Context engineering in Claude Code for a scheduled Notion report

Follow four layers of context engineering that end in a Claude Code routine writing a Notion report every Monday.

Who it's for: Data professionals who want AI to work as a system instead of a chatbot.

The problem: A fresh Claude Code session arrives without context from your other sessions, so you re-explain everything. The same prompt, data and model give different output depending on that context.

The video builds a YouTube channel analyzer across four layers: a CLAUDE.md file, a repo structure with a business rules file, BigQuery and Notion tools, and GSD skills plus a routine that runs every Monday at 9am.

Takeaways:

  • Keep your CLAUDE.md concise.
  • Put stable business rules in their own file.
  • Use routines for reviewable, low-risk reports.
  • Start with repetitive, reviewable work that has the data AI needs.

Resources