Databricks Assistant Skills

Agent Skills for Databricks Assistant to query Genie spaces and Multi-Agent Systems.

Setup

1. Import into your Databricks workspace

   Copy the skills folder into your Databricks workspace at:

   /Workspace/Users/<your-email>/.assistant/skills/
   

   Create the .assistant folder if it doesn't exist.

2. Configure placeholders

   Replace all <INSERT_...> placeholders in the files with your actual values:

   Placeholder	Description	Example
   <INSERT_DATABRICKS_HOST>	Your Databricks workspace URL	https://my-workspace.cloud.databricks.com
   <INSERT_SECRET_SCOPE>	Databricks secret scope name	my_scope
   <INSERT_TOKEN_SECRET_KEY>	Secret key for personal access token	my_token
   <INSERT_CLIENT_ID_SECRET_KEY>	Secret key for service principal client ID	sp_client_id
   <INSERT_CLIENT_SECRET_SECRET_KEY>	Secret key for service principal client secret	sp_client_secret
   <INSERT_CATALOG>	Unity Catalog name for UC functions	my_catalog
   <INSERT_SCHEMA>	Schema name for UC functions	my_schema
   <INSERT_GENIE_SPACE_ID>	Genie space ID (from URL)	01f0f9f0b9c41c74a5adfb46bfc836dd
   <INSERT_MAS_ENDPOINT_NAME>	Model serving endpoint name for MAS	my-mas-endpoint

3. Create secrets (if not already done)

   # Create secret scope
   databricks secrets create-scope <INSERT_SECRET_SCOPE>
   
   # Add personal access token
   databricks secrets put-secret <INSERT_SECRET_SCOPE> <INSERT_TOKEN_SECRET_KEY>
   
   # For Genie UC functions (service principal)
   databricks secrets put-secret <INSERT_SECRET_SCOPE> <INSERT_CLIENT_ID_SECRET_KEY>
   databricks secrets put-secret <INSERT_SECRET_SCOPE> <INSERT_CLIENT_SECRET_SECRET_KEY>

4. Register UC functions (for SQL Editor usage)

   Run the SQL in genie_query_uc_function.sql and mas_query_uc_function.sql to create the Unity Catalog functions.

Folder Structure

skills/
├── genie-skill/
│   ├── SKILL.md                    # Skill configuration for Genie queries
│   ├── genie_query_uc_function.sql # UC functions for SQL Editor
│   └── scripts/
│       └── genie_query.py          # Python script for notebooks
└── mas-skill/
    ├── SKILL.md                    # Skill configuration for MAS queries
    ├── mas_query_uc_function.sql   # UC function for SQL Editor
    └── scripts/
        └── mas_query.py            # Python script for notebooks

Skills

genie-skill

Query Databricks Genie spaces using natural language. Supports:

• Starting conversations
• Asking follow-up questions
• Deleting conversations when done

mas-skill

Query a Multi-Agent System (MAS) endpoint that coordinates multiple specialized agents for cross-domain questions.

Usage Options

Option 1: Python Scripts (Notebooks)

Use the Python scripts directly in Databricks notebooks:

from scripts.genie_query import start_conversation, ask_followup, delete_conversation

response = start_conversation("<SPACE_ID>", "What is our churn rate?")

Option 2: UC Functions (SQL Editor)

For SQL Editor, use the registered Unity Catalog functions:

SELECT my_catalog.my_schema.genie_start_conversation(
    '<SPACE_ID>',
    'What is our churn rate?',
    secret('my_scope', 'sp_client_id'),
    secret('my_scope', 'sp_client_secret')
) as result;

Permissions

Ensure your service principal or user has:

• CAN USE permission on Genie spaces
• CAN USE permission on SQL warehouses
• SELECT permission on Unity Catalog tables used by Genie
• CAN QUERY permission on model serving endpoints (for MAS)

License

MIT