A walkthrough of read-only data exploration with Bauplan and ClaudeCode. We install the Bauplan skills, point Cluadecode at a single table, and watch it work through four phases while building a reusable Python script with the Bauplan SDK and Polars, fixing its own errors and writing the findings to a summary.md. The result is a structured report covering table semantics, data quality, and join candidates, all without touching production.
0:00 - Intro and setup
1:37 - Phases 1 and 2: discovery and schema
3:40 - Phase 3: profiling and anomalies
5:25 - Phase 4 and final report