Session & Memory
/init
Initialize project with
CLAUDE.md. Creates your project's persistent instruction file read at every session start.▸ example
/init
◈ output
✓ CLAUDE.md created (18 lines)
Detected: Node.js · TypeScript · Jest
Added: project structure, test cmds, style notes
Claude will read this at every session start.
/memory
View and manage auto-memory files. Use
# prefix in the REPL to add things to memory without editing files.▸ example
/memory
◈ output
Memory: ~/.claude/projects/.../memory/
user_role.md senior fullstack dev
feedback_testing.md run tests before commit
project_context.md auth refactor (compliance)
[E]dit [D]elete [N]ew
/resume
Resume a previous session.
/resume 1 for most recent, /resume name by name. Sessions are resumable up to 7 days.▸ example
/resume auth-refactor
◈ output
Resuming: "auth-refactor" (2h ago · 34 msgs)
Last action: edited src/auth/middleware.ts
Context restored. Continue where you left off.
/rename
Name the current session for easy resuming later.
▸ example
/rename auth-refactor
◈ output
Session renamed → "auth-refactor"
Resume later with: /resume auth-refactor
/clear
Clear conversation history. Keeps files and memory intact — only the chat context is wiped.
▸ example
/clear
◈ output
Conversation cleared (47 messages removed).
Files, memory, and tool state are intact.
/compact
Condense conversation history to free up context. Optionally focus:
/compact focus on tests▸ example
/compact focus on the failing tests
◈ output
Compacted: 112 messages → 14 (87% reduction)
Summary anchored on: test failures in auth.test.ts
Context freed: ~82k tokens
/context
View context window usage with actionable suggestions on when to compact.
▸ example
/context
◈ output
Context window: ██████████░░░░░░░ 58%
Used: 116,842 / 200,000 tokens
System + tools: 18k
Conversation: 74k
Files in scope: 24k
⚡ Suggestion: /compact when above 75%
/export
Export the conversation transcript to a file.
▸ example
/export
◈ output
✓ Transcript exported
File: claude-session-2026-07-18-14-32.md
Messages: 47 · Code blocks: 12
/rewind
Return to a prior checkpoint. Also triggered by pressing Esc Esc.
▸ example
/rewind
◈ output
Reverted to checkpoint before last response.
3 file edits undone:
src/auth/middleware.ts
src/auth/types.ts
tests/auth.test.ts
/branch
Branch the conversation for parallel exploration without affecting your main session.
▸ example
/branch
◈ output
Branched from turn 23. Branch ID: br_a4f2c9
Explore freely — main session is unaffected.
Merge back or discard when done.
Model & Effort
/model
Switch model mid-session. Options:
sonnet (default), opus, haiku, fable. Also Alt+P while typing.▸ example
/model opus
◈ output
Model → claude-opus-4-8
Effort: medium
Best for: complex reasoning, agentic tasks
/effort
Set reasoning effort:
low, medium, high, xhigh. Use xhigh with Opus for hardest agentic work.▸ example
/effort xhigh
◈ output
Effort → xhigh
Claude will reason more deeply before responding.
Recommended pairing: /model opus
/fast
Toggle fast output mode — uses Opus with optimized output speed.
▸ example
/fast
◈ output
Fast mode ON
Using Opus 4.8 with optimized output speed.
Toggle off: /fast
/usage
View token usage, cost, and plan limits.
/cost and /stats are shortcuts that open specific tabs here.▸ example
/usage
◈ output
Session usage
Input tokens: 142,330
Output tokens: 18,440
Cache hits: 89,200 (63%)
Estimated cost: $0.38
Plan: Pro · Resets: Aug 1
/status
View current session state: active model, permission mode, and settings summary.
▸ example
/status
◈ output
Model: claude-sonnet-4-6
Mode: acceptEdits
Effort: medium · Fast: off
Session: "auth-refactor" (2h 14m)
Working dir: /Users/me/my-project
Autonomous Work
/loop agentic
Run a prompt or command on a repeating interval.
/loop 3h check PRs, /loop 5m /my-skill, or bare /loop for self-paced maintenance. Stop with /loop stop.▸ example
/loop 5m check for new failing tests
◈ output
Loop started ↺
Interval: every 5 minutes
Task: check for new failing tests
Wakeup #1: 14:47
Stop with: /loop stop or Ctrl+C
/goal v2.1.139+
Set a completion condition; Claude keeps going autonomously until the condition is met. Uses a fast model to evaluate after each turn.
▸ example
/goal all tests pass
◈ output
Goal set: "all tests pass"
Running autonomously...
✗ Turn 1: 3 tests failing in auth.test.ts
→ Fixing middleware null check...
✓ Turn 2: All 47 tests passing
Goal reached in 2 turns. Stopping.
/plan
Enter plan mode — Claude outlines the approach before touching any files. Best before large refactors.
▸ example
/plan refactor the auth module
◈ output
Plan: Refactor auth module
Phase 1 — Audit (read-only)
• Read src/auth/*.ts (6 files)
• Identify session token issues
Phase 2 — Restructure
• Split middleware.ts → validator + session
• Update imports in 4 consumers
Phase 3 — Test: jest --testPathPattern=auth
Approve to begin → [Y/n]
/batch
Run batch operations across multiple files or targets in parallel.
▸ example
/batch add JSDoc to all files in src/utils/
◈ output
Batch: 8 files in src/utils/
✓ formatDate.ts — 3 functions documented
✓ validators.ts — 7 functions documented
✓ http.ts — 4 functions documented
... (5 more)
Completed in 12s · 0 errors
/tasks
List all background agents currently running in this session.
▸ example
/tasks
◈ output
Background agents (2 running)
[1] ● pr-reviewer Running 1m 34s
[2] ● test-watcher Running 4m 02s
/tasks stop 1 to stop agent #1
/bashes
List background bash tasks spawned by Claude.
▸ example
/bashes
◈ output
Background bash tasks (1 running)
[1] ● npm run build Running 0m 48s PID 83421
Output: /tasks output 1
Stop: /tasks stop 1
Tip:
/loop vs /goal — use /loop when you want Claude to wake up on a schedule. Use /goal when you want Claude to keep working until a condition is true.
Code Quality
/code-review
Find correctness bugs. Supports effort levels:
/code-review medium, /code-review ultra. Use --fix to auto-apply, --comment to post inline GitHub PR comments.▸ example
/code-review --fix
◈ output
Reviewing diff (14 files, +342/-89 lines)...
● HIGH src/api/users.ts:42 SQL injection via userId
● MED src/auth/session.ts:18 Missing null check
○ LOW src/utils/format.ts:7 Unnecessary re-render
Applying 2 fixes (HIGH + MED)...
✓ Fixed. Review LOW finding manually.
/security-review
Review code specifically for security vulnerabilities — OWASP Top 10, secrets in code, injection risks.
▸ example
/security-review
◈ output
Security review: 23 files on current branch
CRITICAL src/api/upload.ts:31 Path traversal
HIGH src/auth/tokens.ts:14 Token in localStorage
HIGH .env.example Real API key — rotate!
LOW src/utils/log.ts:8 Email in plaintext logs
4 findings · 1 critical · 2 high · 1 low
/claude-api
Get help building apps with the Anthropic SDK — tool use, batch processing, MCP integration, streaming, prompt caching.
▸ example
/claude-api
◈ output
Claude API Reference loaded. Ask me about:
• Model IDs, pricing, rate limits
• Tool use / function calling
• Streaming responses
• MCP server setup & tool definitions
• Batch API & prompt caching
Parallel & Agents
/agents
Open Agent View — an operations screen for running, blocked, and completed Claude Code sessions. Up to 10 subagents can run in parallel.
▸ example
/agents
◈ output
Agent View ─────────────────────────
[1] ● pr-reviewer Running 2m 11s
[2] ● test-watcher Running 8m 03s
[3] ○ docs-builder Complete --
[S]top [V]iew output [N]ew agent
/skills
List installed skills with a type-to-filter search box. Skills are prompt-based commands stored in
.claude/skills/.▸ example
/skills
◈ output
Installed skills (6)
/deep-research Multi-source cited research report
/code-review Review diff at configurable effort
/security-review OWASP-focused security audit
/run Launch and test the app live
/verify Verify a change works in the app
/pr-review Review open PRs [custom]
/add-dir
Expand file access by adding additional working directories to the session.
▸ example
/add-dir ../shared-utils
◈ output
✓ Added to session scope:
/Users/me/shared-utils (47 files · 12.3k lines)
Claude can now read and edit files here.
Config & Permissions
/config
Open the full settings interface interactively.
▸ example
/config
◈ output
Settings ────────────────────────────
Model claude-sonnet-4-6
Effort medium
Theme dark
Permission acceptEdits
Auto-memory enabled
Hooks 3 configured
MCP servers 2 active
[Tab] navigate [Enter] edit [Q] quit
/permissions
Manage permission settings interactively — what Claude can read, edit, and execute.
▸ example
/permissions
◈ output
Permissions ────────────────────────
✓ Read files all paths
✓ Edit files all paths
✓ Bash: git * allowed
✓ Bash: npm test allowed
✗ Bash: rm -rf blocked
[A]dd [R]emove [M]ode [Q]uit
/hooks
View current hook configuration. Hooks fire on events like
PreToolUse, PostToolUse, Stop, etc.▸ example
/hooks
◈ output
Hooks (3 configured)
PreToolUse → lint check before file edits
PostToolUse → run tests after Bash(git commit *)
Stop → notify Slack #dev-alerts
Source: .claude/settings.json
/mcp
Configure MCP servers.
/mcp enable, /mcp disable. MCP expands Claude's tool access to external services.▸ example
/mcp enable github
◈ output
MCP server "github" enabled.
Tools added: list_repos, create_pr,
get_issue (+12 more)
Auth: OAuth (already connected)
Active MCP servers: github, slack
/doctor
Full setup checkup — diagnoses and fixes environment issues. Alias:
/checkup▸ example
/doctor
◈ output
Claude Code Health Check ───────────
✓ CLI version 2.1.205 (latest)
✓ Node.js v20.11.0
✓ Auth logged in (anandhu-eng)
✓ Git 2.43.0
✗ MCP: jira unreachable (timeout)
✓ CLAUDE.md found (18 lines)
1 warning · fix: /mcp disable jira
/bug
Report a bug directly to Anthropic from within your session.
▸ example
/bug
◈ output
Report a bug to Anthropic
Session ID: sess_7f3a2b
Version: 2.1.205
Describe the issue:
> _
[Submit] [Attach transcript] [Cancel]
/login & /logout
Authenticate or sign out mid-session without restarting Claude Code.
▸ example
/logout
◈ output
Signed out of claude.ai (anandhu-eng).
Token cleared from keyring.
Sign back in: /login
Or restart: claude --login
Keyboard Shortcuts
Ctrl+CCancel current operation
Ctrl+DExit session (EOF)
Ctrl+LClear screen (keeps history)
Ctrl+BBackground current operation
Ctrl+X Ctrl+KStop all agents
Ctrl+SStash prompt draft
Ctrl+GOpen external editor
Ctrl+VPaste image from clipboard
Ctrl+OToggle verbose output
Ctrl+RSearch command history
Esc EscRewind last change
Shift+TabCycle permission modes
Alt+PSwitch models while typing
Alt+TToggle thinking mode
TabAccept prompt suggestion
! TabBash history autocomplete
?Show all shortcuts
Quick Prefixes
#
Persistent Memory
# Always use TypeScript
/
Slash Command
/code-review --fix
!
Execute Bash
! git status
@
Reference File
@src/index.ts
&
Send to Cloud
& Build the API
Custom Commands
Create a
.md file in .claude/commands/ (project) or ~/.claude/commands/ (personal) and invoke it with /filename. Use $ARGUMENTS for arguments.--- # .claude/commands/pr-review.md
description: Review open PRs and post GitHub comments
allowed-tools: Bash(gh pr list *), Bash(gh pr diff *), Bash(gh pr review *)
model: opus
---
1. Run `gh pr list --state open --json number,title`
2. For each PR, run `gh pr diff <number>`
3. Review for: breaking API changes, missing error handling, security issues
4. Post: `gh pr review <number> --comment -b "<feedback>"`
Usage:
/pr-review — or schedule it: /loop 3h /pr-reviewPermission Modes
manual (default)
Prompts on first use of each tool
Normal development
acceptEdits
Auto-approves file edits, prompts for bash
Trusted projects
auto
Safety classifier reviews each action automatically
Autonomous with guardrails
plan
Read-only — no edits or execution
Analysis only
bypassPermissions
Skips all prompts entirely
CI/CD automation
Cycle modes with Shift+Tab. Set at startup:
claude --permission-mode auto