mage-inpaint
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions require the agent to execute a locally provided Python script (
mask.py) to apply a visual mask to images. It also utilizescurlcommands to upload these processed images to the Mage service. - [EXTERNAL_DOWNLOADS]: The skill downloads images from user-specified URLs to the local environment for editing. It also instructs the installation of the
PillowPython library to facilitate image manipulation. - [INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection by processing external image data and user-provided descriptions that are directly interpolated into model prompts.
- Ingestion points: Fetches image files from remote URLs and accepts free-form text descriptions for image modifications.
- Boundary markers: Uses a fixed prompt template to contextualize the user input, though it lacks explicit instructions to ignore potentially malicious commands embedded in the input.
- Capability inventory: Includes file system write access, execution of Python scripts, and network operations via
curlfor data uploads. - Sanitization: No specific sanitization or validation of the user-provided text or image metadata is performed before processing.
Audit Metadata