persona-driven-ux-walkthrough
Pass
Audited by Gen Agent Trust Hub on Jun 25, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The troubleshooting documentation suggests a scoped command (
pkill) to terminate hung browser processes related to the Model Context Protocol (MCP) Chrome tool. This is intended for environmental maintenance of the agent's tools and does not represent a security risk. - [PROMPT_INJECTION]: The skill exhibits a surface area for indirect prompt injection as it processes untrusted content from external websites via DOM snapshots and screenshots. This is an inherent risk for browsing skills and is addressed by the descriptive reporting nature of the skill. * Ingestion points:
browser_snapshot(SKILL.md). * Boundary markers: The persona's 'Inner Voice' is output within blockquotes to distinguish it from instructions. * Capability inventory:browser_click,browser_press_key,browser_navigate(SKILL.md). * Sanitization: None mentioned; the skill relies on the agent's contextual interpretation.
Audit Metadata