Name	Name	Last commit message	Last commit date
parent directory ..
README.md	README.md

Claude Code

Anthropic's official AI coding agent for the terminal, IDE, desktop app, and browser.

Vendor: Anthropic | License: Proprietary | Runtime: Node.js (CLI); native binary also available

Installation

Native installer (recommended — auto-updates)

macOS / Linux / WSL:

curl -fsSL https://claude.ai/install.sh | bash

Windows PowerShell:

irm https://claude.ai/install.ps1 | iex

Windows CMD:

curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

Homebrew

brew install --cask claude-code
# or for latest channel:
brew install --cask claude-code@latest

Note: Homebrew does not auto-update. Run brew upgrade claude-code manually.

WinGet

winget install Anthropic.ClaudeCode

npm (alternative)

npm install -g @anthropic-ai/claude-code
claude

Linux package managers (apt, dnf, apk) are also supported — see the setup page.

Configuration Files

File	Scope	Purpose
`~/.claude/settings.json`	Global (user)	User-level settings, hooks, permissions
`.claude/settings.json`	Project (shared)	Project-level settings, committed to git
`.claude/settings.local.json`	Project (personal)	Personal project overrides, git-ignored
`managed-settings.json` / plist / registry	Org (managed)	Admin-controlled, highest precedence
`CLAUDE.md`	Project	Natural-language instructions loaded into context
`~/.claude.json` / `.mcp.json`	User / Project	MCP server configuration

Scope precedence (high → low): Managed → Command line → Local → Project → User

Most settings hot-reload on file change without a session restart (including permissions and hooks).

Hooks

Hook events are defined in settings.json under the hooks key.

Supported Events

Event	When it fires	Can Block?	Notes
`SessionStart`	Session begins or resumes	❌
`Setup`	`--init-only` / maintenance flag	❌
`UserPromptSubmit`	User submits a prompt, before processing	✅ (exit 2 erases prompt)	30 s timeout override
`UserPromptExpansion`	User-typed slash command expands into prompt	✅	30 s timeout override
`PreToolUse`	Before every tool call	✅
`PermissionRequest`	Permission dialog triggered	✅ (exit 2 denies)
`PermissionDenied`	Tool denied by auto-mode classifier	❌	Use `hookSpecificOutput: { hookEventName: "PermissionDenied", retry: true }` to retry
`PostToolUse`	After tool call succeeds	❌	Tool already ran
`PostToolUseFailure`	After tool call fails	❌
`PostToolBatch`	After a parallel tool-call batch resolves	✅ (stops loop)
`Notification`	Agent sends a notification	❌
`MessageDisplay`	While assistant message text renders	❌	10 s timeout override
`SubagentStart`	Subagent is spawned	❌
`SubagentStop`	Subagent finishes	✅
`TaskCreated`	Task created via `TaskCreate`	✅ (rolls back)
`TaskCompleted`	Task marked completed	✅
`Stop`	Claude finishes a turn	✅ (continues conversation)
`StopFailure`	Turn ends due to API error	❌	Output/exit code ignored
`TeammateIdle`	Agent-team teammate about to go idle	✅
`InstructionsLoaded`	`CLAUDE.md` or `.claude/rules/*.md` loaded	❌	Exit code ignored
`ConfigChange`	Config file changes during session	✅	Except `policy_settings`
`CwdChanged`	Working directory changes	❌
`FileChanged`	Watched file changes on disk	❌
`WorktreeCreate`	Worktree created	✅ (any non-zero fails creation)
`WorktreeRemove`	Worktree removed	❌
`PreCompact`	Before context compaction	✅
`PostCompact`	After context compaction completes	❌
`Elicitation`	MCP server requests user input	✅ (exit 2 denies)
`ElicitationResult`	User responds to MCP elicitation	✅ (becomes decline)
`SessionEnd`	Session terminates	❌

Hook Types

Type	Description	Available Events
`command`	Run a shell command (stdin = hook JSON, exit code drives behavior)	All events
`http`	HTTP POST to a URL (body = hook JSON)	All events
`mcp_tool`	Call a tool on a connected MCP server	All events
`prompt`	Single-turn LLM evaluation (yes/no) — 30 s default timeout	`UserPromptSubmit`, `UserPromptExpansion`, `PreToolUse`, `PermissionRequest`, `PostToolUse`, `PostToolUseFailure`, `PostToolBatch`, `Stop`, `SubagentStart`, `SubagentStop` only
`agent`	Spawn a subagent with Read/Grep/Glob tools (experimental) — 60 s default timeout	`UserPromptSubmit`, `UserPromptExpansion`, `PreToolUse`, `PermissionRequest`, `PostToolUse`, `PostToolUseFailure`, `PostToolBatch`, `Stop`, `SubagentStart`, `SubagentStop` only

Note: SessionStart and Setup only support command and mcp_tool types. Most informational/lifecycle events (e.g. Notification, MessageDisplay, InstructionsLoaded, SessionEnd) support command, http, and mcp_tool but not prompt or agent.

Hook Input (stdin JSON)

{
  "session_id": "abc123",
  "transcript_path": "/path/to/transcript.jsonl",
  "cwd": "/current/working/directory",
  "permission_mode": "default|plan|acceptEdits|auto|dontAsk|bypassPermissions",
  "hook_event_name": "PreToolUse",
  "tool_name": "Bash",
  "tool_input": { "command": "rm -rf /tmp/test" }
}

Exit Code Behavior

Code	Meaning
`0`	Success; stdout parsed for optional JSON output fields
`2`	Blocking error on supported events; stderr sent to Claude as error message
Other	Non-blocking warning; first line of stderr shown in transcript

Exit code 2 is not limited to PreToolUse — it applies to all events marked ✅ in the table above.

Command Hook Fields

Field	Required	Description
`type`	Yes	`"command"`, `"http"`, `"mcp_tool"`, `"prompt"`, or `"agent"`
`command`	Yes (command)	Shell command to execute
`args`	No	Argument list; when present, command is exec'd directly (no shell)
`shell`	No	`"bash"` (default) or `"powershell"`; ignored when `args` is set
`timeout`	No	Seconds before cancellation (see defaults below)
`statusMessage`	No	Custom spinner message while hook runs
`async`	No	`true` → run in background, output discarded, Claude proceeds immediately
`asyncRewake`	No	`true` → run in background and wake Claude if exit code 2 (implies `async`). Stderr shown to Claude as system reminder
`if`	No	Permission-rule pattern filter, e.g. `"Bash(git *)"`
`once`	No	`true` → run once per session then deregister (skill frontmatter only)

async vs asyncRewake: async: true is fire-and-forget (output discarded). asyncRewake: true also runs in the background but re-wakes Claude and sends hook output if exit code 2 — use for long background operations that may need human attention.

HTTP Hook Fields

Field	Required	Description
`url`	Yes	URL to HTTP POST
`headers`	No	Additional HTTP headers (key-value object)
`allowedEnvVars`	No	Env var names allowed for interpolation in headers

To block via HTTP hook: return 2xx with JSON { "decision": "block" } (non-2xx = non-blocking error).

MCP Tool Hook Fields

Field	Required	Description
`server`	Yes	Name of an already-connected MCP server
`tool`	Yes	Tool name on that server
`input`	No	Tool arguments; supports `${path}` substitution from hook JSON

JSON Output Fields (stdout from any hook)

These are the top-level universal fields returned in JSON from any hook's stdout:

Field	Default	Description
`continue`	`true`	`false` stops all processing
`stopReason`	—	Message to user when `continue: false`
`suppressOutput`	`false`	Hide hook stdout from transcript
`systemMessage`	—	Warning shown to user
`terminalSequence`	—	Terminal escape sequence (OSC/BEL)
`hookSpecificOutput`	—	Nested object for event-specific decisions (requires `hookEventName` inside)

hookSpecificOutput is a nested object — not a top-level field itself — containing event-specific fields. For example:

{
  "hookSpecificOutput": {
    "hookEventName": "PermissionDenied",
    "retry": true
  }
}

or, to inject context into Claude's context window from a PreToolUse hook:

{
  "hookSpecificOutput": {
    "hookEventName": "PreToolUse",
    "additionalContext": "This file is auto-generated — do not manually edit."
  }
}

additionalContext (a string injected into Claude's context window) is an event-specific field inside hookSpecificOutput, not a top-level field.

Timeout Defaults

Hook type	Default	Overrides
`command`	600 s	`UserPromptSubmit`/`UserPromptExpansion` → 30 s; `MessageDisplay` → 10 s
`http`	600 s	`UserPromptSubmit`/`UserPromptExpansion` → 30 s
`mcp_tool`	600 s	`UserPromptSubmit`/`UserPromptExpansion` → 30 s
`prompt`	30 s	—
`agent`	60 s	—

Configuration Example

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          { "type": "command", "command": "~/.claude/hooks/validate-bash.sh" }
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "Write|Edit",
        "hooks": [
          { "type": "command", "command": "~/.claude/hooks/format.sh", "async": true }
        ]
      }
    ],
    "Stop": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "~/.claude/hooks/background-validate.sh",
            "asyncRewake": true
          }
        ]
      }
    ]
  }
}

Async Hooks

Add "async": true to run a hook in the background without blocking Claude's execution (released January 2026).

Add "asyncRewake": true to run in the background and re-wake Claude if the hook exits with code 2 (useful for deferred validation that may need to surface an error).

Matcher Patterns

Pattern	Evaluated as
`*`, `""`, or omitted	Match all occurrences
Letters/digits/`_`/`\|` only	Exact string or `\|`-separated list
Any other characters	JavaScript regex

MCP tool format: mcp__<server>__<tool> (e.g. mcp__memory__create_entities; all tools from a server: mcp__memory__.*).

Built-in Tools

Tool	Description
`Bash`	Execute shell commands
`Read`	Read file contents
`Write`	Write/overwrite files
`Edit`	Apply targeted string replacements
`MultiEdit`	Multiple edits in one call
`Glob`	File pattern matching
`Grep`	Search file contents
`LS`	List directory contents
`TodoRead` / `TodoWrite`	Manage in-session task lists
`Agent`	Spawn subagents
`WebFetch`	Fetch a URL
`WebSearch`	Web search
`NotebookRead` / `NotebookEdit`	Jupyter notebook support

MCP Support

Config: .mcp.json (project) or ~/.claude.json (user)
Enable servers in settings under mcpServers
Servers expose additional tools available in the agent loop
MCP tool hooks use the mcp_tool hook type targeting connected servers

Agent / Subagent Configuration

Claude Code supports spawning subagents via the Agent tool and via agent-type hooks. Subagents have access to Read, Grep, and Glob tools by default. Background agents and agent-team features (with TeammateIdle hook) are also available.

Permissions

{
  "permissions": {
    "allow": ["Bash(git *)", "Read(**)"],
    "deny": ["Bash(rm -rf *)", "Read(./.env)"],
    "ask": ["Write(./config.json)"]
  }
}

Deny rules take precedence over allow rules.
Rules merge across scopes; higher-precedence sources can only add restrictions, not loosen them.

Skills

Skills and slash commands are two different things in the Claude ecosystem:

	Location	Shape	Invocation
Skill	`.claude/skills/<name>/SKILL.md` (project) · `~/.claude/skills/<name>/SKILL.md` (personal)	A folder whose `SKILL.md` has YAML frontmatter (`name`, `description`) + instructions	Auto-invoked by the model when the `description` matches intent
Slash command	`.claude/commands/<name>.md`	A single Markdown file; can reference `$ARGUMENTS` and embed tool calls	Typed explicitly as `/name`

A skill is a directory, not a file. The description field is what the model uses to decide when to reach for it, so it should enumerate trigger phrases.

Filesystem auto-discovery (Claude Code)

Claude Code (CLI, IDE extensions) scans the filesystem on startup and discovers every skill under .claude/skills/ and ~/.claude/skills/ automatically. Drop a folder in, restart the session, and it's available. No registration step.

Manual upload (Claude Desktop / claude.ai app)

The Claude Desktop app and claude.ai web app do NOT read your local .claude/skills/ folders. They have no filesystem-discovery step at all — pointing them at a repo or a local path does nothing. Skills must be added manually, one at a time:

Enable the sandbox: Settings → Capabilities (a.k.a. Features) and turn on Code execution / file creation — Agent Skills run inside that sandbox, so they're invisible without it.
In the same Capabilities → Skills panel, click Upload skill.
Upload a .zip of a single skill folder, with SKILL.md at the root of the zip (zip the folder's contents, or the folder such that SKILL.md is at the top level — not nested under extra directories).
Repeat per skill — the uploader takes one skill per zip; there's no bulk "import a directory of skills" path.

Common gotchas when a skill "doesn't show up" in the app:

It's only on your disk. A skill living in .claude/skills/ works in Claude Code but is not synced to the app. The app only knows about skills you've explicitly uploaded.
Symlinks don't survive zipping. If your .claude/skills/<name> entries are symlinks into another directory (a common pattern for sharing one source across tools), zipping them produces a broken link. Package from the real skill folder.
SKILL.md is buried. If the zip expands to myskill/myskill/SKILL.md or the file isn't at the root, the app can't find the manifest and rejects it.
Sandbox off. With Code execution disabled, uploaded skills exist but never trigger.

Notes

CLAUDE.md files are loaded recursively from the repo root and from .claude/rules/*.md.
disableAllHooks: true in settings disables all hooks and the custom status line.
Hook agent type has a 60 s default timeout; prompt type has 30 s; command/http/mcp_tool default to 600 s.
Native installer (curl) automatically updates in the background; Homebrew/WinGet/npm do not.
Path placeholders available in hook commands: ${CLAUDE_PROJECT_DIR}, ${CLAUDE_PLUGIN_ROOT}, ${CLAUDE_PLUGIN_DATA}.

Sources

Topic	URL	Fetched	Label
Hooks reference	https://code.claude.com/docs/en/hooks	2026-06-13	[official]
Overview / installation	https://code.claude.com/docs/en/overview	2026-06-13	[official]
Settings reference	https://code.claude.com/docs/en/settings	2026-06-13	[official]
Canonical docs redirect	https://docs.anthropic.com/en/docs/claude-code → https://code.claude.com/docs	2026-06-13	[official]

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

README.md

Claude Code

Links

Installation

Native installer (recommended — auto-updates)

Homebrew

WinGet

npm (alternative)

Configuration Files

Hooks

Supported Events

Hook Types

Hook Input (stdin JSON)

Exit Code Behavior

Command Hook Fields

HTTP Hook Fields

MCP Tool Hook Fields

JSON Output Fields (stdout from any hook)

Timeout Defaults

Configuration Example

Async Hooks

Matcher Patterns

Built-in Tools

MCP Support

Agent / Subagent Configuration

Permissions

Skills

Filesystem auto-discovery (Claude Code)

Manual upload (Claude Desktop / claude.ai app)

Notes

Sources

FilesExpand file tree

claude-code

Directory actions

More options

Directory actions

More options

Latest commit

History

claude-code

Folders and files

parent directory

README.md

Claude Code

Links

Installation

Native installer (recommended — auto-updates)

Homebrew

WinGet

npm (alternative)

Configuration Files

Hooks

Supported Events

Hook Types

Hook Input (stdin JSON)

Exit Code Behavior

Command Hook Fields

HTTP Hook Fields

MCP Tool Hook Fields

JSON Output Fields (stdout from any hook)

Timeout Defaults

Configuration Example

Async Hooks

Matcher Patterns

Built-in Tools

MCP Support

Agent / Subagent Configuration

Permissions

Skills

Filesystem auto-discovery (Claude Code)

Manual upload (Claude Desktop / claude.ai app)

Notes

Sources