align-ui-to-design-system
Pass
Audited by Gen Agent Trust Hub on Jul 31, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill implements a structured workflow that separates discovery (Review mode) from action (Fix mode). Fixes are only applied when the user explicitly authorizes them, reducing the risk of unauthorized or accidental modifications.
- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill reads repository-local design documentation and code to establish a design system baseline. It does not access sensitive system paths (e.g., .ssh, .aws) and contains no hardcoded credentials or external network exfiltration patterns.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests external data such as repository instructions and design documentation. While this is a potential attack surface, the instructions mandate a classification phase (Accidental deviation, Intentional exception, System gap, Uncertain) which forces the agent to categorize findings and provide evidence before recommending or applying changes, serving as a logical mitigation against adversarial instructions hidden in UI code or documentation.
- [COMMAND_EXECUTION]: No dangerous shell commands, privilege escalation (sudo), or unauthorized subprocess executions were found. The skill relies on standard agent file-system and rendering capabilities.
Audit Metadata