onelake-catalog-cli
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- Command Line Interface Interaction: The skill operates by executing shell commands, primarily using the
azCLI andcurlto interact with Microsoft Fabric and Power BI REST APIs. This is the intended mechanism for administrative tasks but requires the agent to have appropriate local execution permissions. - Runtime Token Acquisition: The instructions direct the agent to retrieve authentication tokens using
az account get-access-token. This follows security best practices by not hardcoding credentials, though it necessitates that the environment where the agent runs is pre-authenticated via the Azure CLI. - Processing External Item Metadata: The skill ingests item definitions and table schemas from the Fabric environment to generate metadata suggestions. This data ingestion represents a potential surface for indirect prompt injection, as malicious content within a workspace item could attempt to influence the agent's behavior. The skill mitigates this by requiring explicit owner review and approval for every suggestion before a write operation occurs.
- Scripted Polling Mechanisms: The skill provides a PowerShell script to manage long-running operations when retrieving item definitions. This script involves polling API endpoints and writing temporary files to handle large payloads, which is a common pattern for managing asynchronous cloud operations.
Audit Metadata