text-in-images
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data by instructing the agent to review rendered HTML and interactive components. This ingestion of untrusted data creates a surface for indirect prompt injection if the audited content contains instructions designed to manipulate the agent.
- Ingestion points: The agent reviews rendered HTML and design-system patterns as specified in
SKILL.md. - Boundary markers: The skill does not provide explicit boundary markers or instructions to treat external HTML content as data rather than instructions.
- Capability inventory: The skill provides a client-side JavaScript snippet in
references/rule.mdfor identifying potential issues, but it does not include dangerous capabilities like arbitrary command execution or network exfiltration. - Sanitization: No explicit sanitization or filtering of the ingested HTML is mentioned.
Audit Metadata