ui-agent-patterns
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes architectures where agents process user-supplied natural language descriptions (UIRequest.description) to generate code and perform actions. This establishes a processing pipeline for potentially untrusted data.\n
- Ingestion points:
request.descriptioninterpolated into prompts inSynthesisMasterAgentandHierarchicalUIOrchestratorinSKILL.md.\n - Boundary markers: The prompts use labels like
Request:but do not implement strict escaping or "ignore instructions" guardrails.\n - Capability inventory: The patterns describe agents with permissions such as
write_code,run_tests, andaccess_browseracross various class definitions inSKILL.md.\n - Sanitization: Not present in the provided pattern implementation.
Audit Metadata