Build a Study Helper Agent

Now pull together everything from previous lessons and finish a Study Helper Agent that reads notes, pulls out 3 facts, and only saves after a human has reviewed it.

Let's think about this for a second

This project isn't judged on how eloquently the agent talks. It's judged on: is the goal clear, is the tool access limited, does it avoid guessing, does a write action require approval, and — do you actually understand it? You can absolutely use vibe coding to write the code. But a human still needs to own the agent's behavior. Don't bolt on some heavyweight framework you don't understand.

Let's connect this to everyday life

Steps — 1) write a goal card, 2) create 3 fake note files, 3) connect list_notes and read_note, 4) generate a draft_summary, 5) preview it on screen, 6) put the save action behind a button, 7) run all three tests, 8) write down secrets/logs/limits in a single file. You can finish this with Node.js, or even on paper with fake tool results. If you're running out of time, cut the scope — even an agent that just lists files with one tool clears the lesson. But don't water down your "Done when" criteria to get there.

• Put the goal, tools, and stop rules in a file
• Run all three tests yourself
• End up with only code (or notes) you actually understand

Let's try it out together

Study Helper Agent
Goal: 3 bullets from a chosen note, Burmese, no invented facts
Tools: list_notes, read_note, draft_summary
Auto: list, read, draft
Ask first: save_draft
Never: web search, delete, email
Stop: 3 bullets + quote, or “မတွေ့”, or 5 steps
Tests: happy / missing file / refuse web search

You'll finish a Study Helper Agent you can fully explain.