oracle
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates as an AI design specialist, ingesting untrusted user requirements to generate system architectures, prompt designs, and evaluation suites.
- Ingestion points: System design requirements and AI feature requests provided through user input across all operating modes (ASSESS, DESIGN, EVALUATE, SPECIFY).
- Boundary markers: The skill explicitly recommends and provides templates for XML tags (e.g.,
<instructions>,<context>) and maintains a doctrine of instruction-vs-data channel separation to mitigate injection risks in generated content. - Capability inventory: The skill generates specifications for other agents (Builder, Radar, etc.) and does not possess direct file-writing, network-access, or subprocess execution capabilities within its own instruction set.
- Sanitization: The instructions mandate input sanitization, PII masking, and output validation as critical gates that must be included in any designed architecture before implementation.
Audit Metadata