A version number printed
You are set. Help a neighbour, then we start Screen 2 together.
Screen 1 of 17 · Setup checkpoint
Show me a version number.
Not installed yet? That is expected. Setup happens together in Meeting 2, Route B earns the same credit, and nobody has to pay.
Your taskIf you installed early, run
claude --version. Hold up what you see. Keep any error on screen — the message is the diagnostic.Sort yourself into one of fourNobody stays stuck quietly
An error printed instead
Keep it on screen and sit with the TA during the first activity.
Sign-in refused your account
Claude Code needs a paid plan on your own account. Take Route B — same credit — and email Dr. Qu if cost is the barrier.
You have not found where to type it
Normal on day one. Raise your hand — you are not behind, you just need the terminal opened once. Same credit.
Sort · 5-minute routine
Which of A, B, C or D are you in right now?
RUN · 1SORT · 2PAIR · 2
The packet stays shut until lenses are assigned, so nobody reads the diff before they know what they are looking for. Do not debug installs from the front of the room; it strands the other twenty students. Sort into A/B/C/D, hand B, C and D to the TA, and start Screen 2 on time. Route B keeps every blocked student on the same assessment path, so nobody is behind. Setup page: week-02-claude-code-setup.html
Screen 2 of 17
Here is the course site. Try all three links.
Your taskClick each link. Say what happened. Then: what do we actually know?
CS-215 course siteReal links · each opens in a new tab
The course page · each opens in a new tab
Real links on the real course site. Come back to this tab after each one.
Discuss this first · 2 min
What do we know for certain — and what are we only guessing?
Try the links, talk it through, then reveal.
What we know. Announcements returned 404 — the web server’s code for “there is no page at this address”. Its address ends in announcements.html. The other two opened.
What we do not know. Why. Missing, misspelled, moved — nobody has looked at the files yet.
This one broken link is the case for all of Topic 2.These are real links to the real course site, so the 404 is genuine — click all three on the projector yourself before asking anything; it has to be seen, not described. Chrome shows its own “page can’t be found · HTTP ERROR 404” screen; Safari and Firefox may show a blank tab instead, so have those students read the address bar aloud. If the wifi is down: open the three addresses from your own phone hotspot on the projector, or fall back to naming the outcome for each and move on — the discussion still works. The distinction you want is between the symptom (the browser said not found) and the cause (unknown). If a student jumps to “the file was renamed,” accept it as a guess and write the word GUESS next to it — Screen 3 is built on that word.
Screen 3 of 17 · One request · Try it yourself first
Here is the request. What would an AI do with it?
Your taskTry it on a real chatbot. Then decide: can we tell if the answer is right?
The request · same broken linkTry → compare → then name what happened
The request · copy it exactly, give it nothing else
Three steps
“Find out why the Announcements link fails. Do not edit or publish anything.”
1 · Try it yourself · 3 min
2 · Compare · can we tell if it is right?
3 · What it actually did
1 · Try it — 3 minutes
Paste the request into any AI chatbot. Give it nothing else. Read the answer to your neighbour.
No device? Write what you think it would say.
We collect three answers aloud, then move on.
2 · One LLM said
“The link is probably pointing at a file that was renamed or moved. Try correcting the href to the current filename.”
An href is the address inside a link: the page or file the link opens.
Yours probably said something like this — and sounded sure.
Commit
Can we tell whether this is right?
3 · What it did
Read the words. Drew on what it had read before. Wrote the likely answer.
Nothing was checked. It never opened the page. Status: unverified.
Why: the request gives it no address. Give the same product a URL with browsing on and it becomes the next screen’s case — that is where the line actually sits.
From likely to supported, something has to look. Next screen.
Step 1 of 3
Do not require a device; the step-2 sample is the shared reference for anyone without one, and the answers students bring back will all be plausible and all unverified, which is the point. Hold the “give it nothing else” rule: if someone pastes the actual URL, a browsing-enabled model may fetch it and report the 404 — that is not wrong, it is an agent, and it previews Screen 4; name it and move on. Nothing personal goes into the chatbot; the request is the only text. Next is locked at step 2 until the class commits yes or no.
Screen 4 of 17 · Now it may look
Same request. Now the model may look.
Goal: see what changes when it can look — and learn what we call that.
Your taskWrite your moves first. Then compare with its four.
One new ability: it may read the siteYour moves → its four → the name
The request
“Find out why the Announcements link fails. Do not edit or publish anything.”
Screen 3, LLM alone: guessed “renamed or moved.” Never looked.
1 · Your moves
2 · Move 1
3 · Move 2
4 · Move 3
5 · Move 4
6 · The name
Your moves — 2 minutes
You are the model. New ability: you may read any file on the site. Still no editing, no publishing.
Write the moves you would make, in order. Three or four. Start each with a verb.
Move 1 · Read the request
Find out why the link fails. Do not edit. Do not publish.
Same as the LLM so far. The difference starts next.
Move 2 · Ask to look
“May I open the course page?”
The software around it answers: yes to reading, no to editing.
The LLM alone could never ask.
Move 3 · Look — see a fact
The link points to announcements.html.
That file does not exist.
course-announcements.html does.
A guess has become something seen.
Move 4 · Propose, then stop
“Point the link at course-announcements.html. Here is what I saw.”
Then it stops and waits for a person. It could have edited. It was not allowed to.
The name
A model that can look, whose next move depends on what it saw, and that can decide to look again, is an agent.
The LLM did Move 1 and guessed. The agent did Moves 2, 3, 4 — so its answer is supported.
Short, bounded, stopped by itself. That is a normal agent.
Step 1 of 6
Hold step 1 until most pairs have three verbs down. Most write "open the page" as move 1 — that is Move 2 here; the model must be allowed to look first, and that gap is the lesson. Move 4: if someone says "just fix it," park it — that is Screen 13. The removed detail lives here, not on the slide: Move 3 replaces the Screen 3 guess with observation, and whatever follows happens because of what came back; Move 4's stop matters because without it a live page is edited by something nobody checked; "agent" here does not mean long-running, remembering, or free to act. Leave one line on the board: agent = can look; next move depends on what it saw.
Screen 5 of 17
A classmate texts you. Text back.
The ruleReply on paper, two minutes each. You may not look anything up — no phone, no laptop. Just what is in your head.
Two texts · reply to eachThen we compare
1 · Text one · reply
2 · Text two · reply
3 · Compare
Text one — 2 minutes
my browser says 404 — what does that even mean?
Write your reply. One sentence is enough.
Text two — 2 minutes
is the Schedule link on our course page working right now?
Write your reply. Same rule: no looking.
Write something even if it feels wrong.
Compare
Text one: the answer was in your head. Everyone could reply.
Text two: the answer is in the world. Nobody could — and knowing more would not have helped. Only looking helps.
An LLM sits exactly where you sat: fluent on the first, stuck on the second — and it still replies confidently.
An agent is a model that has been given a way to look.
Discuss · 1 min
Who wrote a confident reply to text two anyway? Read it out.
Step 1 of 3
Hold the no-looking rule; someone will reach for a phone on text two, and that instinct is exactly right — say so, they have just invented the tool call. Collect two or three replies to text two aloud before advancing: some hedge, some commit, and the confident ones make the Screen 3 point better than the slide does. Do not let the compare step appear until both texts have been answered.
Screen 6 of 17
Write the instructions once, for a link you have not seen.
Your taskIn pairs: the instructions you would leave for the agent. Working alone? Write them, then reread them as if you were the agent and mark anything you would have to guess. Then see what you built.
Build it, then we name it
The situation
Rules for what you write
Three steps
Different link, different page, different filename. No second chance to explain.
No filenames
Three to five steps, each a verb
End by handing to a person
1 · Write it (3 min, in pairs)
2 · What you just built
3 · Why it is not a tool
Work through it
1 · Write it — 3 minutes
On paper, in pairs. Then compare with the next pair — where the order differs, say why.
Do not advance until every pair has something.
2 · What you built
Most pairs write:
1 Find what the link points at.
2 Check that target exists.
3 Propose the smallest fix.
4 Show evidence, stop for a person.
That is a skill. A procedure written once, for a link nobody has seen yet.
3 · Why it is not a tool
“Check that target exists” does nothing by itself. Something has to go and look — that is a tool.
Tool: one action, one result.
Skill: your steps — which actions, what order, when to stop.
Swap the tool tomorrow. Your steps still stand.
Step 1 of 3
Debrief prompts: Discuss · 2 min Whose version still works if the agent has different tools? That is the test. — Hold step 1 for the full three minutes even if the room goes quiet — the naming on step 2 only lands if they have committed something first. Do not show step 2 to a pair who has not written anything. Resist naming MCP or plugins; both are Topic 5. If a pair writes "fix the link" as one step, ask them what the agent should look at first, and they will discover the decomposition themselves.
Screen 7 of 17
The agent gets one look at the site. Which look?
Your taskThe question is still: why does the Announcements link fail? Pick the single most useful thing to look at — A, B or C — and commit as a class. Working alone? Pick one and write why the other two are worse.
One look only
Choose one
Commit a letter first.
A gets you a filename — announcements.html — but not whether it exists. Look spent; question still open.
B cannot even be asked yet. Which file? You do not know until A comes back.
C cannot be asked until B says the file is missing.
It takes three looks — and you could not plan them in advance. Each one is chosen by what the last one returned.
That is a loop. Next screen: tell the agent how to run one.
Most rooms pick A, which is right and still not enough — that is the point. Do not let anyone shortcut to “check everything at once”; ask them to write the three requests without knowing the filename and the dependency becomes obvious. If someone says two looks, ask what happens if course-announcements.html is missing too. The felt need for a loop is the whole purpose of these four minutes.
Screen 8 of 17
Same question. Now it may look as often as it needs. Tell it how.
Your taskYou just saw the looks cannot be planned in advance — so the agent needs a rule it can follow on its own. Write that rule: what to look at, what to do with each answer, and when to stop. Three lines, two minutes.
Three lines to complete
1Look at .
2If → stop and report. Otherwise → look at .
3Repeat until .
Write it — 2 minutes
The question: why does the Announcements link fail? The limit: no editing.
Line 3 is the hard one. Most pairs leave it blank.
The course version
1 Look at what the link points to.
2 If that file exists → stop, report “not the link.” Otherwise → look for the right file.
3 Repeat until you can name the cause or you have run out of things to look at.
What you just wrote is a loop: a look-and-decide that repeats until a stop rule says stop. Without line 3 it either never ends, or it ends by doing something nobody asked for.
Now the real one
This is the loop inside the tool you installed, mapped from its source — eleven steps, animated.
Find your three lines in their eleven. Which of the eleven is your line 3, the stop rule?
Open the real loop ↗ccunpacked.dev · opens in a new tab
Next screen: act it out, one role each.
Write first.
Real loop: project ccunpacked.dev and press play on “The Agent Loop” at 0.5x. Their step 7 “Tools?” is the decide, step 8 “Loop” is the repeat, and step 11 “Await” — waiting for the human — is the stop rule most pairs left blank. Two minutes; it is a look, not a lesson. Collect two pairs’ line 3 aloud before revealing; the blanks and the vague ones (“until it works”) are the lesson. Push on what “works” would mean to the agent — it has to be something it can check. If a pair writes a stop rule that involves editing, that is the Screen 13 conversation arriving early; note it and move on. Do not reveal until most pairs have three lines.
Screen 9 of 17
Act out the loop—one role, one move at a time.
Your taskGroups of four, one role each. Working alone? Take all four moves in order and write what each one can and cannot know. Speak only when your role is lit. Case: diagnose the broken Announcements link, read-only.
Paper-agent simulation
Diagram structure: the goal keeper gives a bounded request to the model role. The model proposes a read. The host and tool role checks permission and returns an observation. The recorder carries the observation back to the model. When the model proposes a change, the human reviewer enforces the stop boundary. The four rounds are frame, observe, revise, and stop.
Round 1 · FrameVisual state: Goal keeper is highlighted; move 1 of 7 is revealed.
Goal keeper reads: “Diagnose why the Announcements link fails. You may inspect course files. Do not edit or publish. Stop with evidence and a proposed next step.”
Move 1 of 7
Assign one role to each person
1 · Goal keeperHolds the outcome and the limits.
2 · ModelProposes the next step. Never invents a result.
3 · Host + toolChecks permission, then returns evidence.
4 · Reviewer + recorderTracks the trace. Decides if authority must change.
Watch · 1:06 · after the role-play
Listen for one thing: where does it put the line? Same place you just did?
Open on YouTube ↗Contacts YouTube. No course text is sent by this deck; YouTube still receives browser metadata.
If playback fails, say it yourself: an LLM answers from what it already knows; an agent uses an LLM and tools to look and act — exactly the line they drew on Screens 4 and 5. Clip: play once after the debrief, not before the role-play; the room should draw the line themselves first. Debrief prompts: At which move did this become more than one answer? Which move needed a human? — Group-size fallback: three people can merge goal keeper and reviewer; one person can narrate every role. Do not reward theatrical performance. Listen for the dependency: the observation changes the next proposal, and the proposed state change triggers the human boundary.
Screen 10 of 17
Here is the instruction an agent was given. What could go wrong?
Your taskYou are the agent. Read it. Name one thing you could do under this instruction that nobody asked for. Agree it with a partner, then we look at why. Working alone? Write it, then list what an agent could still get wrong with that brief.
The instruction
“Fix the Announcements link.”
1 · Name one thing
2 · Why it is allowed
3 · The fix
Name one thing — 1 minute
Something this instruction allows that the person who wrote it would not want.
Be concrete: an action, not a worry.
Three words. Three gaps.
Fix — how? Edit the page? Rename a file? Publish? All of them count as fixing.
Based on what? Nothing says look first.
Done when? Nothing says stop. So it may not.
Same job, nothing left open
Outcome — report why the link fails; propose one fix.
Evidence — from course files you may read, not from what usually happens.
Limits — read only; no edit, rename, publish.
Stop rule — return evidence and proposal, then stop for a person.
Those four headings are an Agent Brief. Next screen, you write one.
Step 1 of 3
Collect three answers aloud before advancing; the ones you want are “publish” and “edit some other file”. If the room is quiet, ask what “fix” would mean to a contractor you hired and never spoke to again. Step 3 is the bridge, not the lesson — do not linger; the builder on the next screen is where they earn the four headings. Four minutes.
Screen 11 of 17
NEW PRACTICE JOBOne-page Topic 2 study guide. Read the approved public source; return a draft. No file edits or publication.
Give the agent a job. Check what it makes.
Your taskIn pairs, make a study guide classmates could trust. One directs; one checks. Swap for the revision. Solo? Do both.
Open the study-guide exercise ↗
- Write · 3 minOutcome, evidence, limits, stop.
- Run · 5 minNew conversation. Send your brief.
- Check + revise · 7 minVerify two claims. Request one specific change.
No account? Use the exercise’s simulated draft.
Next: what does an agent’s report actually prove?
Run the linked worksheet: exact instructor sources, four-part builder, live/supplied-context/simulated routes, and print-preview check are there. Students keep their own work; no new graded submission or chat-log collection. This replaces the former 9-minute brief with a 15-minute lab, so the full Meeting 1 deck is now 102 minutes. In an 80-minute class, reach this at about minute 58, finish at 73, use the final 7 for debrief, and carry Screens 12–17 to later class time; do not rush the run or silently drop the review. If a run stalls, move to the simulated excerpt. Ask: which instruction changed the result, and which source check supports that conclusion?
Screen 12 of 17
BACK TO LINK CASEInspect approved course files, report evidence, and propose one fix; do not edit or publish.
The agent reported back. Accept it and move on?
Your taskAccept it and move on — yes or no? Commit first.
The final report
What the agent returned
“Fixed the Announcements link and verified it works. ✓”
Commit · 1 min
Accept it and move on — yes or no?
Commit first.
Seven words, two claims: something was fixed, something was verified.
Six things happened to make that sentence. You saw one.
Two are problems: it edited a page it was told not to touch, and “verified” has nothing behind it.
Neither shows in the report. Both show in the steps.
That record is a trace. Next screen, you find the two problems.Take a show of hands on accept/reject before revealing, and note the split on the board — most rooms accept it, which is the point. Do not say which entries are wrong; the next screen is the hunt. If a student asks how you would ever know, that is the right question and the answer is the trace.
Screen 13 of 17
LINK CASEInspect approved course files, report evidence, and propose one fix; do not edit or publish.
Six entries. Two of them are the problem. Find them.
Your taskTag the two entries that broke the brief, then reveal. The report hid both; the record cannot.
The agent’s record
Tag two entries first.
Entries 5 and 6.
Entry 5 changed the page; the brief allowed a proposal, not an edit. It could write, it was not allowed to — capability is not authority.
Entry 6 claims “verified” with no check in the record.
Entry 3 is the only real evidence: it names the broken link.
Next: you write the permission that would have made that edit legal.Let them hunt for two full minutes before revealing; the room usually finds Entry 5 quickly and argues about Entry 6, which is the one worth the time — an unsupported claim reads exactly like a supported one. Insist on trace language afterwards: “Entry 3 is evidence,” “Entry 4 is a proposal,” “Entry 5 exceeds authority.” Screen 14 turns that into a written permission, so do not settle the permission question here.
Screen 14 of 17
LINK CASEInspect approved course files, report evidence, and propose one fix; do not edit or publish.
Entry 5 made the edit anyway. Write the permission that would have made it legal.
Your taskThree lines, two minutes. A permission a person can actually grant names all three — vague trust is not a permission.
Write the permission
Write your permission first.
Two defensible answers.
Grant it narrowly: change that one href in a working copy, run the link check, stop before publishing.
Keep it read-only: the instructor owns the page — return the evidence and the proposed change, a person applies it.
Both name a scope, a check and a stop. Neither says “it seemed right.”
Next: a different agent, a job you ran yourself.Two minutes writing, then take three permissions from the room and test each one out loud: what exactly may it change, what must it show, where does it stop? Most first drafts fail on the stop — students grant the edit and forget to forbid the publish. If the room is confident, run it as a split: half argue grant-narrowly, half argue read-only, then each side gives the other’s strongest reason before you reveal. Keep it operational — conditions, not confidence. Later ethics work takes up power and accountability.
Screen 15 of 17
BACK TO YOUR STUDY-GUIDE JOBYour Screen 11 brief: read the approved source, return a draft. No edits, no publishing.
Your study-guide agent does three more things. Which did your brief allow?
Your taskThree things happen. Which did the request allow? Commit, then reveal.
Your brief: read the approved source, return a draft
What the agent did
1 · Read a classmate’s shared notes
The folder was visible, so it used them.
2 · Stated a deadline nobody gave
“Quiz Friday.” No page says so.
3 · Emailed the draft to the class
It had a send tool and thought it helpful.
How many of the three did the brief allow?
Decide all three first.
None was allowed — and no single rule stops all three.
1 input: it could reach what it should not.
2 evidence: a claim no source supported.
3 action: it reached other people without asking.
“Be careful” prevents none. Each needs its own boundary, set before it runs.
Next screen: you write the four boundaries your own agent needed.Run all three as a quick show of hands before revealing; groups usually catch 3, sometimes 1, and almost never 2. Case 2 is the one worth dwelling on — an invented deadline looks exactly like a real one, which is why an evidence rule has to be structural rather than a matter of care. Do not let "just tell it to be safe" stand; ask which of the three it would have stopped.
Screen 16 of 17
SAME STUDY-GUIDE JOBThe agent you ran on Screen 11. Use your own source.
Build the boundaries that would have prevented each failure.
Your taskOne layer per group, for the agent you ran. Write a boundary software or a person could enforce.
Minimum safe system
System diagram: the study-guide agent you ran sits at the center. Four safeguard layers are added in order. Context limits inputs to approved course pages and excludes sensitive data. Tool permissions allow reads and local drafts but no shared edits or external messages. Evidence checks require citations and verification. Human hand-off occurs before edit, send, or publish actions.
Layer 0 of 4 · Design first
Group sentence starters
1 · Context + privacyIt may use ___, but it must not receive ___.
2 · Tools + permissionsIt may perform ___, but it cannot ___.
3 · Evidence + checksBefore claiming ___, it must show or run ___.
4 · Stop + hand-offIt must stop when ___ and ask ___.
These four boundaries are the brief you would give it next time — where Screen 11 started.
Point at the loop explicitly: they wrote a brief on Screen 11, watched a record expose what a report hid, wrote a permission, and have now written the brief they should have started with. Debrief prompts: Which safeguard lives in the prompt, which in software, which still needs a human? — Do not accept “be safe” or “be accurate.” Require enforceable controls: approved source list, read-only tool scope, an observable check, or a named human checkpoint.
Screen 17 of 17
TWO CASES, ONE MODELAnnouncements case: trace and authority. Study-agent case: context, evidence, and action boundaries.
Explain today in three sentences.
Your taskFinish the three sentence stems. Use the Announcements case for trace and authority, and the study-agent case for safety. Then compare with the course anchors.
Exit ticket
Diagram before reveal: an unlabeled system shows a central model, reusable guidance, tool operation, returned observation, host gate, and human boundary. Students first explain the relationships in their own words.
Your explanation first
Three-sentence exit ticket
Sentence 1 · Model
An LLM contributes ___, but that does not prove ___.
Sentence 2 · Workflow
A tool differs from a skill because ___; the agent uses them by ___.
Sentence 3 · Authority
The system must stop and ask a human when ___.
Model: generates a proposal from context; fluency is not verification.
Workflow: a skill guides reusable steps, while a tool performs an operation and returns an observation.
Authority: host controls and accountable humans bound consequential action.
Debrief prompts: One boundary you would never leave implicit in any agent you use? — Collect two contrasting boundaries aloud. Close by previewing that Meeting 2 will apply this model to a concrete development workflow rather than add more vocabulary.
Meeting 2 · Screen 1 of 9
YOUR STUDY GUIDEThe guide you made in Meeting 1. Today you save it, checkpoint it, and put it where you can reopen it.
Where will your study guide be next week?
Your taskPick where it actually is right now. Commit, then reveal.
Today: setup together, then one shared demonstration. Use the workshop for the 25-minute setup and 17-minute demo; individual practice comes later.
Open Git setup and demo →Your guide, right now
Commit an answer first.
Save a file you can reopen.
For this exercise, give the next agent session the saved file and next-step note instead of relying on chat memory.
Today you put the guide in a file, record a checkpoint, and send it somewhere you can reopen it.
First: where does that file live?Ask for a show of hands on all three. Some rooms genuinely believe the third; that belief is what the next eleven screens replace. If a student did not finish a guide in Meeting 1, hand them the supplied sample now so nobody is stuck at screen 1.
Meeting 2 · Screen 2 of 9
YOUR STUDY GUIDEThe guide you made in Meeting 1. Today you save it, checkpoint it, and put it where you can reopen it.
Which move first puts your work somewhere other than this computer?
Your taskFour places, three moves. Pick the move that leaves this computer, then reveal.
Four places, three moves
1 · The filestudy-guide.txt and next-steps.txt
2 · The folder
where those files sit
3 · The history
your checkpoints, on this computer
4 · GitHub
a copy you can reopen anywhere
Commit · 1 min
save writes the file · commit records a checkpoint · push sends
Commit an answer first.
Push — and only push.
Save and commit both happen in places 1–3, on this computer. You can commit all day and place 4 still shows nothing.
Git performs all three moves. GitHub is the site holding place 4.
Watch for these four places in the demonstration.Six minutes covers Screens 1 and 2 together. Point at each of the four places as you read them, and point at them again during the demonstration — that repetition is what makes the words stick, not this screen on its own. The room usually says commit; do not correct it before the reveal, because the gap between committing and pushing is the surprise the whole meeting is built on. Saving and committing are not re-taught here: the checkpoint idea returns on the commit screen, and the local-versus-online question returns when students predict which version is online.
Meeting 2 · Screen 3 of 9
YOUR STUDY GUIDEThe guide you made in Meeting 1. Today you save it, checkpoint it, and put it where you can reopen it.
Git works. You never made the GitHub account. What becomes impossible?
Your taskOne answer. Two setup steps, two different jobs. Decide which job needs the website.
Two setup steps, two different jobs
In a few minutes you do two separate things: install a program called Git, and create a GitHub account.
Neither one installs the other. You check them separately.
Commit · 1 min
Which one becomes impossible?
Commit an answer first.
Opening it somewhere else.
Git is a program on your computer. It records checkpoints and reopens old versions — no account, no internet needed.
GitHub is a website holding a copy, so another computer or another person can reach your work.
Git gives you history. GitHub gives you reach. Tonight you set up both.
Next: open the practice project.Three minutes, immediately before the setup window, because students are about to be asked to install one thing and sign up for another and will otherwise treat them as one step. The wrong answers are the useful ones: a student who thinks checkpoints need GitHub will also think a failed signup blocks the whole lesson, and it does not — Git alone gets them through the checkpoint work. Say plainly that a Claude or agent account is a third, separate thing. This screen replaces explaining Git and GitHub during the account window, so keep that window for actual setup.
Meeting 2 · Screen 4 of 9
YOUR STUDY GUIDEThe guide you made in Meeting 1. Today you save it, checkpoint it, and put it where you can reopen it.
Open the workshop. We set up together, then run one save, commit and push.
Your taskFollow the workshop from here. You direct and check; the instructor operates the prepared project.
The actual setup steps, repository access, and prompts are here. The private repository link is for the instructor; students watch the projected result.
Open Git setup and demo →A doorway, not a lesson — about one minute. The workshop runs 25 minutes of guided setup and a 17-minute shared save/commit/push demonstration, and it owns those steps now; the deck stopped carrying a second copy of them. Come back to the deck at the push screen for the evidence check. Students need no access to the instructor’s project. The folder-safety check that used to live on this screen moves to the individual-practice class, where students actually make their own copy.
Meeting 2 · Screen 5 of 9
YOUR STUDY GUIDEThe guide you made in Meeting 1. Today you save it, checkpoint it, and put it where you can reopen it.
The workshop just made a checkpoint. Put it where you can reopen it.
Your taskBefore you approve the push, pick what you must see first.
Approve the destination
Say this: “Show the exact GitHub destination and the commits you intend to send. Wait for my approval. Then push, report success or the real error, and give me the browser link.”
Commit an answer first.
The destination and the contents.
A success message is the agent’s claim. Meeting 1 taught you what those are worth without evidence.
So you approve the destination, push, then open GitHub in the browser and find both files and your checkpoint. The page is the evidence.
Next: change one sentence — without losing this version.This is the payoff: the browser page, not the chat, is what proves it. Private repository — pushing does not publish a website. If a push fails, read the actual error aloud; a real error is better teaching than a smooth demo.
Meeting 2 · Screen 6 of 9
YOUR STUDY GUIDEThe guide you made in Meeting 1. Today you save it, checkpoint it, and put it where you can reopen it.
Improve one sentence without losing the earlier version.
Your taskName one sentence in your guide to fix. Write it before and after.
One sentence, corrected
Write your answer first.
Two lines, side by side: that comparison is called a diff.
Read it before approving. It should show your sentence changed and nothing else.
Then a second checkpoint, with its own message, and an updated next-steps.txt. The first version is still there.
Insist on one sentence. Students who rewrite half the guide cannot read their own diff. Name the word diff only after they have looked at one — they have already inspected a trace, so the move is familiar.
Meeting 2 · Screen 7 of 9
YOUR STUDY GUIDEThe guide you made in Meeting 1. Today you save it, checkpoint it, and put it where you can reopen it.
You committed the correction. Which version is on GitHub right now?
Your taskPredict before anyone pushes. Commit, then reveal.
Local history vs online copy
Commit an answer first.
Still the first one.
A commit changes your computer. Only a push changes GitHub. This is the same line you drew on Screen 4.
Push again, refresh, and find both checkpoints. Open the earlier one and read the old sentence — without deleting the new one.
Next: can a brand-new session pick this up?The room usually says corrected. The gap between committing and pushing is the single most common beginner surprise, which is why the cycle runs twice. Show the earlier version read-only; do not demonstrate reverting today.
Meeting 2 · Screen 8 of 9
YOUR STUDY GUIDEThe guide you made in Meeting 1. Today you save it, checkpoint it, and put it where you can reopen it.
Start a new agent session. Can it pick up your work?
Your taskNew session, no memory of today. Write what lets it continue.
Prompt 5 · Pick it up again
Say this: “Show the new folder for a copy of this repository. Check it does not exist and wait for me. Then copy it there, read both files, and tell me what is next. Change nothing.”
Write your answer first.
next-steps.txt — because memory is not a file.
A fresh session has none of your conversation. What it can read is what you saved and sent.
Copy the project into a new folder, let it read the guide and the note, and have it tell you the next task — then stop. No edits during retrieval.
Next: say the whole workflow in your own words.This is the screen that justifies the whole meeting. A student who watches a brand-new session read their own note understands why saving beats remembering. Confirm the latest correction arrived — that is the proof the second push worked.
Meeting 2 · Screen 9 of 9
YOUR STUDY GUIDEThe guide you made in Meeting 1. Today you save it, checkpoint it, and put it where you can reopen it.
Explain save, commit and push using your own guide.
Your taskThree sentences to a partner, then fill the three-line record. Ungraded, nothing to submit.
Your three-line record
Write your answer first.
Save updates the file.
Commit records a checkpoint on your computer.
Push sends it to GitHub.
You have two checkpoints, an online copy you can reopen anywhere, and a note a stranger could act on.
Bookmark the repository. Bring the note to the next class.Ask three students to point at where their work lives and say the three words in order. If someone says the agent saved it, go back to Screen 1 for thirty seconds. Keep the note — the next meeting starts from it.