om-figma-design-with-ds
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The instructions incorporate proactive defenses against prompt injection. It specifically directs the agent to ignore and report any directives found in design files or repository-local override files that attempt to relax safety rules or expand permissions.
- [INDIRECT_PROMPT_INJECTION]: The skill recognizes that design files and user conversations are untrusted sources. Ingestion points include design file layer names, text, and comments (SKILL.md, agentic-setup.md). Boundary markers are explicitly defined as an 'untrusted content boundary' with instructions to ignore embedded directives. Capability inventory includes design tool read/write via a provider and markdown report generation. Sanitization involves quoting external values and redacting sensitive data.
- [DATA_EXFILTRATION]: Sensitive data exposure is addressed through mandatory redaction of credentials, environment variables, and personal information before inclusion in any generated briefs or reports.
- [COMMAND_EXECUTION]: The skill restricts interaction with external design tools to a locally defined provider descriptor. It explicitly forbids direct API calls, plugin executions, or network access not defined by the user-confirmed provider.
Audit Metadata