What Can I Change

Claude Code has many places where you can change how it works. This page starts from what you want to do, and it points to the right place. Each row links to the page that lists every key, flag, or field.

At a Glance

Question	Answer
Can you use it?	✅ Yes. Find your goal in a table below and follow the link.
Can you change it?	Every row names the surface that changes the goal.
Can you turn it on or off?	The "How" column names the switch.
Where does it live?	The pages in Configure, Extend, and How It Works.
Status	Documented Undocumented

The Surfaces

Surface	Use it when	Page
Settings keys	You want a stored value that stays between sessions.	Settings Files and Scopes
Environment variables	You want a value for a shell, a script, or a CI job.	Environment Variables
CLI flags	You want a change for one run only.	CLI Flags
Hooks	You want your own code to run on an event, and maybe block it.	Hooks
Skills and commands	You want a reusable prompt or a /name shortcut.	Skills and Commands
Subagents	You want a helper with its own prompt, tools, and model.	Subagents
Plugins	You want to bundle and share several of these.	Plugins
MCP servers	You want Claude to use tools from another program or service.	MCP
Mod API	You want code that changes Claude Code itself, such as new UI or event handlers.	Mod API Overview

Behavior and Safety

I want to ...	Status	How	Page
Allow or block a tool call without a prompt	Documented	permissions.allow, permissions.deny, and permissions.ask rules	Permission Rules
Start in a different permission mode	Documented	--permission-mode flag or the permissions.defaultMode setting	Permission Modes
Let a classifier approve safe actions	Documented	Auto mode and its rules	Auto Mode
Limit what shell commands can touch	Documented	The sandbox settings	Sandbox
Block a command with my own logic	Documented	A PreToolUse hook	Hook Events
Understand why a call was allowed or asked	Documented	Read the flow	Permission Flow
Let managed policy lock a value	Documented	Managed settings	Enterprise Settings

Models and Cost

I want to ...	Status	How	Page
Pick the model	Documented	/model, --model, the model setting, or ANTHROPIC_MODEL	Model Settings
Change how hard the model thinks	Documented	The effort and thinking settings	Models and Effort Variables
Use Bedrock, Vertex, or Foundry	Documented	Provider variables	Models and Providers
Change when old messages are summarized	Documented	Compaction settings and variables	Context and Compaction Variables
Understand how a model is chosen	Documented	Read the flow	Models
Route requests through a proxy or custom CA	Documented	Network variables	Network and Proxy Variables

Instructions and Memory

I want to ...	Status	How	Page
Give Claude standing project advice	Documented	CLAUDE.md	Memory and Rules
Apply advice only to some files	Documented	A rule file with paths in .claude/rules/	Memory and Rules
Turn off auto memory	Documented	autoMemoryEnabled: false or CLAUDE_CODE_DISABLE_AUTO_MEMORY=1	Memory and Rules
Change the tone or style of replies	Documented	An output style	Output Styles
See the real system prompt	Undocumented	Captured prompts on this site	Prompt Captures
Understand how the prompt is built	Documented	Read the build steps	System Prompt
Add text to the system prompt	Documented	--append-system-prompt	CLI Flags

Tools and Automation

I want to ...	Status	How	Page
Run something after every file edit	Documented	A PostToolUse hook	Hooks
Add my own /command	Documented	A skill or a file in .claude/commands/	Skills and Commands
Start a helper for one kind of task	Documented	A file in .claude/agents/	Subagent Frontmatter
Connect a database or an API	Documented	claude mcp add or .mcp.json	MCP Config
Limit which MCP servers may run	Documented	MCP allow and deny settings	MCP Settings and Policy
Run Claude Code from a script	Documented	Print mode with -p	Headless Mode
Drive it from my own program	Documented	Stream JSON and the control protocol	Control Protocol
Share my setup with a team	Documented	A plugin and a marketplace	Plugin Marketplaces

Screen and Terminal

I want to ...	Status	How	Page
Show my own status line	Documented	The statusLine setting	Status Line
Change theme, spinner, or notices	Documented	Interface settings	Interface Settings
Change terminal behavior with variables	Documented	UI and terminal variables	UI and Terminal Variables
Change keyboard shortcuts	Documented	/keybindings opens ~/.claude/keybindings.json	Slash Commands
Add a screen element or dialog	Documented	A mod that uses $.ui	Mod API UI

Sessions, Files, and Privacy

I want to ...	Status	How	Page
Continue or branch an old session	Documented	--continue, --resume, /branch	Sessions and Storage
Keep sessions for a shorter time	Documented	cleanupPeriodDays	Core Settings
Stop Claude Code from saving anything	Documented	CLAUDE_CODE_SKIP_PROMPT_HISTORY or --no-session-persistence	Sessions and Storage
Move the whole ~/.claude folder	Documented	CLAUDE_CONFIG_DIR	Files on Disk
Turn off telemetry or debug output	Documented	Telemetry variables	Telemetry and Debug Variables
Know where a file lives	Documented	The file map	Files on Disk

What You Cannot Change

Some parts are fixed in the program. You cannot edit the built-in tools, the permission engine, or the request format from a setting. You can still watch or shape them with a hook or a mod. The Architecture and Agent Loop pages show where each part sits.

warning

Do not use Undocumented items in shared setups without a plan. They can change in any release without notice.

Example

You want Claude to run your formatter after each edit and to ask before it runs git push. Two small settings do both.

.claude/settings.json

{
  "permissions": {
    "ask": ["Bash(git push:*)"]
  },
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|Write",
        "hooks": [{ "type": "command", "command": "bunx prettier --write ." }]
      }
    ]
  }
}

The first part is a permission rule. The second part is a hook. Both live in the shared project settings file.

Related Pages

• How to Read These Docs
• Files on Disk
• Environment Variable Groups
• Undocumented Items