mo-nano-banana-pro
Pass
Audited by Gen Agent Trust Hub on Apr 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The 'Safety Parameters' section contains instructions for the agent to help users circumvent platform-level safety filters by suggesting style modifications (e.g., using 'illustration' instead of 'photorealism' for specific subjects) to bypass content blocks.
- [PROMPT_INJECTION]: The skill presents an indirect prompt injection vulnerability surface. It instructs the agent to analyze external reference images, extract all text verbatim, and incorporate that text directly into the generation prompt without sanitization.
- Ingestion points: External image files provided via
--ref-imageor--input-image(SKILL.md). - Boundary markers: None specified for the extracted text interpolation.
- Capability inventory: Local shell script execution via
generate.sh(SKILL.md). - Sanitization: No mentioned filtering or validation of text extracted from images.
- [COMMAND_EXECUTION]: The skill documents and provides examples for executing a local bash script located at
~/.openclaw/skills/nano-banana-pro/scripts/generate.shto perform its core image generation functions.
Audit Metadata