Huxley's Game Maker: A guided 8-day game design curriculum

The constraints

What made this hard?

Four constraints. The first was attention. A curriculum that takes itself too seriously dies in the first session. A curriculum that's too loose doesn't earn the next session. The pacing had to alternate structure and creative work, with completion checklists short enough that a child can finish one in a single sitting, and stretch tasks for the days he has more energy.

The second was sequencing. Most "learn game design" content is genre-specific, which fails when the learner doesn't know what genre yet. The curriculum had to start with the highest-leverage decisions (concept, theme, core feeling) and end with implementation, so the learner is always making decisions with enough context to judge them. Day one is "pick your game". Day eight is "the game comes alive" with a working playable prototype.

The third was creative honesty. There's a category of children's curriculum that pretends every kid is a future game director when really the work is mechanical. This curriculum had to treat the learner like an actual designer; the villain ethics module insists "great antagonists need justification, they can't just be evil". That kind of substantive design thinking is the work, and dumbing it down would have produced something he wouldn't take seriously.

The fourth was scope. Eight themed days, each with a substantive engaging blurb, a real-world fact, thinking prompts, work steps, completion checklist, notebook exercises, learner feedback prompt, embedded Canva template, linked resources, and stretch tasks, plus the populated Canva templates the learner uses to capture decisions. Manually writing all of that for a curriculum aimed at one specific learner is its own project. Per-day Claude research scoped against the design objective, plus Claude-built Canva templates, compressed it into the actual build window.

The approach

How did Tincture frame the problem?

Eight themed days, each building on the last, each with a specific creative output.

Day 1: Pick Your Game (genre, theme, core feeling, elevator pitch).

Day 2: Build Your Characters (hero, villain, supporting cast with character design cards).

Day 3: Write Their Stories (backstories, the villain monologue rule, relationship mapping, world bible).

Day 4: Design Your First 3 Levels (mood boards, top-down maps, difficulty curve graphing, level theming).

Day 5: Game Mechanics & Rules (core mechanics as IF/THEN statements, win/lose conditions, enemy AI logic, paper playtesting).

Day 6: Sound & Music Design (themes, action effects, mood playlist, boss-fight audio moment).

Day 7: First Code Sprint (Cursor, Python, a moving rectangle, collision detection).

Day 8: The Game Comes Alive (replace the rectangle with the first character, sixty-second playtest).

Every day has the same structure: an engaging blurb, a real-world game design fact (chess has six pieces, Minecraft took one person six days to prototype), thinking prompts, three to four concrete work steps, a completion checklist, structured notebook exercises, a specific Claude AI prompt the learner pastes for design feedback, embedded Canva templates, linked resources and videos, and stretch tasks.

The discipline-typed badge system (Think blue, Create purple, Design orange, Build green, Code yellow) is the secret weapon. The learner sees that game design draws on five different skill types, and the badges cluster differently for each kid, so a "I'm not really a code person" doesn't kill the curriculum on day seven. Most of the work is Think, Create, and Design. Code is the last fifth.

The curriculum content per day is generated through a structured Claude research pipeline, scoped against the day's design objective, the kind of game design thinking the day should teach, and the format constraints (kid-readable, completable in a single sitting, ending with creative output). For each day, Claude returns the engaging blurb, the real-world fact, the thinking prompts, the work steps, the notebook exercises, the learner feedback prompt to paste into Claude.ai, the linked resources (Game Maker's Toolkit, Extra Credits, HeroForge, Itch.io), and the stretch tasks. Each output is reviewed against the design intent and edited where the model drifted before being dropped into the curriculum's structured data array.

The Canva template layer is the second AI surface. Each day's Canva template is the artefact the learner populates with their decisions (character design cards on Day 2, level mood boards on Day 4, sound mood boards on Day 6, and so on). Building those templates manually for eight days would have been its own project. Instead, Claude generates the structured content for each template (the prompts, the field labels, the example placeholders), and the Canva API turns that content into the actual Canva file. The learner gets a populated template with structured prompts rather than a blank canvas, which is the difference between "I don't know what to put here" and "I know what's being asked of me".

The build

What was shipped?

A web app with eight day-themed modules, each containing the engaging blurb, the real-world fact, prompts, three to four work steps, the completion checklist, notebook exercises, Claude AI prompt for feedback, embedded Canva template, linked resources, stretch tasks, and a discipline-typed badge unlocked on completion.

The curriculum content per day is the output of a structured Claude research pipeline scoped against the day's design objective and format constraints, with each generation reviewed against the intent before being dropped into the inline data array. The Canva templates the learner populates per day are built using Claude content generation plus the Canva API: Claude generates the structured prompts, field labels, and example placeholders for each template, and the Canva API turns that structured content into the actual Canva file. Eight templates, all populated, ready for the learner to walk into.

Real-world resources embedded throughout: Mark Brown's Game Maker's Toolkit, Extra Credits, HeroForge for character modelling, Itch.io for inspiration. Paper playtesting built into Day 5, insisting on testing with humans before any code is written, catching design flaws early.

A light/dark theme toggle with a carefully designed colour token system, because a curriculum a child uses on his phone in the evening needs to not blast his face with light at bedtime.

Discipline badges: Think (blue), Create (purple), Design (orange), Build (green), Code (yellow). Earned on day completion. Visible on a progress dashboard showing how the badges cluster for this learner specifically.

All content lives as a structured data array in a single 934-line HTML file. Vanilla JavaScript, Supabase Postgres for progress tracking, Vercel deployment. No build step. No framework. Sessions load instantly.

The outcome

What were the results?

Live and in use. The villain ethics module alone has sparked more conversation than any homework assignment, because a thirteen-year-old defending why his antagonist is doing the right thing in his own head is a different conversation than "what did you do at school today".

The structural outcome is what got built into the design thinking, not the game itself. The curriculum teaches that the best games have simple mechanics with deep combinations (the chess example), that prototypes can ship in days not months (the Minecraft example), and that paper playtesting catches design flaws code doesn't. Those are transferable creative-work principles, applied through games.

The methodological outcome is the badge system. Most children's curricula collapse skill types into a single "you're doing well" gauge. The discipline-typed badges show that "doing well" looks different in different parts of the work, and that the same learner can be a Think-and-Design kid even if he's not a Code kid. That framing alone has shifted how this learner thinks about what he's good at.

The build-side outcome is the AI pipeline. The curriculum content and the Canva templates would have been a multi-week writing-and-design project without Claude doing the per-day research and the Canva API doing the template building. The pipeline compressed that into the actual build window, which is what made a one-learner curriculum economically possible at this depth.

What it took

What tools and methods were used?

Cursor and Claude Code for the build itself. Claude (Claude.ai for deep research, Claude Code for content structuring) for the per-day curriculum research pipeline. Claude plus the Canva API for building the populated Canva templates. HTML5, CSS3, vanilla JavaScript. Supabase Postgres for progress tracking. Vercel for deployment. Embedded Canva templates accessed by URL. Real-world resources linked rather than reproduced. Claude AI prompts embedded per day, designed so the learner can paste a prompt into Claude.ai and get specific design feedback on his own work.

The methodological underpinning is the practice's standard pattern for one-user education products: build with the same rigor as a curriculum for a thousand learners, but optimize the architecture for the specific user, which here meant short sessions, light/dark theme for bedtime use, embedded prompts so AI feedback is one paste away, and content written for a child who's already smart about games rather than one being introduced to them.

The other move worth naming: two AI surfaces with different rules. As a build tool, AI generates curriculum content and populates Canva templates. Claude does the research, the structuring, and the template prompts. The human reviews against the design intent before content ships. As a learner-facing surface, AI is strictly a feedback layer. The Claude prompts embedded in each day let the learner paste their own work into Claude.ai and get structured feedback, never generated game concepts. AI for the builder is generation; AI for the learner is feedback. That distinction is the right one for AI in education content. AI used for feedback teaches; AI used for generation in front of a learner replaces.

The takeaway

What's the transferable principle?

Most children's curricula choose between rigor and accessibility. They either treat the kid like a future expert (and lose them) or they treat them like a beginner (and bore them). The work that lands does both. Substantive design thinking, scaffolded carefully enough that the learner can actually do the work.

For Huxley's Game Maker, that meant the villain ethics module being substantive (great antagonists need justification, they can't just be evil) while the day structure was light enough to finish in a sitting. The rigor lives in the content; the accessibility lives in the pacing.

The second transferable principle, broader than education: AI as feedback is more valuable than AI as generation in front of the learner. The Claude prompts in this curriculum let the learner submit his own work and get structured feedback, which keeps him doing the work. AI prompts that generate concepts for him would have eliminated the work, which would have eliminated the learning.

The third principle, specific to the build side: AI generation in the right place is what makes the depth feasible. Eight days of substantive content plus eight populated Canva templates is too much to write and design manually for one learner. Per-day Claude research scoped against design intent, plus Claude-generated Canva templates via the Canva API, compresses the build window without diluting the depth. The boundary is the move: AI generates the build artefacts, the human reviews against the spec, the learner gets feedback from AI on their own work. Three roles for AI, three different rules for each.