Claude Code Slash Commands Field Guide

Commands are an interface, not magic words

A slash command changes Claude Code's operating state, tool access, workflow, or interface. It does not replace a clear task brief.

Accuracy note. These are Claude Code terminal commands, not universal shortcuts for normal Claude chat. Availability varies by version, platform, plan, plugins, preview access, and environment. Type / inside Claude Code to inspect your exact list.

Three kinds of commands

Core commands

Control the active terminal session, model, context, configuration, tools, and diagnostics.

/help/status/model/compact

Bundled workflows

Package multiple steps into a repeatable review, verification, planning, or automation routine.

/verify/simplify/autofix-pr

Optional commands

Depend on platform, plan, preview access, integrations, or the environment Claude Code runs in.

/chrome/mobile/sandbox

The operator sequence

Inspect

Use /help and /status.

Scope

Define deliverable, files, constraints, and success.

Equip

Set model, permissions, MCP, hooks, and skills.

Execute

Run bounded work with visible checkpoints.

Verify

Inspect the diff, test, and observe the real flow.

The nine command families

Session control

Initialize, reset, resume, and inspect sessions.

/help/status/init/clear/resume/exit

Context and memory

Preserve durable context and compress conversational noise.

/compact/memory/add-dir/export/goal/focus

Models and thinking

Match model and reasoning effort to task complexity.

/model/effort/fast/ultraplan/config/theme/color

Tools and permissions

Connect capabilities, enforce least privilege, and add checks.

/permissions/mcp/hooks/agents/skills/sandbox

Review and verification

Inspect changes, review risks, simplify, and test the real flow.

/diff/security-review/simplify/verify/run/doctor

Agents and automation

Delegate bounded work, observe progress, and keep a stop path.

/tasks/background/loop/schedule/autofix-pr/stop

Apps and remote work

Move work across browser, IDE, desktop, mobile, GitHub, and Slack.

/chrome/ide/desktop/mobile/remote-control/teleport

Account and cloud

Manage authentication, usage, cloud providers, and onboarding.

/login/logout/stats/setup-bedrock/setup-vertex

UI and diagnostics

Tune the terminal and collect evidence when something breaks.

/feedback/insights/keybindings/statusline/tui

Five command sequences that solve real work

Start a repository safely

/status→/init→/permissions→plan→/diff→/verify

Creates durable instructions, scopes access, inspects changes, and verifies the real flow before trust.

Recover a bloated session

/status→/compact focus on current decisions→/memory→/resume

Preserves durable decisions while removing conversational noise.

Add an external tool

/mcp→/permissions→/hooks→/status

Connects the service, limits access, adds deterministic checks, and confirms state.

Review a risky change

/diff→/security-review→/simplify→/verify

Inspects scope, checks security, removes needless complexity, and tests behavior.

Automate recurring work

/skills→/agents→/schedule→/tasks→/stop

Packages the workflow, delegates bounded work, observes execution, and preserves a kill switch.

Production safety checklist

Scope the workspaceGrant added directories deliberately.

Set permissionsGate secrets, destructive commands, production writes, and external sends.

Constrain MCPUse least-privilege credentials and separate read tools from write tools.

Add hooksValidate inputs, lint changes, scan secrets, and log risky tool calls.

Inspect the diffGenerated code is not evidence that behavior is correct.

Verify the real flowRun tests, open the app, reproduce the workflow, and inspect side effects.

Keep a stop pathBound retries and expose a kill switch for background work.

Export decisionsPreserve context the next operator must inherit.

Official references

Official Claude Code command reference → Claude Code overview → Memory and CLAUDE.md → Model Context Protocol → Hooks → Permissions →

Build a Claude Code operating system

A command is useful once. A configured system of memory, skills, tools, permissions, hooks, and verification compounds across every run.

Book a GTM strategy call →