Setting up Claude Code for your team

The five layers — in order of leverage

CLAUDE.md, settings.json, hooks, rules/, skills and agents. What each one does, which ones most teams actually need, and the order to set them up so you are not configuring things before you understand why.

Instructions Claude follows vs. rules it cannot break

CLAUDE.md is a suggestion. Hooks are a guarantee. This distinction is the mental model the whole system is built on — and the reason most Claude Code setups have gaps in them.

What to actually write in CLAUDE.md

Four starting templates — solo project, team backend, agency client repo, and non-code ops setup — with annotations explaining what each section does and why. The answer to the question every setup article skips.

The five decisions your team makes together

Who owns CLAUDE.md. What the deny list blocks. How to write hooks that work on every machine. The split between personal and shared config. And when to graduate to rules/ — which is later than most people think.

Keeping CLAUDE.md accurate over time

CLAUDE.md decays quietly: it gets too long, instructions contradict each other, the architecture changes but the file does not. The four failure modes and the one maintenance habit that prevents all of them.

Beyond config: the agent layer

Once your .claude folder is set up, this is what comes next. What Managed Agents are, how the hosted agent loop works, and how it compares to building your own agent harness. For teams ready to go further than configuration.