azure-hosted-copilot-sdk

Pass

Audited by Gen Agent Trust Hub on May 14, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • Project Context Ingestion: The skill analyzes files such as AGENTS.md and package.json to determine the codebase configuration. While this is necessary for tailoring the deployment, it represents a surface for indirect prompt injection if these user-controlled files contain instructions designed to influence the agent's behavior.
  • External Template Scaffolding: It uses azd init to download application templates from the azure-samples repository. This is an expected development workflow using templates from a well-known source to bootstrap infrastructure and application code.
  • CLI Tool Execution: The instructions call for the use of official command-line tools including az, azd, and gh. These tools are used for their intended purposes of resource management, project initialization, and authentication within the developer's environment.
  • Credential Management: The skill describes an automated process to retrieve authentication tokens via the gh CLI for storage in Azure Key Vault. This process is part of the standard deployment automation provided by the official Microsoft templates and is used to securely manage secrets.
  • Azure Authentication Standards: The provided guidance correctly emphasizes the use of ManagedIdentityCredential for production environments and DefaultAzureCredential for local development, adhering to Azure's recommended security patterns for passwordless authentication.
Audit Metadata
Risk Level
SAFE
Analyzed
May 14, 2026, 04:31 PM