design-video

Pass

Audited by Gen Agent Trust Hub on Apr 29, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill employs role-play instructions, directing the agent to adopt the persona of an award-winning UI/UX designer. This is a common behavior-shaping technique but is highlighted as it affects agent identity.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its primary ingestion point: external video data. An attacker could embed visual instructions in the video to influence the agent's code generation.
  • Ingestion points: Multimodal video data passed via the $ARGUMENTS variable in SKILL.md.
  • Boundary markers: Absent. There are no specific instructions to treat text or data found within the video as potentially untrusted.
  • Capability inventory: The agent can perform file system operations including directory creation and writing markdown or source code files (plan.md, phase-XX.md, HTML/CSS/JS files).
  • Sanitization: Absent. The skill does not validate or sanitize the visual content processed by the vision agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 29, 2026, 09:18 AM
Security Audit — agent-trust-hub — design-video