Module 4 — Built-in slash commands (1/2): session, context, configuration and model

The first three modules laid the foundations. This module and the next are the reference: every command, its arguments, its aliases, its role, when to type it. Module 5 covers review, parallelism, automation and integrations; here, eight families centered on the session — help, conversation, context, appearance, model, memory, permissions, account.

Command grammar

A command is only recognized when it starts your message. What follows becomes its arguments. Since v2.1.199, skills are an exception: /skill-a /skill-b do XYZ loads each skill up front and passes the rest to each of them, up to six chained. In the tables, <arg> marks a required argument, [arg] an optional one.

A command typed while Claude is replying is queued; some run immediately — /status, /tasks, /usage — and in full-screen rendering (v2.1.234+), dialogs like /theme or /help open without waiting. /model, /effort, /permissions and /add-dir now take effect on the current turn after confirmation.

Three kinds of entries in the / menu: built-in commands coded into the CLI, bundled skills (Skill, a prompt handed to Claude) and bundled workflows (Workflow, several subagents in the background). Module 5 covers skills and workflows; here, built-ins only. Type / followed by a few letters to filter: highlighting ignores the separators :, _, - (/adddir highlights /add-dir, /new highlights /clear via its alias).

(a) Help and state

Command	Arguments	Aliases	Role	When to type it
/help	—	—	Shows help and the list of available commands	On your first session, or when a name escapes you
/status	—	—	Opens the Settings UI on the Status tab (version, model, account, connectivity, session type)	Check you are on the right model and the right account
/doctor	—	/checkup	Diagnostic: installs, PATH, settings, unused skills/MCP, slow hooks, oversized CLAUDE.md	Strange behavior, monthly hygiene
/release-notes	—	—	Changelog picker; renders in the transcript without polluting the context	After an update
/powerup	—	—	Interactive lessons with animated demos	Onboarding
/insights	—	—	HTML report on your recent sessions: projects, usage patterns, sticking points	Personal quarterly retro
/team-onboarding	—	—	Markdown onboarding guide built from the last 30 days; shareable link (Pro, Max, Team, Enterprise)	New hire in the team

/doctor deserves a special mention: since v2.1.206, it slims down a committed CLAUDE.md by cutting what Claude can infer from the code and proposes to migrate the rest into skills or nested CLAUDE.md files. From a terminal, claude doctor (no slash, no session) prints a read-only diagnostic.

(b) Session and conversation

Command	Arguments	Aliases	Role	When to type it
/clear	[name]	/reset, /new	New conversation with empty context; the name shows up in /resume	Between two unrelated tasks
/resume	[session]	/continue	Resume a conversation by ID or name, or open the picker	Pick up the morning's thread
/rename	[name]	—	Renames the session; auto-named without an argument	Pin a long-running session
/branch	[name]	—	Creates a branch of the conversation here and switches to it; the original stays available via /resume	Explore a tangent without losing the thread
/fork	[prompt]	—	Copies the conversation into a background session and stays here; v2.1.212+	Delegate a long task in parallel
/subtask	<task>	—	Forked subagent that inherits the conversation; the result flows back here; v2.1.212+	Side question that needs real work
/export	[filename]	—	Exports the conversation as plain text; file or clipboard	Archive, send to a colleague
/copy	[N]	—	Copies the Nth reply; code-block picker; w writes a file	Paste a patch into an external editor
/recap	—	—	One-line summary on demand	Resuming after a break
/rewind	—	/checkpoint, /undo	Rewinds code and/or conversation, or summarizes from a message	An edit broke the build
/exit	—	/quit	Quits the CLI; detaches an attached background session without stopping it	End of day
/stop	—	—	Stops the attached background session; transcript and worktree are kept	A /fork that no longer converges

Three useful nuances: /branch switches you here, /fork copies elsewhere, /subtask delegates to a subagent that reports back. /rewind checkpoints do not cover shell commands run or files outside the repo — see module 8.

(c) Context and cost

Command	Arguments	Aliases	Role	When to type it
/context	[all]	—	Context-occupation grid, saturation alerts, suggestions; all expands each item	Before a big task, or when Claude "forgets"
/compact	[instructions]	—	Summarizes the conversation; optional directive to target what to keep	Long sessions, before a new subtask
/autocompact	[auto|<tokens>]	—	Auto-compaction threshold (500k, auto); saved in settings; v2.1.221+	Multi-hour sessions
/btw	[question]	—	Side question that does not pollute history; without an argument, last reply (v2.1.212+)	"what was that env var again?"
/usage	—	/cost, /stats	Session cost, plan limits, statistics; /stats opens the Stats tab	End of task, end of day
/usage-credits	—	—	Usage credits in the browser; sends a request to your admin on Team/Enterprise; formerly /extra-usage	When the plan is saturated

/compact takes a directive: /compact keep only the payments plan and drop the explorations is infinitely more useful than a blind compact. What survives every compaction: CLAUDE.md, rules without paths: and the active skill — Claude re-reads them from disk on the next turn.

(d) Configuration and appearance

Command	Arguments	Aliases	Role	When to type it
/config	[key=value ...]	/settings	Opens Settings, or applies a key (/config theme=dark, model=sonnet) since v2.1.181; /config --help lists keys	Change a preference without leaving the session
/theme	—	—	Color theme: auto, light, dark, colorblind, ANSI, themes from ~/.claude/themes/ or a plugin	On install or for a colorblind theme
/tui	[default|fullscreen]	—	Terminal renderer; fullscreen = full-screen without flicker	Wide screen, long session
/focus	—	—	Focus view: prompt, one-line summary of tool calls, response; full-screen only	Screen share, quick review
/color	[color|default]	—	Prompt-bar color: red, blue, green, yellow, purple, orange, pink, cyan; without an argument, random	Distinguish several parallel sessions
/statusline	—	—	Configures the status line; auto-detected from the shell prompt	Once per machine
/keybindings	—	—	Opens the keyboard-shortcut file	Customize Esc, Shift+Enter, etc.
/scroll-speed	—	—	Wheel speed; full-screen only, not in the JetBrains terminal	Miscalibrated mouse
/terminal-setup	—	—	Shift+Enter for VS Code, Cursor, Devin Desktop, Alacritty, Zed; Option+Enter and silenced bell on Apple Terminal; clipboard for /copy on iTerm2	Once per terminal
/voice	[hold|tap|off]	—	Voice dictation; requires a claude.ai account	Long spoken descriptions

(e) Model and reasoning

Command	Arguments	Aliases	Role	When to type it
/model	[model]	—	Change model and save as default; without an argument, opens a picker; s switches for the session only; left/right arrows adjust effort	Move from opusplan to sonnet before execution
/effort	[level|auto|status]	—	Sets the effort: low, medium, high, xhigh, max, ultracode, auto; status displays it; max and ultracode are session-only	Bump to high for a tough debug
/fast	[on|off]	—	Toggles fast mode; applies without waiting for the end of turn since v2.1.242; requires v2.1.205+	Small repetitive tasks with low latency
/advisor	[model|off]	—	Enables an advisor model consulted at key moments (fable, opus, sonnet, or a full ID); without an argument, opens a picker	Architectural decisions, plan reviews

A model or effort change invalidates the prompt cache; Claude Code prints the warning before applying. A good habit: plan in opusplan, execute in sonnet, set the effort once for the whole session.

(f) Memory and project

Command	Arguments	Aliases	Role	When to type it
/init	—	—	Generates a starter CLAUDE.md; CLAUDE_CODE_NEW_INIT=1 triggers an interactive flow that also covers skills, hooks, personal memory and imports an existing config from another agent	First session in a repo
/memory	—	—	Edits CLAUDE.md and CLAUDE.local.md at every scope, including those that do not exist yet; toggle and audit of automatic memory	Every time you correct Claude twice
/add-dir	<path>	—	Adds a working directory for the current session; autocomplete suggests paths; .claude/ configurations from the added folder are not loaded	Monorepo, sibling repos next to each other
/cd	<path>	—	Moves the session into another working directory without losing the conversation; autocomplete since v2.1.206	Switch subproject without restarting

/add-dir and /cd do not serve the same purpose: /add-dir extends the accessible surface without changing the root; /cd moves the root and reloads the new folder's CLAUDE.md.

(g) Permissions

Command	Arguments	Aliases	Role	When to type it
/permissions	—	/allowed-tools	Interactive dialog to manage allow/ask/deny rules per scope, working directories, recent refusals in auto mode, and auto-mode classifier rules	After a refusal that gets in the way, before a headless run
/sandbox	—	—	Toggles sandbox mode (supported platforms only)	Run untrusted code, brand-new project
/fewer-permission-prompts	—	—	Bundled skill: scans your transcripts for common read-only Bash and MCP calls, then adds a prioritized allowlist to the project's .claude/settings.json	After a week of repeated "ask" prompts

Module 7 walks through rule patterns (Bash(...), Edit(...), WebFetch(domain:...), mcp__server__tool), the default, acceptEdits, plan, auto, dontAsk, bypassPermissions modes, and the precedence of settings files.

(h) Account

Command	Arguments	Aliases	Role	When to type it
/login	—	—	Sign in to your Anthropic account	First time, or after a /logout
/logout	—	—	Sign out of your Anthropic account	Shared machine
/privacy-settings	—	—	View and change your privacy settings; reserved for Pro and Max plans	Retention, telemetry
/upgrade	—	—	Opens the upgrade page in the browser; hidden for Enterprise plans	End of Free plan, need Max
/passes	—	—	Share a free week of Claude Code; visible only for eligible accounts	Recommend to a colleague
/mobile	—	/ios, /android	QR code to install the Claude mobile app	Want to drive from a phone
/desktop	—	/app	Continue the current session in the Desktop app; macOS and Windows x64 with a Claude subscription	Move from terminal to desktop window

The ten commands to know by heart

Over hundreds of hours of Claude Code, ten commands come back on loop. Knowing them by heart saves half a day per week:

1. /init — on the first launch in a repo.
2. /memory — as soon as you correct Claude twice in a row.
3. /model and /effort — plan in opusplan, execute in sonnet, set effort by difficulty.
4. /context — before every big task, to see the room you have left.
5. /compact <directive> — the directive makes all the difference.
6. /clear — new unrelated task, clean context.
7. /permissions — after a refusal, before a non-interactive run.
8. /status and /doctor — one checks state, the other fixes it.
9. /rewind — the insurance policy that lets you dare a refactor.
10. /usage — end of session; you learn fast what costs what.

Running example: a Kiosque day told through commands

A typical day of the learner on Kiosque, written only in commands typed:

• 09:02 — claude in the repo, /status (right account, Sonnet active).
• 09:03 — /resume kiosque-paiements to pick up yesterday's work; /recap shows the auto summary.
• 09:05 — /context: 47% used, mostly by paiements.py. /compact keep the plan and the error analysis, drop the code snippets read.
• 09:20 — new lead: refactor notifications.py in TDD. /branch notifications-tdd.
• 09:22 — /model opusplan, /effort high; Claude proposes a plan.
• 09:45 — approved, switch to execution: /model sonnet, /effort medium. Refuse .env, approve Edit(tests/**).
• 10:30 — a test fails for lack of allowlist. /permissions adds Bash(pytest -k *) as project allow.
• 11:15 — Lea drops in for review. /focus for the screen share.
• 12:40 — /btw how do I regenerate the test Stripe key? — the answer stays out of history.
• 14:10 — wrong lead. Esc Esc then /rewind restores three messages earlier.
• 16:30 — /subtask run make lint on app/ and bring back the top 20 warnings — a subagent runs while notes are being written.
• 17:20 — overnight run: /fork chain the Alembic migrations and restart the API.
• 17:45 — /usage (37,000 cache-read tokens, 4,200 write, cost under threshold), /rename kiosque-notifications-tdd, /export notes/2026-09-09.md, /exit.

Summary

• Grammar: command at the start of the message, arguments after; up to six chainable skills; some commands run immediately, others wait for the end of turn.
• Three kinds of entries in the / menu: built-ins, bundled skills (Skill), bundled workflows (Workflow).
• Context: /context to see, /compact (with a directive) to free up, /autocompact to automate the threshold.
• Session: /branch switches here, /fork copies elsewhere, /subtask delegates to a subagent that reports back.
• Cost: /model, /effort, /fast are your three levers; /usage measures; changing model or effort invalidates the cache.
• Memory: /init then /memory are your two everyday commands; /add-dir extends, /cd moves.
• Permissions: /permissions, /sandbox and /fewer-permission-prompts; patterns and modes are covered in module 7.
• Ten commands to know by heart cover 90% of sessions.

Next module: Built-in slash commands (2/2): quality, parallelism, automation and integrations — /diff, /code-review, /security-review, /tasks, /batch, /loop, /schedule, /mcp, /plugin and the IDE, web, Chrome integrations.