The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves data from multiple Azure management tools, establishing a surface for indirect prompt injection if the source records contain malicious payloads.
Ingestion points: Outputs from list_benefit_utilization, get_benefit_recommendations, query_costs, list_reservation_transactions, and execute_query are consumed by the agent.
Boundary markers: The instructions lack specific delimiters or guardrail prompts to ignore potential commands embedded within the retrieved tool data.
Capability inventory: The skill possesses the ability to execute resource discovery via execute_query and access sensitive financial data via query_costs.
Sanitization: A validate_query step is utilized to verify query structure and property paths prior to execution.
[DYNAMIC_EXECUTION]: The skill dynamically generates and runs KQL (Kusto Query Language) queries for the Azure Resource Graph at runtime.
It leverages generate_query to adapt query templates based on the specific resource types found in the user's environment.
Execution occurs via execute_query only after the query has passed a syntax validation check.

commitment-analysis