Skill Development Guide

How to extend or customize this skill for your own use cases.

Skill Structure

Agent Skills follow a convention:

skill-directory/
├── SKILL.md          # Required — the skill definition the agent reads
├── assets/           # Optional — templates, configs, reference files
├── references/       # Optional — domain docs, schema references
├── scripts/          # Optional — automation scripts
└── examples/         # Optional — working examples for the agent to reference

The SKILL.md file is the core. The agent reads it to understand:

  • When to activate the skill (trigger phrases)

  • What to generate (schemas, templates, examples)

  • How to generate it (workflow steps, constraints)

Customizing for Your Domain

1. Update the Examples

Replace the energy-domain examples with your own:

examples/
├── your-bronze/
│   ├── dataflowspec/your_bronze_main.json
│   ├── schemas/your_table_schema.json
│   └── expectations/your_table_dqe.json
├── your-silver/
│   └── dataflowspec/your_silver_main.json
└── your-gold/
    └── dataflowspec/your_gold_main.json

2. Update the References

Edit references/energy-domain-mapping.md to describe your tables and their relationships:

## Your Domain Tables

| Table | Description | Key | CDC Pattern |
|-------|------------|-----|-------------|
| raw_orders | Customer orders | order_id | SCD1 |
| raw_products | Product catalog | product_id | SCD2 |

3. Add Domain-Specific Expectations

Create expectation templates for your data quality rules:

{
    "expect": [
        {"name": "valid_order_id", "constraint": "order_id IS NOT NULL", "tag": "completeness"},
        {"name": "positive_total", "constraint": "order_total > 0", "tag": "range"}
    ],
    "expect_or_drop": [
        {"name": "valid_status", "constraint": "status IN ('pending','shipped','delivered')", "tag": "validity"}
    ]
}

4. Update SKILL.md Triggers

Modify the “When to Use” section to include your domain terminology:

## When to Use

- User says "generate a Data Flow Spec for orders"
- User needs CDC ingestion for the e-commerce data platform
- User asks for order-to-shipment pipeline using Data Flow Specs

Adding New Patterns

To add support for a new pattern:

  1. Create an example Data Flow Spec in assets/dataflowspec-templates/

  2. Add the pattern to the “Patterns” section in SKILL.md

  3. Add a corresponding example in examples/

  4. Update references/patterns-guide.md

Testing Your Skill

  1. Install the skill in your assistant’s skills directory (see Getting Started)

  2. Open a new chat or notebook session

  3. Test with progressively complex prompts:

    • Start with: “What patterns does the dataflow-spec-builder support?”

    • Then: “Generate a simple bronze Data Flow Spec for my_table”

    • Then: “Create a complete medallion pipeline with templates and DQ”

Tips for Effective Skills

  • Be specific in SKILL.md — Include complete JSON schemas, not just descriptions

  • Include working examples — Agents learn from examples in the skill directory

  • Use unique trigger phrases — Avoid terms that overlap with built-in Databricks features

  • Add a “When NOT to Use” section — Helps the agent avoid false positives

  • Keep the skill focused — One skill per framework/pattern, not a catch-all