emil-design-eng
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides a controlled framework for design engineering that prioritizes local deterministic tools and the preservation of the user's existing design system and authorization boundaries.\n- [DATA_EXFILTRATION]: Integration with the SandBase API for media generation tasks includes explicit security instructions. The agent is required to send only the minimum necessary data, provide unit pricing and total estimates, and obtain explicit user confirmation before executing any paid or external call.\n- [INDIRECT_PROMPT_INJECTION]: The skill involves inspecting and processing local files and external generative outputs, which creates a standard ingestion surface. This is mitigated by instructions to verify real output against user constraints and source material.\n
- Ingestion points: Project files, runtime environment details, and SandBase tool outputs (
sandbase_run_get).\n - Boundary markers: Explicit instructions to preserve the user's chosen stack and authorization boundaries, and to avoid silently switching providers.\n
- Capability inventory: File system access for design work and external API execution via
sandbase_run.\n - Sanitization: Verification gates require inspecting rendered interfaces, contrast, and performance rather than trusting generated code structure alone.
Audit Metadata