Tutor notes. For the person running the sessions.
Session 3: tutor notes
Objective: a complete second sprint with the full ceremony set, review (their demoing, you as stakeholder), a design pass, a live-data slice, and their first real retrospective, ending with the turn: they should walk out knowing why the next session's project can't be built the way this one was.
Timings (2h)
| Time | Segment |
|---|---|
| 0:00 | Sprint review, their demo, board honesty, velocity in one sentence |
| 0:20 | README + accessibility pass |
| 0:30 | /design-shotgun → /design-html → ship it through the loop |
| 0:55 | Scrape → dashboard slice → /skillify (the heart of the session) |
| 1:35 | /retro + /learn |
| 1:50 | The turn + /context-save |
Delivery notes
- Be a real stakeholder in the review. Ask "what does it cost?", "what's next?", and one question they can't answer, feeling that is the lesson. If last session's solo loop didn't happen, review what is live without drama; the committed-vs-done gap is the velocity lesson teaching itself.
- Design shotgun paralysis: time-box the comparison board to ten minutes. Agile design is picked, not perfected; note the losers' best ideas on the backlog and move.
- Pick the scrape target in advance. Vet 2–3 candidate pages the day before: publicly accessible (or their own logins), stable structure, data they genuinely care about. A failed scrape at 1:10 kills the best part of the session. Supermarket offers and society events pages work; anything behind aggressive bot-detection doesn't. If nothing fits inside their product, a standalone dashboard page in the same repo is fine, don't force the integration.
- /setup-browser-cookies needs a consent conversation. It copies their real browser's login cookies into the agent's browser. Explain it in those words before they run it, and prefer a target page that doesn't need it at all. This is the course's first data-privacy teachable moment. Don't skip it to save five minutes.
- The /skillify beat is the finale, protect it. After it runs, open the generated skill file and let them read it: "this is what a tool is." That's this session's reveal, and it sets up everything gstack does later ("all those commands are files like this one").
- The scrape returns a mess. Fine, messy data is real data. Cleaning it with the agent is a bonus lesson, not a derailment. Time-box it.
Running the retro
You've run a hundred of these; the only note is restraint: their patterns, their words, you facilitate. It's their first: keep it to three lists (went well / harder than it should have been / surprised me) and let /learn do the banking. The deluxe version with receipts comes at the end of the course; don't spend the emotional beats in this session.
The turn: set up the next session properly
The closing pivot is load-bearing: agile just worked, so why change? Have their homework answer ready ("what couldn't start that way?"), the capstone is the answer. Three platforms, one shared database, decisions that are expensive to reverse. Don't reveal the product (Shelf Life, on their actual phone, the next session's dopamine); do plant the shape: "big, multi-platform, decided up front." If they say "can't we just sprint it?", perfect, hold the question and promise the next session answers it. That tension is the course's engine for the next fortnight.