Human Approval & Applied Single-Agent Builds

Put a dry-run and approval gate in front of consequential actions.

Lessons

7 lessons

Deliverable: SQL or vectors: answer the same 20 questions over a product catalogue with CSV RAG (M19) and with the text-to-SQL agent, and record which wins for each question type.
Assessment: Answer the five Interview questions without notes, then run the lab.
Interview probe: Larkspur's status agent published "a root-cause report within 48 hours". The reviewer approved a preview that said 72 hours, and the audit log shows the approve. How did this happen, and what do you change?
Common mistakes: Generating the dry-run preview inside the review node before interrupt()
Counting sale lines in generated SQL where the question needs summed Quantity
Asking the model to merge checked parts instead of rendering a template
Trusting an evidence URL the agent's own searches never returned

How confident are you with this module?

1 means "I could not explain it yet"; 5 means "I could teach it". Rate again any time.

Your notes on this module

Saved in this browser when you leave the box.