ai-architecture-advisor
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references a large volume of external research papers and documentation in
data/sources.jsonto support its architectural recommendations. These links target trusted organizations including Anthropic, Meta, Google, Amazon, and NVIDIA, as well as well-known academic services like arXiv and Nature. These resources are utilized for technical guidance and are documented neutrally. - [INDIRECT_PROMPT_INJECTION]: The skill identifies and processes user-provided constraints to recommend AI architectures, creating a surface for potential injection. Ingestion points: Natural language problem statements and data constraints described in SKILL.md. Boundary markers: The skill implements an 'Intake' protocol requiring six questions to be answered before providing a recommendation. Capability inventory: Routing to 36 specialized skills and updating local history via
append_learning.py. Sanitization: No explicit input sanitization or escaping mechanisms are specified for the user-supplied constraints.
Audit Metadata