Claude / Operating Manual
The Claude Code operating manual.
The terse-lookup layer of the Claude Manual. One page per capability, written for people who already know what they are after and want the shape of it fast. If you are new to Claude, read Start Here first. For orchestration and autonomy, see Advanced.
Topics
Deep guides organized by what you are trying to do. The best place to start when you have a goal, not a keyword.
- Primitives the principles assumeThe generic Claude primitives the KMS principles build on, as one digest: subagents, skills, CLAUDE.md, memory, hooks, MCP, long context, prompt caching, fast mode, tool selection, context window management.
- Memory & ContextClaude Code only remembers what you write down, and everything you write down loads every turn. How to keep the right things present and the rest out, across sessions and within one.
- Models & CostModel choice is the biggest cost lever you have, roughly a 30x spread for the same work. Default to the cheapest model that clears the task, override per project, and reach up only for the hard parts.
Reference
Quick reference for each individual capability, one page each.
- SkillsProgressive-disclosure knowledge units, loaded by name at startup, with full instructions pulled only when Claude matches them to the task.
- Slash CommandsControl sessions and automate workflows with built-in and custom `/` commands.
- Memory & CLAUDE.mdInstruction files that load every session, giving Claude persistent project and personal context without re-explaining yourself each time.
- Context managementSteer what stays in the window and what persists across sessions. /compact directs compaction, /clear resets, /memory edits the memory files.
- Remote & WebRun and steer Claude Code beyond one terminal: remote sessions, cross-device Remote Control, and the desktop GUI.
Setup
Install, config, permissions, model selection.
- Installation & SetupGet Claude Code installed, authenticated, and ready to use in one sitting.
- Permissions & SettingsConfigure what Claude can run without asking via settings.json allowlists at project or user scope.
- Model SelectionWhich Claude model to run, when to flip to fast mode, and how fallbacks work.
- IDE & VS CodeRun Claude Code inside VS Code or JetBrains instead of a standalone terminal.
Releases
Running changelog of models and Claude Code drops.