One connected map of the Claude stack.
Everything for understanding how Claude Code actually works, in one place: the mental model that frames the whole stack, a field guide to every primitive, a self-organizing concept graph, a section on evals, a side-by-side comparison matrix, and a decision guide for picking the right tool. Pick your way in below — go deep on one idea, or broad across the whole stack. Reviewed 2026-06-02.
🧠 Mental model
How Claude Code really works — prompt, context, harness, and the loop. Start here if the stack still feels like a pile of nouns.
→🧱 Primitives
The field guide: hooks, skills, MCP, subagents, agent teams, dynamic workflows, agent view, conventions, policies — one full page each.
→🕸 Concept graph
Every idea wired to its neighbours — the self-organizing map of atomic notes across the whole stack.
→⚖️ Comparison matrix
The primitives side by side: who invokes each, hard vs. soft enforcement, and when to reach for which.
→🧭 Decide
A decision guide that turns “I want to do X” into “reach for Y” — the fast path to the right primitive.
→🧪 Evals
How to know an AI product actually works: offline vs online evals, freezing the world, noise, graders and their blind spots, safety gates. Plain-language lessons, learned the hard way.
→🏷️ Legend
What the badges mean — hard (client-enforced) vs. soft, plus preview and experimental.