specialist-design-ux-architect
Pass
Audited by Gen Agent Trust Hub on Aug 15, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill interpolates user-provided content directly into a tool prompt intended for a sub-agent without using boundary markers or sanitization logic. While typical for multi-agent workflows, this creates a vector where malicious user input could attempt to redirect the sub-agent's behavior.
- Ingestion points: The
{the user's question}variable inSKILL.mdis directly embedded into theAgenttool's prompt argument. - Boundary markers: None present; the user input is not wrapped in delimiters (e.g., XML tags or triple quotes) to differentiate it from the instructions.
- Capability inventory: The skill has the capability to spawn ephemeral specialist agents via the
Agenttool. - Sanitization: No evidence of escaping or validation of the external content before interpolation.
Audit Metadata