Claude Code Workshop — Canonical Curriculum

WEEK 1Understanding AI + Claude Code Essentials

📖Before classASYNC

~55 min

Readings

1

How AI works Probabilistic AI: outputs vary, iteration is the workflow. Context is everything: "context is not storage." AI tool landscape: chat interfaces vs agents.

30 min

2

Why Claude Code is different — the harness Claude can read files, write files, run commands, check results, and loop. Read before class so the demo lands as recognition, not introduction.

10 min

🧠 Quiz — answer before class (10 min)

1.You ask Claude the same question twice and get different answers. Is this a bug?

2.Why does adding more context usually improve Claude's output?

3.What is the main practical difference between Claude.ai and Claude Code?

4.True or false: Claude Code can create and edit files on your computer.

5.What does "context window" mean in plain terms?

Instructions — how to prepare for class

Install Claude Code CLI: npm install -g @anthropic-ai/claude-code and confirm it launches

Have a real work folder open in VS Code during class (not a tutorial — something from your actual job)

Bring your pre-L1 planning exercise output (the task plan you asked Claude to generate)

🎥In classLIVE

57 min content + ~15 min FAQ

#	Subtopic	Type	Time	What happens
1	Recap of reading	—	5 min	"You read that Claude Code has a harness. Today you watch it run. You read about the approval gate — today you use it."
2	The agentic loop	Demo	15 min	Instructor runs Claude Code live: two sequences, a file gets created on screen. Participants have read about the harness — now they feel it run for the first time.
3	The approval gate	Demo	12 min	Instructor generates a plan, reads it aloud, identifies what's wrong, redirects it. Full cycle: plan appears → read → redirect → corrected plan. Participants observe.
4	System prompts / CLAUDE.md concept	Concept	5 min	Why persistent instructions matter. One file that loads every session. Primes the after-class exercise.
5	Tier classification	Demo	8 min	Instructor classifies 5 real tasks live, explaining the reasoning out loud. Participants observe — they do it themselves in the after-class exercise.
6	Validate before fan-out	Demo	7 min	Instructor shows: plan approved on 10 files without canary → disaster. Then with canary → caught at step 1.
7	Session discipline + exercise preview	Demo	5 min	Instructor creates `todo.md`, adds one task, saves state. Then briefly opens the after-class exercise doc and shows what "done" means.
—	FAQ	—	~15 min	—

✏️After class — exerciseASYNC

~40 min · share before Week 2

Exercise 1 — Write your CLAUDE.md (25 min)

Open your real work folder. Write a CLAUDE.md from scratch — no template.

It must contain:

A WHY section: 2-3 sentences on what this folder/project is for
At least 3 constraints: what Claude must or must not do (e.g., "Never delete files without asking")
A HOW section: how you want Claude to work with you (tone, format, what to check)
Under 60 lines total. No style preferences.

Exercise 2 — Tier classification (15 min)

Take 10 real tasks from your actual work. Classify each as T1 / T2 / T3. Write one sentence explaining your reasoning for each.

WEEK 2Skills, Rules & MCPs

📖Before classASYNC

~50 min

Readings

1

Rules files concept What rules files are, why they're separate from CLAUDE.md, what paths: vs globs: means and why the wrong one silently affects 20+ files.

15 min

2

Skills concept The 3-rep heuristic. The context-first pattern. Why you should never write a skill for a workflow you haven't done manually first.

15 min

3

MCPs concept What MCPs connect. The friction-twice gate: don't install one until you've hit the absence of it twice.

5 min

🧠 Quiz — answer before class (10 min)

1.What is the difference between CLAUDE.md and a rules file?

2.You've typed the same 5-step workflow explanation to Claude three times this week. What should you do?

3.True or false: you should install an MCP the first time you think it might be useful.

4.What is the risk of using globs: instead of paths: in a rules file?

5.What is the "trigger" field in a SKILL.md?

Instructions — how to prepare for class

Bring your CLAUDE.md from last week's exercise — have it open during class

Bring your tier classification list — instructor may reference it during the skills demo

Have your real work folder open in VS Code

🎥In classLIVE

47 min content + ~28 min FAQ

#	Subtopic	Type	Time	What happens
1	Recap of reading	—	5 min	"You read that rules files auto-load constraints. Today you watch one fail in the exact way the reading described. You read the 3-rep heuristic — today you see a skill built from scratch."
2	Self-audit: CLAUDE.md review	Demo	5 min	Instructor opens a fresh Claude session and reviews one participant's CLAUDE.md live. Group sees what a fresh-context review finds — and what the implementer missed.
3	Rules files	Demo	10 min	Wrong config (`globs:`) fails silently — 20 files, zero warnings. Then correct config (`paths:`). The failure mode from the reading, now visible on screen.
4	Skills	Demo	12 min	Instructor builds a skill from scratch end-to-end: names the workflow, writes SKILL.md, invokes it. Then briefly shows what the after-class skill exercise looks like.
5	Hooks — enforcement over promises	Demo	12 min	Side by side: rule as text gets ignored → hook catches the same violation automatically. First contact — no reading on hooks before class.
6	MCPs: decision gate + exercise preview	Demo	8 min	Instructor shows the friction-twice check live with real examples. Then shows the optional MCP setup exercise and what a good submission looks like.
—	FAQ	—	~28 min	—

✏️After class — exerciseASYNC

~65 min · share before Week 3

Exercise 1 — Write your first rule (20 min)

Write one constraint rule file for your real work folder.

It must:

Use paths: (not globs:)
Enforce one real constraint you actually want
Verify it auto-loads: start a new Claude Code session and confirm the rule appears

Exercise 2 — Write your first skill (30 min)

Write a skill for a workflow you already did manually in Week 1 or 2.

It must:

Have a clear trigger phrase (something you would actually type)
Have at least 3 steps
Be for a workflow you've done manually — not something theoretical
Work when you invoke it (test it before sharing)

Exercise 3 — MCP setup (optional, 15 min)

Only if you identified a friction-twice use case in class: install one MCP, configure it.

WEEK 3Build a Feature

📖Before classASYNC

~55 min

Readings

1

Define done before you start The 3-criteria rule: before any code, define 3 verifiable success criteria. "It looks right" is not a criterion. "The record saves and appears in the list" is.

10 min

2

Design preview before code Sketch in plain text before asking Claude to list files. Catches misunderstandings before a diff exists. Costs 2 minutes, saves 30.

10 min

3

Consumer sweep When you change a convention, grep every file that depends on it before calling it done. Highest-recurrence error: one change, stale references everywhere.

10 min

🧠 Quiz — answer before class (10 min)

1.Give one example of a bad success criterion and a good one for the same feature.

2.What is a "design preview" and when do you do it?

3.True or false: you should deploy first and check if it works after.

4.You rename a field in one file. What do you do before calling it done?

5.What is the compounding loop in one sentence?

Instructions — how to prepare for class

Bring your first rule and first skill from last week — have them live in your folder

Pick one real feature to build (small: one button, one form, one script). You'll use this during class.

Have your real work folder open in VS Code

🎥In classLIVE

56 min content + ~19 min FAQ

#	Subtopic	Type	Time	What happens
1	Recap of reading	—	5 min	"You read the 3-criteria rule — today you write criteria live before we build anything. You read about design preview — today everyone sketches before touching Claude."
2	Show and tell — rules & skills	Demo	5 min	Instructor picks 2-3 participant submissions from last week. Shows what a strong rule and skill look like. Brief, high-energy opener.
3	Define done — gate	Exercise	8 min	Participants write 3 success criteria for their feature live. Instructor doesn't proceed to the next block until everyone has 3 written. This is a hard gate — not optional.
4	Design preview	Exercise	10 min	Participants sketch their feature in plain text live. Instructor reviews 2-3 sketches with the group — shows what a useful sketch looks like vs. a vague one.
5	Build-feature walkthrough	Demo	20 min	Instructor runs a complete feature build: preview → scan → branch → build → self-review → deploy. Then opens the after-class exercise and shows exactly what participants will replicate.
6	Compounding loop ritual	Demo	8 min	Instructor writes one capture live at the end of the demo: one sentence on what went wrong. Shows the format and where it goes. Then previews the after-class compounding exercise.
—	FAQ	—	~19 min	—

✏️After class — exerciseASYNC

~65 min · share before Week 4

Exercise 1 — Build and ship your feature (45 min)

Build the feature you defined in class. You already have the success criteria and the design sketch.

Steps:

Run the build: approve the plan only after checking step 1 first (canary check)
Run the consumer sweep: grep for any files that depend on what you changed
Deploy and verify: URL returns 200, all flows work, no console errors
Check your 3 success criteria — all must pass

Exercise 2 — First compounding entry (20 min)

Write your first real mistake-log entry.

Format (3 sentences):

What happened
Why it happened
What you'll do differently (this becomes a future rule)

WEEK 4Agentic Skills

📖Before classASYNC

~55 min

Readings

1

Parity testing — what the agent can't do Map every flow: which can Claude test automatically, which can only a human check? Email, PDFs, QR codes — Claude can't click those. Tests passing ≠ product working.

15 min

2

The first-user-path test Run one test with a completely fresh account — no pre-seeded data, no pre-configuration. Every test with pre-loaded state hides what real new users hit.

10 min

🧠 Quiz — answer before class (10 min)

1.All your automated tests pass. Is your product working? Explain.

2.Give one example of a flow Claude can test automatically and one it cannot.

3.What is the first-user-path test and what does it catch?

4.You wrote a rule that keeps getting skipped. What's the correct response?

5.What are the three graduation artifacts?

Instructions — how to prepare for class

Complete the Week 3 exercise: feature deployed, PR link ready to share

Have at least 2 entries in your mistake log

Write your pattern scan before class: what did you explain to Claude more than twice? What workflow did you run manually that Claude could have run?

🎥In classLIVE

53 min content + ~22 min FAQ

#	Subtopic	Type	Time	What happens
1	Recap of reading	—	5 min	"You read that tests passing ≠ product working. Today you watch a project fail the parity test — something automated tests missed. You read about the first-user-path test — today you see what it catches."
2	Parity testing	Demo	15 min	Instructor takes a real participant project. Maps every flow on screen. Identifies flows Claude cannot test. Shows the parity gap in practice.
3	The first-user-path test	Demo	10 min	Instructor runs a fresh-account test — no pre-seeded data. Shows what breaks. Participants recognise the same gap in their own projects.
4	Hook conversion + exercise preview	Demo	8 min	Instructor converts one rule from a participant's CLAUDE.md into a hook on screen. Then shows what the after-class hook conversion exercise looks like.
5	Graduation	Exercise	20 min	Each participant presents: one shipped feature + test, one parity test, one hook from a real cohort failure. Reads one entry from their mistake log aloud. Group: what did the system learn?
—	FAQ	—	~22 min	—

✏️After class — exerciseASYNC

~40 min

Exercise 1 — Hook conversion (15 min)

Convert one rule from your CLAUDE.md into a hook.

Steps:

Find a rule you've violated — or that Claude has skipped — at least once
Convert it to a PostToolUse or PreToolUse hook
Verify it fires: trigger the condition it guards against, confirm the hook catches it
Remove the text rule from CLAUDE.md — the hook replaces it

Exercise 2 — Pattern scan final output (25 min)

Write your final pattern scan. This is the artifact that shows the system is learning.

For each pattern found:

What you explained to Claude more than twice → candidate for a new rule or skill
What you ran manually that Claude could have run → candidate for a new skill or hook
One rule or skill you're committing to write next week