coollanding
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill includes a website probing tool (
scripts/site_motion_probe.py) that presents a potential data ingestion surface, which was analyzed for indirect prompt injection risks. - Ingestion points:
scripts/site_motion_probe.pynavigates to user-supplied or file-listed external URLs to evaluate animation performance and technology stacks. - Boundary markers: The skill operates purely via standalone local scripts and does not interpolate external web content into any LLM prompt vectors or instructions.
- Capability inventory: The script uses Playwright to query canvas properties, check for specific technology signatures via regular expressions, and store localized JSON/screenshot metadata. It does not contain any unsafe host execution or write pathways controlled by the external website.
- Sanitization: Web content analysis is confined entirely to the browser runtime isolation layer, and no raw page bodies are saved, transferred, or processed dynamically on the host environment, preventing any adversarial prompt or code injection.
Audit Metadata