Do-this-exactly walkthroughs with the exact prompts to type, the console output you should see, and checkpoints along the way. Pick anything that would make your week better and build it.
A real, live https://USERNAME.github.io/SITE URL — a website you built and published without ever leaving the Claude Code console — that you can text to a friend right now.
Your own static website live on a *.pages.dev URL on Cloudflare's global network, deployed entirely from the Claude Code console with wrangler — plus a working mental model of when you'd reach for Workers instead.
A published, multi-section personal site (about, work, contact) that you designed, built, previewed, and put on a real URL in a single sitting.
Your site answers at https://yourdomain.com with a padlock, and you can explain — record by record — every piece of DNS you created to get there.
A GitHub Actions workflow, written and reviewed with Claude Code, that deploys your Cloudflare site automatically on every push to main — and you have watched both a green run and a deliberately-broken red run, then fixed it.
A working contact form on your live site that drops a real email into your inbox within seconds of a visitor hitting Submit — powered by a tiny Cloudflare Worker and an email API, with no server to run or patch.
A small script you designed with Claude Code that runs automatically every morning at a time you chose, writes you a "good morning" note, logs every run, and that you know how to check, read, and reschedule.
A pipeline you built that emails you a formatted briefing every morning at 7:00 — your product's key numbers with context, plus a relevant market signal — so you start the day already knowing what changed.
A monitor you built with Claude Code that checks the pages you care about every N minutes and emails you the moment one goes down or its content meaningfully changes — running on a schedule, without you watching.
A tested nightly backup — your chosen folders copied to a second location as dated snapshots, a backup.log that proves each run happened, a scheduled job that runs at 2 a.m., and a restore you have actually performed at least once.
Every Monday at 8:00 a polished weekly report — totals, week-over-week trends, a chart or two, and a short written narrative — appears as an HTML file (optionally PDF, optionally emailed), generated from your real data with zero manual work.
Three mornings a week, a folder on your machine fills with fresh, on-brand draft posts generated overnight from your own idea backlog — each one a starting point you edit and publish, never auto-posted.
A working skill you designed installed in ~/.claude/skills, firing automatically on your real requests, plus a correct mental model of what a skill IS and exactly when Claude reaches for it.
A personal continuity system — a tracker folder with one RESUME file per project and a Claude Code skill that updates it when you stop and reads it back when you return — so you can walk away from any project for weeks and be productive again in about a minute.
A layered CLAUDE.md setup — a personal global file with your standing rules plus a project file that makes any Claude Code session behave like a briefed teammate on day one — and the judgment to know what belongs in each layer.
A project where every file Claude edits is automatically formatted and linted the moment the edit happens — and a clear mental model of hooks as GUARANTEES the harness enforces, versus CLAUDE.md and skills as instructions the model usually follows.
Two or three personal slash commands you actually reuse — each one turning a paragraph you used to retype into a slash, a name, and a tab.
From the Claude Code console you can say "summarize what came in today and draft replies to the two that need me" against your real inbox — with permissions you understand and a setup where a reply can never leave your outbox without you pressing send.
Every morning — on demand via /today or automatically via cron — a brief of your real calendar that flags conflicts and back-to-backs, writes a prep line per meeting from your own notes, and shows where your focus time actually is. Read-only; it advises, you decide.
A pipeline that turns a 60-email morning into a 5-minute one — every incoming mail auto-sorted into your own small label taxonomy, a twice-daily digest of only what needs you, and a weekly review that reads your inbox back to you as a list of commitments you've made and not kept.
A pipeline where you drop any meeting artifact — typed notes, a Zoom/Teams/Meet transcript, even a photo of a whiteboard — into a meetings/ folder, run one command, and get clean minutes, an owner-per-item action list merged into your master tracker, and ready-to-review follow-up email drafts. Nothing is sent automatically.
A local, private SQLite database distilled from your own mail — who you know, when you last spoke, what about, and who's owed a reply — queryable in plain English from Claude Code, with a weekly nudge on relationships going cold.
A scheduled collector that gathers the external numbers your business decisions depend on — competitor prices, market rates, category news, app-store reviews (your pick) — into a growing local dataset with a timestamped history, so "what changed this month?" finally has an answer.
Your recurring spreadsheet chore — merging exports, cleaning messy columns, monthly rollups, cross-file lookups — reduced to one console command, plus the confidence to hand Claude Code any .csv or .xlsx and just ask it questions.
A private, always-current dashboard page — big KPI numbers, a trend chart, a "what changed" line, and a last-updated stamp — that only you can open, on any device, fed by your real data.
A tracker that watches the prices and stock you care about — competitors' products, supplies you buy, the gadget or flight you're waiting on — keeps a real price history, and alerts you only when something moves in a way worth knowing about.
A reusable recipe — point Claude Code at any API and get a clean daily report — proven three times on three genuinely different APIs (a no-key one, a key-required one, and one of your own services), so you leave owning the pattern, not just one script.
You can sit down with any database you have access to — a SQLite file from earlier lessons, your app's Postgres/MySQL, an exported .db from some tool — and get trustworthy answers by asking Claude Code in plain English, including knowing exactly when and how to distrust an answer.
A polished command-line tool on your PATH — with --help, sensible flags, a safe dry-run, and error messages you can act on — for a chore you actually do every week, plus the skill of specifying tools-for-yourself well.
A personal app of your own — with an icon on your phone's home screen, opening full-screen, working on the subway with no signal — built and shipped entirely from the Claude Code console, with zero app-store anything.
Two working MCP servers connected to your Claude Code — one that reads something you actually care about (your database, your browser, a service you use) plus one daily-driver service — along with a correct mental model of what MCP is and a security checklist you actually apply.
You've run a real, multi-part job as parallel subagents, watched them work at the same time, merged their results — and you can tell, before you start, when splitting a job will help and when it'll just make a mess.
One email at 7:00 every morning that is genuinely the first thing worth reading in your day — your business numbers with what changed, what in your inbox needs you, your day's shape with prep notes, and the external signals you track — synthesized by Claude into a briefing with a lead, not four reports stapled together.
A real phone number you own that answers with an AI-generated voice, handles your callers' most common questions from YOUR business rules, takes structured messages, and hands anything important to a human — built and tested from the Claude Code console in one sitting.
The method behind all 41 lessons, extracted and made yours — a repeatable five-move procedure that turns ANY idea ("I wish something would just…") into a scoped, buildable, watched system, plus a decision table that tells you exactly which lesson holds each piece you'll need.
A Claude Code-built control panel for a fleet of Android devices you own — discover and inventory every connected device, push the same update or config to all of them at once, run a health check across the fleet, relaunch the kiosk app on any that drifted, and log the whole run — turning "walk to each of 20 tablets" into one audited command, using ADB the way a real device-fleet admin does.
A definitions.yaml stating what "good" means for the five things you can't afford to lose, plus a checker Claude built that evaluates all of them on your chosen cadence and alerts only when a state actually changes — the foundation every other Category G lesson stands on.
A sentinel that every 15–30 minutes compares what's actually listening (open ports, firewall rules — local and at your cloud edge) against a customer-approved baseline file, and alerts the moment a new opening appears or an approved rule vanishes.
A monitor you built with Claude Code that reads events from your own badge/door access system every 1–5 minutes (or instantly by webhook), checks each against rules you wrote in plain language, and alerts you — with the who, where and when attached — the moment one of your own rules is broken.
A watcher you built with Claude Code that, every 5–15 minutes, PROVES each of your cameras, your recorder, and your status screens is genuinely alive — producing fresh images and actually recording — and alerts you by name the moment one goes dark, so you never again discover after an incident that camera 7 froze three weeks ago.
A read-only watcher you built with Claude Code that, every 5–15 minutes, checks YOUR OWN systems and cloud tenant against YOUR definition of normal account-health — failed-login bursts, impossible sign-ins, a new admin appearing, MFA switched off — and stays silent on an ordinary day, with evidence attached when it isn't.
A read-only sentry you built with Claude Code that watches everything in your world with an expiry date — TLS certificates, domain registrations, tokens with known lifetimes, backup freshness, and disk space — and gives you one calm heads-up weeks before any of them can lapse, plus a weekly "here's everything and how much runway it has" digest so silence never has to mean "broken."
One phone-friendly page of green / amber / red tiles — every watcher you built across Category G reporting into a single board through one shared JSON contract, with a customer-written escalation policy, a dead-man switch that turns silence into a STALE-amber tile, and an external ping watching the board itself — so a five-second glance from bed replaces the worry.
The ability to take any contract, lease, or agreement you're about to sign, get a plain-English explanation of every clause, a flagged list of terms that are unusual or work against YOU, and a sharp set of questions to bring to a real lawyer — turning "I guess I'll just sign it" into informed consent.
The ability to understand a lease before you sign, know what your landlord can and can't legally do, decode a deposit-deduction letter or eviction notice, and draft a firm, factual, dated response that cites the lease and requests specifics — while knowing exactly when the situation has left "write a letter" territory and needs a tenant-rights org or a real lawyer today.
A clear, fair client agreement — scope, payment schedule, revisions, kill fee, IP ownership, late-payment terms — drafted with Claude Code, where you can defend every clause you put in front of a client, plus a reusable template with fill-in fields and a plain-language cover note, and a clear sense of when a deal is big enough to have a lawyer bless the template once.
The ability to take any dense government form, benefits letter, or bureaucratic notice and get it translated into plain language — what it's asking, what it means for you, the deadline, what documents to gather, and a drafted response — so the system stops being a wall you bounce off, while knowing that Claude explains and prepares but never makes the official determination.
The ability to run a real benefits, permit, or aid application (or renewal) from start to submitted — a document checklist built for your exact situation, help understanding and honestly answering every field, a tracked map of the multi-stage deadlines, and a saved record proving exactly what you submitted and when — while knowing Claude organizes the application but never decides your eligibility or invents your answers.
The ability to take a denial — of a benefit, an insurance claim, a permit, a refund — and understand exactly WHY it was denied, judge honestly whether you have grounds, gather the evidence that answers the actual reason, draft a clear factual appeal, and file it before the (often short) deadline — while knowing the point where this needs legal aid or a professional advocate.
A private, fully local log of symptoms, vitals, and medications-taken that you keep over days or weeks, which Claude Code turns into a clear, objective one-page summary — a timeline, patterns in what you recorded, and questions to ask — that you hand your doctor so the appointment starts from facts instead of fuzzy memory.
One accurate, current medication and care list — every drug, dose, timing, and prescriber, plus allergies, key contacts, and upcoming appointments — that Claude keeps clean and reprints on demand as a wallet card, a fridge sheet, or a hand-off summary for a new caregiver or the ER, so the whole circle of care works from the same facts instead of scattered memory.
The ability to take a confusing medical bill and its EOB, have Claude line them up so you can see what was billed, what insurance paid, and what you actually owe and why — spot the common, checkable errors, organize the whole paper trail locally, and draft a factual itemized-bill or billing-question request — so you walk into the call or dispute informed, not guessing.
A repeatable workflow where Claude Code helps you draft honest alt text, format captions and transcripts, rewrite dense copy into plain language without losing meaning, and run a first automated accessibility audit of a page — producing a prioritized, human-checkable fix list — so what you publish works for people using screen readers, captions, or plain-language needs, while you keep the real verification (assistive tech, disabled users) where it belongs.
A systematic remediation pass over an existing site or document set — an inventory of every accessibility issue across many pages, a triage by real user impact rather than raw error count, high-impact and mechanical fixes applied with Claude as small reviewable diffs you approve, verification that each fix broke neither layout nor meaning, and an honest remediation report of what's fixed, what's deferred, and what still needs human and assistive-tech testing.
The ability to turn a specific access need — a parent who can't read small text, a family member with a tremor, low vision, hearing loss, memory trouble, or your own RSI — into concrete, working adaptations Claude Code helps you build: bumped-up readable views, a one-tap launcher, read-aloud for documents, high-contrast or large-print reformatting of things they actually use, and a plain how-to they keep — built with the person, to their preference, so the result is dignity and independence, not a tech project.
A set of Claude Code helpers for the one-person-does-everything nonprofit or community group — personalized donor thank-you drafts pulled from a donations CSV, a volunteer roster and shift-reminder tracker, a rolling deadline calendar for grant reports, filings, and renewals, and a monthly update assembled from real numbers — where Claude drafts and a human personalizes, verifies, and sends, so time goes back to the mission instead of the admin.
A repeatable way to scaffold a grant proposal and its later report with Claude Code — a reusable "org facts" file of your true mission, history, and verified numbers; a funder-requirements checklist so you answer exactly what's asked; first drafts of narrative sections built only from that verified data; a budget table that adds up; and a compliance check against the funder's limits — so your energy goes to the story, and you never submit a number you can't back.
The ability to engage local government as a prepared citizen — draft a clear public-records request that asks for exactly the right records, track it to its legal response deadline, digest the dense documents, agendas, or minutes you get back into plain language, and prepare a short, factual public comment — so an ordinary resident can participate on equal footing instead of being worn down by process.
A private, LOCAL dashboard built from your own bank and card CSV exports — spending by category, the subscriptions you forgot you're paying, a month-over-month "what changed," and progress toward one goal you set — turning "I have no idea where it goes" into a clear picture you own and that never leaves your machine.
A calm, private, fully local picture of everything you owe — balance, rate, minimum, due date, and who holds it — ordered so you can see what to deal with, a transparent model of what a given monthly amount does over time, and a factual hardship-letter or payment-plan-request draft you can send a creditor — turning "I'm afraid to open the envelopes" into "I can see it, and here's my next move," while being clear about where a real professional takes over.
A calm, fully local system that turns the annual pile — bank and brokerage statements, 1099s and W-2s, receipts, that letter you didn't understand — into an organized, labeled record: Claude decodes what each document is and why it matters, builds a have-vs-still-need checklist for taxes or an advisor meeting, flags the questions worth asking, and produces a clean summary and folder you hand your tax pro — so you arrive prepared and pay them for the expert part, not to sort your paperwork.
Two current documents — a printable household EMERGENCY PLAN (who to call, where to meet, medical info, shutoffs) on the fridge, and a DIGITAL-ESTATE INVENTORY pointing a trusted person to everything that exists and how to reach it — stored encrypted, reachable by the one or two who'd genuinely need it, holding pointers, never raw secrets.
A documented home inventory and recovery file — a room-by-room list of what you own (photos, model/serial, rough value, purchase info where you have it), your real insurance policy facts (carrier, where the policy number lives, coverage limits, deductible, claim phone), and a step-by-step first-48-hours plan after a fire, flood, or storm — organized locally and backed up OFF-SITE so it survives the very disaster it's for.
A gentle, organized way to help an aging parent (or yourself) get the essentials in place — a checklist of the documents that matter as pointers to where they live and who to call, a warm agenda for the hard conversation so it's caring rather than clinical, and a shared "if something happens, here's what you need to know and who to reach" file the family can actually find — so this gets done on a calm day, with dignity, instead of a terrible one.
Opening a project uses the same free magic-link sign-in as the training — sign in once, use both.