vp-pixel-art
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes a local utility script,
scripts/tiling_probe.py, which is used to validate the mathematical logic of Wang tile generation. The script relies exclusively on Python's standard library to perform calculations and generate diagnostic PNG images and JSON reports. It writes only to a user-specified output directory. - [INDIRECT_PROMPT_INJECTION]: The skill identifies an attack surface by ingesting external data such as user briefs and reference images, which could potentially contain malicious instructions.
- Ingestion points: Found in
SKILL.mdandreferences/modes-and-capabilities.md, where the agent is instructed to process text briefs and reference images. - Boundary markers: The skill incorporates a robust "review gates" framework in
references/review-gates.mdto establish strict evidence boundaries and acceptance contracts before processing candidates. - Capability inventory: The skill uses local file writing (diagnostic script) and host-level image generation/observation tools.
- Sanitization: The instructions mandate rigorous technical validation against fixed constraints (palette, grid size, transparency) and manual visual inspection to ensure the integrity of the generated output.
Audit Metadata