The Agent Skills Directory

Command Execution: The skill utilizes subprocess calls to the Azure CLI (az) to facilitate authentication and discover workspace details. This is a standard pattern for Fabric-integrated tools. It also employs the system's browser to open task status pages, using a controlled wrapper that filters sensitive environment variables (like bearer tokens) and validates the target URL to ensure it points to supported Fabric hosts.
Sensitive Data Management: The skill handles Microsoft Workload Context (MWC) and Power BI tokens. It follows security best practices by storing these credentials in local files with restricted access permissions (chmod 600) rather than passing them through command-line arguments or environment variables where they could be exposed to other processes.
Network Operations: Several scripts use the Python standard library to communicate with official Microsoft Fabric and OneLake service endpoints (api.fabric.microsoft.com and onelake.dfs.fabric.microsoft.com). These network interactions are necessary for orchestrating remote data engineering tasks and uploading task instructions.
Indirect Prompt Injection Surface: The skill ingests "User outcomes" which are interpolated into handoff contracts for a remote SparkCore service. To mitigate potential prompt injection risks, the skill implements instruction length validation via a dedicated script, uses clear markdown boundary markers to separate user intent from the execution plan, and enforces robust URL parsing to prevent redirection to untrusted portals.

project-osmos