gpt-image-2
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The script allows downloading remote images via the
--image-urloption. It employscurlwith connection timeouts and validates the downloaded content's MIME type and file extension (e.g., .png, .jpg, .webp) before processing, which is standard for image generation tools. - [COMMAND_EXECUTION]: The skill uses Bash and Python to manage API requests and file validation. It follows security best practices by using shell arrays for command arguments and
sys.argvfor passing data to Python subprocesses, preventing shell injection vulnerabilities. - [CREDENTIALS_UNSAFE]: The skill requires a
FREVANA_TOKEN. It correctly prioritizes environment variables (FREVANA_TOKEN) and provides a secure fallback for interactive input (read -s), ensuring that sensitive credentials are not logged or exposed. - [INDIRECT_PROMPT_INJECTION]: The skill processes external image data and user-provided prompts. It mitigates potential risks by strictly validating file extensions and sizes (50MB limit) and by utilizing structured JSON for API payloads.
- [DATA_EXFILTRATION]: Network communication is restricted to the official vendor API endpoint (
ai-factory.frevana.com). Local file access is scoped to the images provided by the user via command-line arguments, with validation checks ensuring only supported image formats are accessed.
Audit Metadata