data-model

What This Skill Does

The data-model skill provides a professional framework for designing robust analytical schemas. It shifts the focus from 'star schema by default' to intentional, requirement-driven data modeling. The skill guides you through a six-stage methodology covering business question alignment, grain definition, conformed dimensions, measure classification, Slowly Changing Dimension (SCD) strategies, and physical warehouse optimization. By enforcing explicit grain definitions and strict key management, the skill helps prevent common BI pitfalls like fan traps, chasm traps, and inconsistent metric reporting. It ensures that your data warehouse layers remain scalable, readable, and performant as your organization's data footprint grows.

Installation

To integrate this skill into your OpenClaw environment, execute the following command in your terminal:

clawhub install openclaw/skills/skills/clawkk/data-model

Use Cases

• Architecting New Warehouses: Designing clean dimensional models from raw data sources in dbt or BigQuery environments.
• BI Refactoring: Identifying why reports show incorrect duplicate counts or why joins are failing due to fan traps.
• Schema Evolution: Determining how to handle changing source data attributes, whether through SCD Type 2 history tracking or SCD Type 1 overwrites.
• Performance Tuning: Applying partitioning, clustering, and late-arriving fact strategies to large-scale analytical tables.

Example Prompts

1. 'I am designing a retail analytics warehouse. Can we walk through the six-stage data model process to define the grain for our order line items and sales fact table?'
2. 'My BI dashboard is reporting inflated revenue figures when I join customers to orders. Can you help me identify if this is a fan trap and how to restructure the model?'
3. 'We have a requirement to track historical status changes for our project entities. Should I use an SCD Type 2 table or a snapshot fact table for this analysis?'

Tips & Limitations

• The Grain is King: Always resolve the 'one row per what' question before writing any code. If you cannot define the grain in a single sentence, the model is not ready.
• Avoid Fact Sprawl: Do not add every attribute to your fact tables; stick to measures and keys. Use degenerate dimensions sparingly.
• Tooling Matters: When working with the agent, be sure to specify your warehouse technology (e.g., dbt, BigQuery, Snowflake, Redshift) so the agent can provide platform-specific syntax for clustering or window functions.
• Limitations: This skill is focused on analytical modeling (OLAP). It is not designed for transactional schema (OLTP) design or application database normalization tasks.