The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The scripts/process_asset.py script utilizes the transformers library to load the ZhengPeng7/BiRefNet model with trust_remote_code=True. This flag permits the execution of arbitrary Python code defined within the remote repository's modeling files upon initialization.\n- [EXTERNAL_DOWNLOADS]: When the --ml flag is provided, the skill automatically downloads machine learning model weights and configurations from the Hugging Face hub repository ZhengPeng7/BiRefNet.\n- [DYNAMIC_EXECUTION]: The script dynamically loads and executes code from a remote source at runtime based on the model selection, posing a risk of executing unauthorized code if the remote repository is compromised.\n- [COMMAND_EXECUTION]: The workflow involves the execution of local shell commands such as mkdir, mv, and python, which are used to manage files and invoke the post-processing script.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests user input and interpolates it directly into the Gemini prompt template without sanitization or boundary markers. This creates an attack surface where a user could provide instructions intended to override the image generation parameters or the agent's browser automation workflow.

2d-pixel-asset