skills/everettfish/ip_illustration_for_yourself/ip-illustration-character-system/Gen Agent Trust Hub
ip-illustration-character-system
Pass
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to run local Python scripts (
scripts/make_windows_icon.py) via the shell to convert images into Windows-compatible icons. - [DYNAMIC_EXECUTION]: Included Python scripts (
scripts/clear_rect_alpha.pyandscripts/make_windows_icon.py) perform file system writes and directory creation. These scripts use thePILlibrary to manipulate image data based on command-line arguments. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data from the user, such as articles for illustration or reference images for 'possession' (subject replacement). Although the instructions advise the agent to treat text within images as content rather than instructions, the process of interpolating user-provided text into image generation prompts and script arguments presents a vulnerability surface.
- Ingestion points: Processes user-uploaded photos, article text, and external meme references.
- Boundary markers: Uses structured prompt blocks like
IDENTITY LOCKandSCENEto delineate instructions from user-derived data. - Capability inventory: Executes local Python scripts via subprocess calls and writes to the local file system.
- Sanitization: Lacks explicit validation or sanitization steps for user-provided strings before they are used in script arguments or prompts.
- [EXTERNAL_DOWNLOADS]: The skill fetches several layout and style reference images from the author's GitHub repository. These are identified as vendor-owned resources and are necessary for the skill's stated purpose.
Audit Metadata