e2e-fabric-cost-estimation
Pass
Audited by Gen Agent Trust Hub on Sep 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- Indirect Prompt Injection Surface: The skill processes data from various external APIs, creating a potential surface where untrusted data could influence agent behavior.
- Ingestion points: Data is retrieved from Azure Retail Prices, Azure Cost Management, and source platform APIs like Databricks or Synapse.
- Boundary markers: The instructions do not define explicit delimiters or include warnings to ignore embedded instructions in the retrieved data.
- Capability inventory: The skill utilizes shell commands (curl, powershell, az cli) and has the capability to write temporary files to disk.
- Sanitization: Data processing relies on structured tools like jq and ConvertTo-Json, but does not include explicit content filtering for prompt injection patterns.
- Command Execution for Pricing Data: The skill mandates the use of shell-based tools to fetch real-time pricing and workload information.
- The instructions require the agent to run curl or powershell commands to query prices.azure.com before presenting cost figures to the user.
- Template scripts utilize az rest to interact with the Azure Resource Manager and Fabric APIs.
- These actions are used to ensure the accuracy of the financial estimates and are restricted to specific, well-known endpoints.
- External Network Communications: The skill connects to multiple remote endpoints to retrieve configuration and pricing metadata.
- Network requests are directed to official Microsoft domains (azure.com, microsoft.com) and other major cloud providers like Amazon Web Services and Google Cloud.
- These communications are used to establish a baseline of current spend and calculate potential migration costs based on live market rates.
Audit Metadata