microsoft-code-reference
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill instructions do not contain any malicious patterns. The workflow emphasizes local work and primary sources, using the external SandBase API only as a fallback. Explicit gates are present to prevent unauthorized costs and to ensure data minimization.- [COMMAND_EXECUTION]: The skill utilizes vendor-specific tools (
sandbase_run,sandbase_inspect) to execute external search and research tasks. These operations are governed by schema-defined arguments and require user confirmation for execution, mitigating the risk of arbitrary or unauthorized command execution.- [INDIRECT_PROMPT_INJECTION]: While the skill ingests and processes external data, which is a potential vector for indirect prompt injection, it includes specific quality gates that require the agent to cross-check claims, record publication dates, and identify evidence gaps. These instructions significantly reduce the risk of the agent blindly following instructions embedded in retrieved content.
Audit Metadata