The Agent Skills Directory

Standard Command-Line Fallback: The skill provides instructions to use Azure CLI or Azure PowerShell as a fallback when specific platform operations are unavailable. This is a controlled capability intended for standard Azure management tasks.
Evidence: The references/tools-and-safety.md file outlines a fallback process using az rest or Invoke-AzRestMethod, requiring explicit user approval before execution.
Query Validation and Safety: The skill implements a pattern for Resource Graph queries that includes a validation step before execution to ensure accuracy and safety.
Evidence: Several files, including references/aks-anomaly.md and references/aks-enablement.md, specify a generate_query, validate_query, then execute_query sequence.
Data Ingestion Surface: The skill processes external data returned from Azure cost tools and Resource Graph, which is a potential surface for indirect prompt injection. The skill mitigates this by instructing the agent not to parse or aggregate these results in a local interpreter.
Ingestion points: Data enters the context via query_costs, query_aks_costs, and execute_query (referenced in SKILL.md and references/aks-anomaly.md).
Boundary markers: The skill uses explicit evidence labels (e.g., "Actual cost", "Retail price") and instructs the agent to separate categories like #idle charges# from actual workloads.
Capability inventory: Capabilities include network requests to Azure APIs via MCP tools and fallback command execution via az or Invoke-AzRestMethod (in references/tools-and-safety.md).
Sanitization: The skill requires the use of validate_query for Resource Graph and explicitly forbids local interpretation or aggregation of tool results to prevent unsafe parsing of external data.

cost-analysis