripts and configuration (avatar/voice IDs) to HeyGen's official API (api.heygen.com) for video production.
  • Database access: Logs the original YouTube URL and the resulting HeyGen video URL to a PostgreSQL database specified in the user's environment variables.
  • [COMMAND_EXECUTION]: The skill is designed to be executed via the command line, facilitating the automation of the multi-step video generation pipeline.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 8, 2026, 08:20 AM
Security Audit — agent-trust-hub — youtube-to-heygen-longform