Nano Banana 2 Image Generation Master
Fail
Audited by Gen Agent Trust Hub on Mar 6, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill documentation encourages the execution of shell commands where complex prompts, potentially containing unsanitized user input, are passed as direct arguments to Python scripts (e.g., generate_fal.py, generate_euri.py). This creates a significant risk of shell command injection.
- [PROMPT_INJECTION]: The skill provides explicit instructions to 'neutralize' model biases and bypass normalization filters. It includes a list of 'blockers' like 'anatomy normalization' and 'body proportion averaging' to be used in negative prompts to override standard model safety and alignment protocols.
- [EXTERNAL_DOWNLOADS]: Scripts within the skill (generate_kie.py, get_kie_image.py) automatically download image files from URLs returned by the Kie.ai API. This process does not validate the source URL or the content, posing a risk of interacting with untrusted infrastructure.
- [CREDENTIALS_UNSAFE]: Python scripts in the skill programmatically read from a .env file located in the parent directory to access sensitive API keys (KIE_API_KEY). This exposes the local filesystem and sensitive environment variables to the skill's execution context.
Recommendations
- AI detected serious security threats
Audit Metadata