design-site-icon
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from external websites and local project files to inform design choices.
- Ingestion points: The workflow in
SKILL.md(Steps 1 and 2) explicitly directs the agent to "check the URL given by the user," "check its actual icon files and page usage," and analyze reference websites. - Boundary markers: There are no boundary markers or explicit instructions to ignore embedded directions within the ingested data.
- Capability inventory: The skill has the capability to write files via
scripts/build_preview.pyand integrate assets into the local project structure in Step 6. - Sanitization: While
scripts/build_preview.pyuseshtml.escapefor HTML generation, the agent's high-level reasoning is exposed to instructions embedded in the external content it researches. - [COMMAND_EXECUTION]: The skill workflow involves executing local Python scripts (
build_preview.pyandvalidate_svg.py) to automate design tasks. - The
build_preview.pyscript reads multiple SVG files and writes a new HTML preview file to a user-defined or agent-defined directory. - The
validate_svg.pyscript reads and parses SVG source code to verify compliance with safety and design rules, including checking for embedded scripts or external network resources.
Audit Metadata