The Agent Skills Directory

[SAFE]: The skill implements a 'privacy-first' approach for sensitive data. It instructs the agent to identify potential PII/PHI based on table names and requires explicit user consent before sending sample rows to the LLM.
[SAFE]: A 'Plan-First' requirement is enforced, ensuring the agent presents a numbered TODO list for user approval before any database tools are invoked.
[SAFE]: The instructions mandate cardinality checks and user confirmation of primary keys, preventing erroneous data comparisons that could lead to false results.
[SAFE]: The skill uses specialized tools (data_diff, warehouse_list) for its primary functions and restricts sql_query to schema inspection and metadata discovery, reducing the risk of arbitrary command execution.
[SAFE]: Documentation provided for SQL Server and Microsoft Fabric references standard industry libraries (mssql, @azure/identity) and official authentication methods.

data-parity