Here are 42 commonly used Claude Code slash commands, grouped by category, each with a description, when to use it and an example. Based on the official Commands documentation, with notes on how Claude Code on the web (claude.ai/code) and the Claude desktop app differ: neither shows the terminal’s interactive pickers, so some commands need an argument. Availability also depends on your plan, Claude Code version and organization policy — send /help in a session to see what is actually supported.
/compact /context /btw /recap /autocompact /clear
VibeSlash/compact [what to keep]Compacts the context by summarizing the conversation so far, freeing up context space while the session continues.
Use when the conversation is long, responses slow down, or you are warned that context is nearly full; optionally specify what the summary should keep.
/compact keep test output and open issues/context [all]Shows current context usage as a colored grid, with optimization suggestions.
Use to see what is filling the context (tool output, memory files, MCP, etc.); add all for an item-by-item breakdown.
/context all/btw [question]Asks a side question without adding the question or answer to the conversation history.
Use to quickly check something without interrupting or polluting the context of the current task.
/btw which test framework does this repo use?/recapImmediately generates a one-sentence summary of the current session.
Use to catch up quickly after stepping away, or to bring someone else up to speed.
/recap/autocompact [auto|<size>]Sets the auto-compact window: how much context can be used before it is compacted automatically.
Pass a size such as 500k, or auto to restore the model default; saved to user settings.
/autocompact 500k/clear [name]Starts a new conversation with empty context.
Use when switching to an unrelated task; to free up space and keep working on the current task, use /compact.
/clear/model /effort /fast /plan /advisor /goal /output-style
/model <model>Switches the model.
Give a model alias (e.g. opus, sonnet, haiku) or a full model ID.
/model sonnet/effort <low|medium|high|xhigh|max|auto>Adjusts reasoning effort; status shows the current value, and ultracode on/off toggles ultracode.
Raise it for complex problems and lower it for simple changes to go faster; max applies only to the current session.
/effort high/fast [on|off]Toggles fast mode for faster output.
Requires fast mode to be enabled for your account.
/fast on/plan [task description]Enters plan mode, where Claude analyzes and proposes a plan before changing any code once you approve.
Use before large changes; you can attach a task description to start right away in plan mode.
/plan refactor the login authentication flow/advisor [model|off]Turns the advisor tool on or off, which consults a second model at key points.
The argument can be fable, opus, sonnet, or a full model ID; off turns it off.
/advisor opus/goal [completion condition|clear]Sets a goal that Claude keeps working toward across multiple turns until the condition is met.
Suited to tasks with a clear completion criterion, such as "keep fixing until all tests pass"; clear cancels early, and no argument shows the current goal.
/goal all unit tests pass and lint reports no errors/output-style [style]Lists or switches output styles.
With no argument, lists available styles; requires Claude Code 2.1.269 or later.
/output-style concise/code-review /security-review /simplify /verify /run
/code-review [effort] [--fix] [--comment] [PR number|branch|path]Reviews the current changes (or a given PR, branch, or path) for correctness bugs.
Effort ranges from low to max; --fix applies fixes directly, --comment posts results to the PR, and ultra runs a multi-agent deep review in the cloud.
/code-review high --fix/security-reviewChecks the current branch's changes against the default branch for security vulnerabilities (injection, auth, data exposure, etc.).
Use before merging; the repository needs an origin remote.
/security-review/simplify [target]Reviews changes for reuse, simplification, efficiency, and abstraction-level issues and fixes them directly (does not look for bugs).
Use to tidy up code after finishing a feature and before committing; you can specify a path or PR. To find bugs, use /code-review.
/simplify src//verifyActually builds and runs the app to confirm a change works, rather than relying only on tests or type checks.
Use after changing the UI or runtime behavior when you need to confirm the real effect.
/verify/runLaunches and drives the project's app to show the change in action.
Use to see a page or service running, or to take screenshots.
/run/loop /batch /subtask /schedule /tasks /deep-research
/loop [interval] [prompt]Repeats a prompt at an interval while the session stays open; without an interval, Claude sets its own pace.
Use to watch deployments, poll CI, or check status periodically.
/loop 5m check whether the deployment has finished/batch <instruction>Makes large-scale parallel changes by splitting work into 5 to 30 independent units, each completed by a background subagent in its own worktree.
For repo-wide migrations and bulk refactors; Claude presents a plan for your approval first.
/batch migrate src/ from JavaScript to TypeScript/subtask <task>Sends a background subagent that inherits the full conversation to handle a side task, returning its result to the current conversation when done.
Keep the main thread moving while a subagent investigates or handles a side issue.
/subtask investigate why integration tests fail intermittently on CI/schedule [description]Creates, edits, lists, or runs scheduled tasks (Routines) that execute in the cloud.
For checks or reports that run daily or weekly; Claude guides you through setup conversationally.
/schedule summarize unmerged PRs every weekday at 9 a.m./tasksShows and manages the current session's background tasks, including completed subagents.
Use to see which commands or subagents are still running in the background.
/tasks/deep-research <question>Searches the web in parallel, cross-checks sources, and produces a research report with citations.
Use for multi-source research or comparing options; runs in the background.
/deep-research compare cold-start speed of popular frontend build tools/init /update-config /fewer-permission-prompts /doctor /debug /reload-skills /mcp /config
/initAnalyzes the project and generates an initial CLAUDE.md project guide.
Run the first time you use Claude Code in a repository, then refine it as needed.
/init/update-config [request]Takes a settings change described in natural language (allow a command, set an environment variable, add a hook) and has Claude edit settings.json.
For options like theme or model, use /config or /model.
/update-config stop asking before running npm test/fewer-permission-promptsScans history for common read-only commands and adds them to the allowlist in the project's .claude/settings.json.
Use when permission prompts appear too often.
/fewer-permission-prompts/doctor [prompt-audit [path]]Checks for problems with the installation, settings, extensions, and CLAUDE.md, and has Claude fix them after you confirm.
The prompt-audit subcommand reviews how instruction files are written instead.
/doctor/debug [problem description]Enables debug logging for the current session and reads the logs to troubleshoot.
Use when Claude Code itself misbehaves; logging starts from when this command is run.
/debug MCP server keeps failing to connect/reload-skillsRescans skill and command directories so skills added or changed during the session are available immediately.
Use after writing new .claude/skills or .claude/commands in the repository.
/reload-skills/mcp [reconnect|enable|disable <server|all>]Manages MCP server connections and authentication.
reconnect all reconnects every server that failed or is awaiting authentication.
/mcp reconnect all/configOpens Claude Code settings.
For cloud session settings, use environment variables or a .claude/settings.json committed to the repository instead.
/config/rename /color /usage /teleport /export /help
/rename <name>Renames the current session.
Makes it easier to find in the session list.
/rename fix order sync/color <color|default>Sets the prompt bar color for the current session.
Options are red, blue, green, yellow, purple, orange, pink, and cyan; default restores the default.
/color purple/usageShows this session's cost, plan usage, and activity stats.
Use to see how close you are to your usage limit.
/usage/teleportGives you a claude --teleport <session-id> command to take over this cloud session in a local terminal.
Use to continue a cloud session locally; run the command in a local checkout of the repository.
/teleport/export [file name]Exports the current conversation as plain text, writing directly to a file if a file name is given.
Use to keep a record or hand off work.
/export notes/session.txt/helpShows help and available commands.
Use to see which commands the current environment actually supports.
/help/claude-api /dataviz /design /slides
/claude-api [subcommand]Loads Claude API and Managed Agents reference material (models, SDK usage, caching, etc.).
Use when building apps that call Claude; subcommands include migrate (migrate model versions), cost-optimize, and build-eval.
/claude-api migrate/dataviz [request]Provides design guidance for charts and dashboards: choosing chart forms, assigning colors by role, and checking color-blind safety and contrast.
Use before asking Claude to draw charts or build data dashboards.
/dataviz a line chart of monthly order volume/design [request]Turns UI sketches, page flows, or posters into an editable Claude Design canvas.
Requires a session that supports Artifacts and an account with the Design template enabled.
/design settings page for a mobile banking app/slides [request]Generates a Claude Slides presentation from a brief.
Requires a session that supports Artifacts and an account with the Slides template enabled.
/slides platform team quarterly reviewClaude and Claude Code are trademarks of Anthropic, PBC. Codex and ChatGPT are trademarks of OpenAI. DeepSeek is a trademark of its respective owner. VibeSlash is an independent, unofficial open-source project and is not affiliated with, endorsed by, sponsored by or authorized by Anthropic, OpenAI or DeepSeek. These names are used only to describe compatibility; all trademarks belong to their respective owners.