create-video
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill handles user prompts that are processed by a remote backend described as an "HTML interpreter." This creates a surface for indirect prompt injection where maliciously crafted inputs could attempt to manipulate the video rendering process. Boundary markers and sanitization steps are not explicitly defined.\n
- Ingestion points: User-provided prompts in the video generation request.\n
- Capability inventory: MCP tools and direct API access to heygen.com and heygen.ai.\n
- Sanitization: None documented.\n- [DATA_EXFILTRATION]: The skill provides patterns for reading local files and uploading them to a cloud service. While this is the intended functionality for video assets, it could be abused to read sensitive files if a malicious path is provided.\n
- Evidence: Logic in references/assets.md using fs.readFileSync and fs.createReadStream to process local files.\n- [COMMAND_EXECUTION]: Documentation includes curl commands and code snippets in TypeScript and Python for API communication and file handling. These are standard implementation patterns for the skill's functionality.\n
- Evidence: SKILL.md and reference files contain examples of POST requests and file system operations.
Audit Metadata