Is the language too technical? Start with the Glossary
All the course terms — system prompt, token, harness, diff, agent… — in plain English, with an everyday analogy for each one. You don't need to memorize anything: it's a cheat sheet to consult when you get stuck.
Why this course
Most of what’s taught about prompting is a collection of isolated tricks. This material is different: real prompts, leaked and official, examined as engineering. Four ideas guide everything.
Rule with rationale > bare instruction
The best prompts don't command—they explain. A rule with the reason behind it generalizes to unforeseen cases; a prohibition without justification breaks at the first edge case.
Consolidation > accumulation
A good prompt doesn’t grow by stacking exceptions. It gets shorter by replacing ten specific rules with one principle that encompasses them. You can see this in real diffs between versions.
Safety by default
The safety section comes early and is principle-based, not a list of blocks. It signals precedence — safety before usefulness — and describes the default, not the mechanics.
Layers: harness × behavior
The stable prompt (behavior) is one thing; the context the harness injects each turn is another. Separating the layers is what prevents incorrect conclusions when reading a leak.
The 4 tracks
From vocabulary to lab. Each track is self-contained, but the order tells a story: first you read, then compare, and finally write.
🧬 Anatomy & Patterns
The vocabulary—the anatomy of the 6 sections, the 12 named patterns, and the anti-patterns.
Enter the track →🔬 Evolution by Diffs
What sets it apart—learning from real changes: Opus 4.6→Fable 5, safety, autonomy, personalities.
Enter the track →🏛️ Philosophies & Specialization
Comparative overview — four provider philosophies and specialization by product type.
Enter the track →🚀 Building Yours
The lab—from the minimum viable prompt to your version-controlled repo with a changelog and tests.
Enter the track →How the course works
Four steps, one for each track. The theory from one becomes the tool for the next.
Named patterns
Learn the vocabulary: the 6 universal sections, the 12 patterns, and the antipatterns. Without names, you can't compare anything.
Real diffs
Read what changed between versions of real prompts. The changes reveal intent better than any best-practices manual.
Comparison between providers
Put Claude, GPT, Gemini, and Grok side by side. Four different philosophies for the same problem — and what each product requires.
Laboratory
Write your own. From the minimum viable prompt to a version-controlled repository, with a changelog and tests — treating prompts as code.
Start reading prompts as engineering
Thirty minutes in Track 1 and you’ll never look at a system prompt the same way again. The course is free and the collection is open.