azure-foundry-agent

Pass

Audited by Gen Agent Trust Hub on Jun 21, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local shell command az account show --query "{subscriptionId:id, name:name}" -o json to retrieve the current Azure subscription ID and account name. While functional for its purpose, it accesses system environment metadata.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it interpolates the raw user message directly into the query parameter of the foundry_agents_connect tool. There are no boundary markers or sanitization steps to prevent instructions inside the user message from potentially manipulating the behavior of the Azure Foundry agent.
  • Ingestion points: User input captured via the <user-message> placeholder in SKILL.md.
  • Boundary markers: None provided in the tool call structure.
  • Capability inventory: The skill has the capability to connect to and invoke remote Azure Foundry agents via the azure-mcp-foundry tool.
  • Sanitization: No evidence of input escaping or validation is present.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 21, 2026, 02:10 PM
Security Audit — agent-trust-hub — azure-foundry-agent