The official reference is command-centered now
The Commands page documents built-ins like /clear and /compact alongside bundled skills like /run, /verify, and /simplify.
Overview
Searchable like a command table, readable like a workflow guide
Claude Code now documents built-in commands and bundled skills in one command system. This page puts the “what do I type now,” “where do I extend it,” and “when do I move work into the background” answers first.
30+ command contexts
4 workflow categories
Official docs linked
Recently Checked
The mental model worth remembering first
Running-app verification is now a first-class workflow
/run, /verify, and /run-skill-generator standardize launching, driving, and observing an app instead of relying only on tests.
Think of custom commands as skills
.claude/commands/deploy.md still works, but .claude/skills/deploy/SKILL.md is the more capable structure for new workflows.
Parallel work is split across agents, tasks, background, and batch
/agents opens the manager, /tasks lists background work, /background detaches the current session, and /batch decomposes large changes.
Keep external tools and policy automation separate
MCP prompts run as /mcp__server__prompt, while hooks run automatically on session, turn, and tool-call events.
Command Reference
Search commands
Filter by command, purpose, or category when you need the exact thing to type.
/clearSessionStart a clean conversation
Use this when the old context is no longer useful. The previous thread remains available through /resume.
/compact [instructions]SessionSummarize a long thread to recover context
Keep going in the same task while freeing context. Optional instructions can steer what the summary preserves.
/contextSessionInspect where the context window is going
Use this before compacting to see which parts of the conversation and project state are consuming context.
/btw / /sideSessionAsk a quick side question without bloating the main flow
Useful for small checks or related questions that should not become part of the main work history.
/resume [session]SessionResume a previous session
Choose by session ID, name, or picker. Alias: /continue.
/rewindSessionReturn to an earlier point
Restore a previous conversation or code checkpoint. Aliases include /checkpoint and /undo.
/usageSessionCheck usage, cost, and limits
/cost and /stats are aliases into the usage view.
/configSetupOpen Claude Code settings
The main entry point for theme, model, output style, and other interactive preferences.
/effortSetupAdjust reasoning effort for the current work
Balance speed and depth by changing effort on models that support it.
/model [model]SetupChange model for the current session
Open the model picker or pass a model name. Supported models also expose effort controls.
/permissionsSetupManage tool permission rules
Adjust allow, ask, and deny rules for reads, writes, Bash, and MCP tools. Alias: /allowed-tools.
/doctorSetupDiagnose installation and settings
The first health check to run when your Claude Code environment feels off.
/init / /memorySetupCreate and manage project memory
/init drafts CLAUDE.md. /memory manages memory files and auto-memory entries.
/review [PR]WorkflowReview a PR or local changes
Ask Claude Code to inspect changes from a reviewer’s perspective inside the current session.
/code-review / /ultrareviewWorkflowRun a deeper PR review flow
The official docs prefer /code-review ultra for deep cloud review, with /ultrareview retained as an alias.
/security-reviewWorkflowReview pending changes for security risk
Looks for risks like injection, auth problems, and data exposure in current branch changes.
/simplifyWorkflowSimplify recent changes and improve quality
A bundled skill for reviewing recent files and applying quality, simplicity, and efficiency improvements.
/diffWorkflowInspect current and per-turn diffs
Review the full git diff and the changes made across recent Claude turns.
/run / /verifyWorkflowLaunch the app and verify behavior directly
/run drives the running app, while /verify builds, launches, and observes whether a code change works.
/run-skill-generatorWorkflowSave a project-specific launch recipe as a skill
Use it when standard project inference is not enough and the app needs env files, services, or custom launch steps.
/loop [interval] [prompt]WorkflowRun a prompt repeatedly
Useful for waiting on deployment, polling state, or running a recurring check. Alias: /proactive.
/schedule / /routinesWorkflowCreate routines on Anthropic-managed cloud infrastructure
Describe a recurring routine and Claude walks through creating, updating, listing, or running it.
/agents / /tasks / /backgroundWorkflowManage parallel agents and background work
Open the agent manager, inspect background tasks, or detach the current session so it keeps running.
/batch <instruction>WorkflowSplit large changes across parallel agents
A bundled skill that researches the codebase, decomposes work, then uses isolated worktree agents after approval.
/skills / /skill-nameExtendTurn repeatable playbooks into skills
Personal skills live in ~/.claude/skills/. Project skills live in .claude/skills/. Legacy .claude/commands/ files still work, but skills are the more capable structure for new workflows.
/plugin / /reload-pluginsExtendInstall and reload plugins
Plugins distribute skills, hooks, MCP servers, and other extensions. Use /reload-plugins after plugin changes.
/reload-skillsExtendForce-refresh the skills list
Use this when a newly added or edited skill has not appeared in command discovery yet.
/mcp / /mcp__server__promptExtendManage MCP servers and run MCP prompts
/mcp handles connection status and OAuth. Server prompts run like /mcp__github__pr_review 456.
/hooksExtendInspect lifecycle hooks
Review hooks for events like PreToolUse, UserPromptSubmit, and SessionStart.
No commands found. Try another keyword or the All filter.
Practical Routines
Fast ways to use this in real work
Starting a new repository
Run /init for CLAUDE.md, check access with /permissions, then turn launch steps into a skill with /run-skill-generator.
When a long task gets messy
Use /context and /diff to inspect state, /compact if the direction is right, or /rewind if the path is wrong.
Before opening a PR
Run /code-review for deep review, add /security-review for auth, permissions, or input-handling changes, and use /simplify for quality cleanup.
Splitting work and recurring checks
Use /batch or /agents for large parallel work, /schedule for recurring checks, MCP for integrations, and hooks for hard guardrails.
Official Sources
Jump to the official docs
Built-in commands, bundled skills, MCP prompt commands, and background work commands.
Skill locations, SKILL.md structure, and how legacy .claude/commands/ files fit in.
MCP server setup, prompt command execution, and managed MCP configuration.
Hook events, configuration schema, inputs, outputs, and blocking behavior.