designscope
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from user-provided URLs and Figma links as part of its core analysis workflow. • Ingestion points: Step 2 in SKILL.md and capture-flows.md uses web_fetch and extract_css_vars.py to ingest remote HTML and CSS properties. • Boundary markers: Instructions do not explicitly define boundary markers or 'ignore' directives for the ingested external content. • Capability inventory: The skill utilizes bash, web_fetch, view_file, and write_to_file for its operations. • Sanitization: Data is processed via regex-based scripts and LLM interpretation without a formal sanitization layer before prompt interpolation.
Audit Metadata