ux-researcher-designer
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external research data, including user interviews and surveys, which constitutes a surface for indirect prompt injection. If the processed data contains malicious instructions, the agent could potentially be influenced during synthesis.
- Ingestion points: Data is ingested via Workflow 1 and Workflow 4 within SKILL.md and processed by the scripts/persona_generator.py script.
- Boundary markers: The workflows suggest using structured JSON for input data, which provides some structural delimitation, but there are no explicit instructions for the agent to ignore embedded commands in qualitative fields.
- Capability inventory: The skill includes a Python script for data processing, but the script is limited to standard library operations for data aggregation and lacks network access, file system write permissions, or dynamic code execution capabilities.
- Sanitization: No explicit sanitization or filtering logic is defined in the instructions or the accompanying script to mitigate injection in the qualitative data streams.
Audit Metadata