This PR adds a fetch_db_schema method to the DatabaseTool class, enabling agents to retrieve a compact, human-readable schema summary for all or selected tables in a connected SQL database. This allows agents to preload schema information without requiring runtime introspection via SQL queries.

Features

• New method: fetch_db_schema(thread_context, tables=None)
  • Returns a string like:

    table "users":
      id int
      name varchar
      email varchar
    table "accounts":
      ...
    

  • If tables is provided, only those tables are included; otherwise, all tables are listed.
• Tool registration: The new method is registered in get_tools() so it is available to agents and tool discovery.
• Error handling: If a table cannot be introspected, the error is included in the output for that table.

Why?

• Allows agents to preload schema context for better SQL generation and reasoning.
• Reduces the need for repeated schema introspection queries at runtime.
• Supports workflows where schema is needed as part of agent memory or prompt context.

Tests

• Added tests/test_database_tool.py with tests for both all-tables and single-table schema fetches using a sample SQLite database.
• Tests verify that the output format matches the requested compact, readable style.

Example Usage

dbtool = DatabaseTool()
agent = Agent(..., tools=[dbtool], memories=[dbtool.fetch_db_schema("accounts", "users", "sales")])