azure-role-selector
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill interacts with repository content and external data sources, which presents a surface for indirect prompt injection.\n
- Ingestion points: The workflow involves inspecting repository files, runtime inputs, and external documentation via SandBase tools (SKILL.md).\n
- Capability inventory: The agent can execute discovered capabilities via the
sandbase_runtool and apply or deploy Azure configurations (SKILL.md, references/sandbase-api-map.md).\n - Boundary markers: The skill instructions enforce a 'SandBase boundary' and require explicit user authorization before performing deployments or executing paid calls (SKILL.md).\n
- Sanitization: Security is maintained through verification tests and outcome validation rather than explicit input filtering (SKILL.md).
Audit Metadata