Core Features

Slash Commands

Type / in a session to see available commands¹:

Command	Description
/help	Show available commands
/resume	Pick a previous conversation to continue
/continue	Continue most recent session (also -c flag)
/context	See what’s consuming context space
/memory	View and manage CLAUDE.md files and auto memory
/init	Generate a CLAUDE.md for your project
/agents	Create and manage custom sub-agents
/permissions	View and manage tool permissions
/hooks	Create and manage automation hooks
/mcp	Connect MCP servers
/compact	Manually compress context
/model	Switch between Claude models
/fast	Toggle fast output mode
/status	View session info
/doctor	Diagnose common issues
/cost	Track token usage and costs
/clear	Clear conversation history
/login	Switch accounts mid-session

Bundled skills (invoke like slash commands): /simplify, /batch, /debug, /loop, /claude-api

Custom Slash Commands (Skills)²

Users can create custom commands by placing markdown files in:

• .claude/skills/<name>/SKILL.md — Project-scoped (shared via git)
• ~/.claude/skills/<name>/SKILL.md — Global (personal)

These become available as /<name> and support:

• $ARGUMENTS placeholders for parameters
• !command“ syntax for dynamic context injection (shell output)
• Frontmatter for model, tools, and behavior control
• Supporting files organized in the skill directory

Example:

---
name: fix-issue
description: Fix a GitHub issue
allowed-tools: Bash, Read, Edit, Write, Grep, Glob
---
 
Fix GitHub issue $ARGUMENTS following project standards.
1. Read the issue: !`gh issue view $0`
2. Implement the fix
3. Run tests
4. Commit with descriptive message

Invoke with /fix-issue 123.

Permissions Model

Claude Code has a layered permissions system³ controlling what actions require user approval.

Permission Modes

Mode	Behavior
default	Asks on first use of each tool type
acceptEdits	Auto-accepts file edits, asks for shell commands
plan	Read-only analysis only (no modifications)
dontAsk	Auto-denies unless pre-approved in rules
bypassPermissions	Skips all checks (use with extreme caution)

Permission Rules

Configured in settings files with allow/deny/ask lists:

{
  "permissions": {
    "allow": [
      "Bash(npm run test *)",
      "Bash(git commit *)",
      "Read(~/.zshrc)"
    ],
    "deny": [
      "Bash(git push *)",
      "Read(./.env)",
      "Read(./secrets/**)"
    ],
    "ask": [
      "WebFetch"
    ]
  }
}

Rule Syntax

Pattern	Matches
Bash	All bash commands
Bash(npm run build)	Exact command
Bash(npm run *)	Commands starting with npm run
Bash(* --help)	Commands ending with --help
Read(./.env)	Specific file
Read(/src/**)	Directory patterns
WebFetch(domain:github.com)	Specific domain
Edit(/docs/**)	Edit operations on pattern
Agent(Explore)	Specific sub-agent type
mcp__github__*	All tools from an MCP server

Tool Safety Tiers

Tier	Tools	Default Behavior
Read-only	Glob, Grep, Read	No approval needed
File modifications	Edit, Write	Ask on first use per session
Shell execution	Bash	Ask every time (unless pre-approved)

Memory

CLAUDE.md Files (Explicit Memory)

CLAUDE.md files serve as persistent instructions⁴ loaded automatically at the start of each session:

Location	Scope	Use Case
./CLAUDE.md	Project root	Project conventions, architecture, build commands
.claude/rules/*.md	Project rules dir	Scoped rules (e.g., testing.md, api-design.md)
./CLAUDE.local.md	Project (gitignored)	Personal project overrides
~/.claude/CLAUDE.md	Global	Personal preferences across all projects

What to include:

• Build and test commands (npm test, cargo build)
• Code style conventions (naming, formatting, indentation)
• Architecture overview and key abstractions
• “Never do X” rules
• Frequently used workflows

Best practices:

• Keep under 200 lines (every line costs tokens on every request)
• Use markdown headers and bullets for structure
• Be specific: “Use 2-space indentation” not “Format code properly”
• Move detailed reference material to skills (loaded on-demand)
• Use @path/to/file imports for additional context files
• Scope rules to file types with YAML frontmatter paths: field

Auto Memory

Claude automatically saves learnings across sessions without explicit user action:

• What it saves: Build commands, debugging insights, architecture notes, code style preferences, workflow habits
• Storage: ~/.claude/projects/<project>/memory/MEMORY.md
• Loading: First 200 lines load every session; full directory available on-demand
• Scope: Machine-local only (not shared across devices)
• Control: Enable/disable with /memory command or autoMemoryEnabled setting

Sessions

• Independent: Each new session starts with fresh context (no conversation history from prior sessions).
• Directory-tied: Resume only shows sessions from the current directory.
• Resume: claude --continue resumes the most recent session; claude -r <session-id> resumes a specific one; /resume lets you pick interactively.
• Fork: --fork-session creates a new session with preserved history context.
• Cross-branch: When you switch git branches, files update but conversation history stays.
• Parallel: Use git worktrees to run parallel Claude Code sessions in separate directories.

IDE Integration⁵

VS Code

• Official extension from the Marketplace (search “Claude Code”)
• Embedded chat panel alongside the editor
• @file mentions to reference files
• Inline diffs and plan review
• Resume past conversations
• Full keyboard shortcuts

JetBrains IDEs

• Plugin for IntelliJ, PyCharm, WebStorm, etc.
• Interactive diff viewing
• Selection context sharing
• WSL and remote development support

Desktop App

• Visual diff review
• Multiple sessions side-by-side
• Schedule recurring tasks
• Cloud session management
• Remote Control support

Browser (VS Code Web)

• No local setup needed
• Cloud execution
• /teleport to move between web and terminal

All interfaces share the same CLAUDE.md files, settings, MCP servers, skills, and sub-agents.

Git Integration

Claude Code has deep git awareness:

• Reads git status, diffs, and logs to understand current state
• Creates commits with descriptive messages
• Creates branches and pull requests via gh CLI
• Respects uncommitted work and avoids destructive operations by default
• Supports code review workflows (/review, PR comment analysis)
• Checkpoints every file edit for easy rollback

Chrome Extension

claude --chrome

Enables browser control for web testing, form filling, and debugging web applications.

Footnotes

References

• Claude Code Documentation
• Claude Code Slash Commands
• Claude Code Permissions
• Claude Code Memory (CLAUDE.md)
• Claude Code IDE Integrations
• Claude Code Skills

Footnotes

1. Claude Code Slash Commands ↩

2. Claude Code Skills ↩

3. Claude Code Permissions ↩

4. Claude Code Memory (CLAUDE.md) ↩

5. Claude Code IDE Integrations ↩