Build Snowflake data objects with Claude Code and PRPs

Follow a four phase product requirements prompt workflow as Claude Code builds two Snowflake data objects.

Who it's for: Data analysts and engineers who want to use AI before their data is clean.

The problem: A data object takes many iterations to settle the grain, sources and edge cases, and AI lacks the context in your head.

A product requirements prompt (PRP) front loads that research. You fill out INITIAL.md, generate-data-object-prp writes the PRP, prp-data-object-execute builds and checks the object in development, and a human promotes it. Two examples use Snowflake TPC-H sample data.

Takeaways:

  • Your data does not have to be in a final place to use AI.
  • Start each PRP review with its confidence assessment.
  • Keep production promotion in human hands.
  • Adapt the framework for dbt or ticket work.

Resources