technology-selection
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill guides the agent in building RAG and multi-step agent systems that ingest external data, creating an attack surface that the skill proactively addresses with design-time remediations.\n
- Ingestion points: The skill describes patterns for document ingestion from repository files (PDF/markdown) into vector databases (references/rag.md).\n
- Boundary markers: Guidance includes mandatory similarity score thresholds for retrieval and schema validation for non-deterministic model outputs (SKILL.md, references/llm.md).\n
- Capability inventory: Orchestrated tool-calling and multi-step reasoning are enabled through the Microsoft Agent Framework (references/agentic.md).\n
- Sanitization: The instructions enforce bounded iteration limits (MaximumIterations), token budgets, and cost ceilings to mitigate malicious or runaway agent loops (references/agentic.md).
Audit Metadata