nano-banana-image
Pass
Audited by Gen Agent Trust Hub on May 5, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
subprocess.runto orchestrate task execution between its internal Python scripts (nano_banana_image.py,nano_banana_batch.py, etc.) and to verify the environment viasmoke-test.sh. - [COMMAND_EXECUTION]: The skill integrates with the
lark-clitool to facilitate the optional uploading of generated images to Feishu/Lark cloud storage, as documented in the skill's functionality. - [EXTERNAL_DOWNLOADS]: The skill communicates with OpenRouter (
openrouter.ai) to fetch generated image data. OpenRouter is a well-known service for accessing AI models via API. - [CREDENTIALS_UNSAFE]: The skill manages the
OPENROUTER_API_KEYusing environment variables or a local.envfile, which is a recommended practice for handling API credentials. - [PROMPT_INJECTION]: User-provided prompts are structured into a JSON specification before being sent to the image generation model. While this represents a surface for indirect prompt injection, the risk is mitigated as the skill only processes image data (base64) returned by the model and does not execute text-based instructions from the model's response.
Audit Metadata