ce-frontend-design

Pass

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill and its associated persona (ce-design-iterator) instruct the agent to use browser automation tools like agent-browser, Playwright, Puppeteer, or Cypress. This involves executing shell commands to navigate to URLs, take screenshots, and interact with web elements (e.g., agent-browser scrollintoview, agent-browser screenshot).
  • [PROMPT_INJECTION]: The 'Competitor Research' functionality in the ce-design-iterator persona creates a surface for indirect prompt injection. By navigating to external, potentially attacker-controlled websites to 'extract specific techniques' for code implementation, the agent could ingest malicious instructions hidden in the site's content.
  • Ingestion points: External competitor websites accessed via the browser tool (e.g., in ce-design-iterator.md).
  • Boundary markers: Absent. The instructions do not define delimiters or warnings to ignore embedded instructions in the external data.
  • Capability inventory: The agent has the capability to capture screenshots, read DOM snapshots, and directly modify the local codebase to implement found 'techniques'.
  • Sanitization: Absent. There is no logic for validating or filtering content retrieved from external sites before it influences code generation.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 23, 2026, 09:14 PM
Security Audit — agent-trust-hub — ce-frontend-design