Your Setup

What Claude loads before it answers.

Six sketches for the parts people confuse: project files, personal files, loading order, commands, skills, and subagents. Use this when the setup starts to feel invisible.


Anatomy of the .claude folder, a tree of settings, commands, rules, skills, and agents.
01

The .claude/ folder

Project instructions, permissions, custom commands, rules, skills, and subagents all live here. This is the shared setup for the project.

Project .claude versus global home .claude, one shared with the team, one personal.
02

Project config vs your config

One config sits in the project and can be shared. One sits in your home folder and belongs only to you. Both can load into the same session.

CLAUDE.md loading order, personal, project, and global layers merging into the system prompt.
03

How CLAUDE.md loads

Your personal notes, the project instructions, and your global rules all stack into one set of instructions at the start of every session. Higher in the stack wins.

Writing a markdown file that becomes a custom slash command Claude can run.
04

Custom commands

A filename, a description, and a few lines of markdown can become a command you run on demand.

Commands you trigger manually versus skills Claude invokes automatically.
05

Commands vs skills

Same idea, different trigger. Commands wait for you to call them. Skills watch for the right moment and fire on their own.

A main agent spawns a subagent that does the exploration and returns a compressed answer.
06

Subagents

A subagent does the messy exploration in its own context window, then hands back the useful answer. The main session stays cleaner.

← Back to home base