original-design

Pass

Audited by Gen Agent Trust Hub on Jul 13, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill facilitates a visual inspiration workflow that involves browsing external websites, which introduces a surface for indirect prompt injection.
  • Ingestion points: The agent is directed to search and analyze content on platforms such as Pinterest, Are.na, cosmos.so, and Dribbble as specified in references/inspiration-browsing.md (Phase 1.5 and Phase 6).
  • Boundary markers: The instructions do not define delimiters or explicit "ignore embedded instructions" warnings when processing content from these external sources.
  • Capability inventory: The skill leverages powerful GUI and browser automation tools such as claude-in-chrome and Computer Use and provides instructions to invoke several downstream production skills based on the results of the design session.
  • Sanitization: The instructions do not include steps for sanitizing, escaping, or validating external input before it is reflected to the user or used to populate the project's design brief template.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 13, 2026, 12:44 AM
Security Audit — agent-trust-hub — original-design