InContext

A working reference on the craft of context for AI — the files, patterns, and small choices that decide whether a model is useful or useless on your problem.

Start with foundations View on GitHub

The bet: the unit of intelligence is not the model — it’s the model plus its context apparatus. A better apparatus on the same model beats a better model on a worse apparatus, and the gap is widening. More →

---

What’s here

Foundations — the vocabulary and the underlying ideas. What a context resource is, why context behaves like a budget rather than a bucket, what models actually attend to.

Per-tool — practical references for the artifacts in use today. CLAUDE.md, AGENTS.md, Cursor rules, system prompts.

Patterns — recurring shapes that work across tools and models. Scope-by-directory, layered context, role + task + constraints, retrieval before generation.

Skills — the small set of named, reusable workflows worth keeping installed. Headlined by an XP-style fixed loop for shipping with an agent.

Reference — quick lookups. Format cheat sheets, directives that work, common failure modes.