Vibecoding.cz Patrick Zandl o AI vývoji

Claude Code Cheat Sheet

Dense, current, and practical: commands, settings, skills, hooks, MCP, worktrees, routines, and setup paths in one place.

Verified against official Anthropic / Claude Code docs on September 5, 2026 (v2.1.261).
What’s new → Settings →
Language
Install

Node 18+, then install and run Claude in your project.

npm install -g @anthropic-ai/claude-code
Sign in

Interactive login is the normal path. API keys override subscriptions when present.

/login
Resume

Continue the latest session or resume a specific session by ID.

claude -c
Permissions

Use /permissions or settings.json to control allow / ask / deny behavior.

/permissions
MCP

Add tool servers with claude mcp add, then inspect them with /mcp.

claude mcp add ...
Use this page like a board, not a tutorial Use the left navigation on desktop, search at the top, and jump directly to the raw command or key you need.

Install & first run

Official install paths as of September 5, 2026:

CommandDetails
Standard NPM install
npm install -g @anthropic-ai/claude-code 		Official standard install path. Avoid sudo.
Native install (beta)
curl -fsSL claude.ai/install.sh | bash 		macOS, Linux, and WSL beta install path.
Windows PowerShell
irm https://claude.ai/install.ps1 | iex 		Native Windows PowerShell install path.
Health check
claude doctor 		Verify installation type and fix common issues.

High-signal starter prompts

Good first prompts to keep Claude useful and scoped:

Codebase orientation

Explain this project architecture, key entry points, build/test commands, and the files I should read first.

Safe implementation start

Before coding, inspect the repo, identify the relevant files, propose the smallest safe plan, then implement it and run verification.

Targeted bug hunt

Investigate this failure, find the root cause, patch it with the smallest change that fixes it, and explain the regression risk.

CLI launch patterns

Launch-time patterns worth memorizing:

CLI launch patterns

CommandPurposeCategory
claude Start interactive REPL Interactive
claude "query" Start REPL with an initial prompt Interactive
claude -p "query" One-shot query, then exit Non-interactive
cat file | claude -p "query" Pipe content into print mode Non-interactive
claude -c Continue the most recent conversation Session
claude -c -p "query" Continue via SDK / print mode Session
claude -r "<session-id>" "query" Resume a specific session by ID Session
claude update Update Claude Code to latest version Maintenance
claude mcp Configure MCP servers from the CLI MCP
claude agents List all sessions — running, blocked, or done (Agent View, research preview) Agents
claude project purge [path] Delete all Claude Code state for a project (transcripts, tasks, file history, config); supports --dry-run Maintenance
claude ultrareview [target] Run /ultrareview non-interactively from CI or scripts; prints findings to stdout, exits 0/1. --json for the raw payload, --timeout <minutes> to override the 30-minute default, --post to post findings to a GitHub PR as a comment (--no-post is default). Requires v2.1.227+ Review
claude auto-mode reset Restore the default auto-mode configuration (confirms first; --yes to skip) Maintenance

Flags

CommandPurpose
--verbose Verbose turn-by-turn logging
--max-turns Limit agentic turns in print mode
--model Override model for the current launch
--permission-mode Start in default / plan / auto / bypass-style mode
--permission-prompt-tool Use an MCP tool to answer permission prompts in non-interactive mode
--resume Resume a specific session by ID
--continue Resume the latest conversation in the current directory
--dangerously-skip-permissions Skip permission prompts; use with caution
--output-format json Machine-readable output for scripting
--forward-subagent-text Include subagent text and thinking in stream-json output (v2.1.211+)
--restricted Start in restricted mode for shared/eval-harness machines: removes tools that run commands or code and WebFetch (unless named in --tools), confines file tools to the working directories, loads only managed settings and --settings, refuses bypassPermissions. Also CLAUDE_CODE_RESTRICTED=1. Requires v2.1.248+
--permission-prompts none Choose who answers permission prompts in print mode; none denies anything that would prompt instead of asking, for unattended headless hosts. Requires v2.1.259+
--append-subagent-system-prompt-file Read the subagent system-prompt addition from a file instead of the command line (sibling of --append-subagent-system-prompt); -p only. Requires v2.1.261+

Built-in commands

Built-in commands documented in the current Commands reference:

Essential commands 11
CommandPurposeCategoryAvailability
/btw [question] Ask a side question without polluting main context; without a question, reopens the overlay on your most recent side question this session (v2.1.212+) Context Built-in
/clear [name] Start a fresh conversation; previous one stays in /resume, optionally labeled with a name Session Built-in
/compact [instructions] Compact the conversation with optional focus Context Built-in
/context [all] Visualize context usage and optimization hints. Since v2.1.216, when the conversation exceeds the context window the output includes a warning showing how far over the limit you are and which command frees space Context Built-in
/effort [level|auto|status] Change effort level immediately (low/medium/high/xhigh/max/ultracode; auto resets to default; status prints the current level). Also available in non-interactive mode (-p) with a level argument since v2.1.205, session-only. Since v2.1.257, press s on a level to change it for the current session only, matching the /model picker Model Built-in
/model [model] Change the model immediately and save it as default for new sessions. Also available in non-interactive mode (-p) with a model argument since v2.1.205, session-only Model Built-in
/permissions Manage allow / ask / deny rules and working dirs. Since v2.1.246 includes an Auto mode tab for viewing and editing auto-mode classifier rules Settings Built-in
/plan [description] Enter plan mode directly Planning Built-in
/resume [session] Resume a session by ID, name, or picker Session Built-in
/rewind Rewind conversation and/or code to a checkpoint Session Built-in
/status Open status view with version, model, account, connectivity Settings Built-in
Session, context & memory 9
CommandPurposeCategoryAvailability
/add-dir <path> Add a working directory for file access Context Built-in
/artifacts List artifacts you own or that are shared with you, then attach one to the session, open it in your browser, or copy its link. Requires v2.1.208+; attaching with Enter requires v2.1.216+ Session Built-in
/autocompact [auto|<tokens>] Set the auto-compact window (how full the context gets before auto-compaction), e.g. 500k, or auto for the model-tuned default; without an argument opens a dialog showing the current window. Requires v2.1.221+ Context Built-in
/branch [name] Fork the current conversation into a branch Session Built-in
/cd <path> Move this session to a new working directory, keeping the conversation and its prompt cache; prompts to trust the workspace if unfamiliar. To grant access to an extra directory without moving, use /add-dir. Requires v2.1.169+ Context Built-in
/exit Exit the CLI Session Built-in
/memory Edit memory files and auto-memory settings Memory Built-in
/recap Generate a one-line summary of the current session on demand Session Built-in
/rename [name] Rename the current session; without a name, auto-generates one from history. Also available in non-interactive mode (-p) since v2.1.205 Session Built-in
Models & planning 4
CommandPurposeCategoryAvailability
/advisor [model|off] Enable/disable the advisor tool that consults a second model at key moments (fable, opus, sonnet, or a full model ID); fable requires Fable access and its usage-credits consent is set up via /model fable Model Built-in
/fast [on|off] Toggle fast mode. Since v2.1.205, also works in non-interactive mode (-p) but only in a session launched with fast mode set via --settings Model Plan / platform specific
/goal [condition|clear] Set a completion condition — Claude keeps working across turns until it's met; shows live elapsed/turns/tokens overlay; clear to remove Autonomous Built-in
/ultraplan <prompt> Removed in v2.1.222; the browser-backed ultraplan feature no longer exists. Use /plan for planning Planning Deprecated / removed
Agents, tasks & automation 13
CommandPurposeCategoryAvailability
/agents Since v2.1.198 prints a reminder to ask Claude to manage subagents or edit .claude/agents/ Agents Built-in
/autofix-pr [prompt] Watch a PR remotely and push fixes for CI / review feedback Automation Plan / platform specific
/background [prompt] Detach the session to run as a background agent and free the terminal; monitor with claude agents. Alias: /bg Background Built-in
/fork [prompt] Copy the current conversation into a new background session (its own row in agent view) while you keep working here; since v2.1.212 (was an in-session subagent before — use /subtask for that). Since v2.1.221 the copy creates its own worktree by default before editing files (was: worked in the original session's checkout) Agents Built-in
/install-github-app Set up the Claude GitHub Actions app Automation Plan / platform specific
/install-slack-app Install the Claude Slack app Automation Plan / platform specific
/list-agents List subagents and other Claude Code sessions Claude can message, with the name to use for each (also /peers); needed for cross-session SendMessage. Requires v2.1.224+ and cross-session messaging enabled; teammate rows and the first line showing this session's own name require v2.1.239+ Agents Built-in
/schedule [description] Create and manage routines Automation Built-in
/stop Stop the current background session (transcript and worktree kept); to detach without stopping use /exit Background Built-in
/subtask <directive> Hand a side task to an in-session subagent whose result comes back into this conversation (the old /fork behavior, renamed in v2.1.212) Agents Built-in
/tasks List and manage background tasks Tasks Built-in
/workflow-authoring Load the reference for writing dynamic workflow scripts: script API, resume behavior, quality patterns, worked examples. Claude normally loads it on its own before writing a script; run it yourself before editing a saved script by hand. Available when dynamic workflows are enabled; requires v2.1.248 or later Automation Bundled skill
/workflows Watch, pause, resume, or save running and completed dynamic workflows Automation Built-in
Review, safety & privacy 7
CommandPurposeCategoryAvailability
/diff Review uncommitted changes, including edits Claude has made so far. Since v2.1.260, in fullscreen mode it opens a persistent side panel that updates live as you keep working; in the classic renderer it opens a full-screen viewer instead Review Built-in
/pr-comments [PR] Removed in v2.1.91; ask Claude directly instead Review Deprecated / removed
/privacy-settings Manage privacy settings Privacy Plan / platform specific
/review [PR] Run a fast single-pass, read-only review of a GitHub PR by number; with no argument, lists open PRs to pick from. Since v2.1.202 (reverted from the v2.1.186–2.1.201 multi-agent engine); for a multi-agent review at a chosen effort level use /code-review <level> <pr#>, or /code-review ultra for a cloud review Review Built-in
/sandbox Toggle sandbox mode where supported Safety Plan / platform specific
/security-review Review current changes for security issues Review Built-in
/ultrareview [PR] Deep multi-agent cloud code review; preferred form is now /code-review ultra (this remains an alias) Review Plan / platform specific
Extensions & setup 25
CommandPurposeCategoryAvailability
/batch <instruction> Parallel large-scale codebase change orchestration Bundled skill Bundled skill
/claude-api [migrate|upgrade|managed-agents-onboard|prompt-audit|cost-optimize] Load Claude API reference for the project language. Run cost-optimize to profile the project's Claude API spend and work through savings (caching, token hygiene, batch, effort, model choice) one change at a time; requires v2.1.247 or later Bundled skill Bundled skill
/dataviz [request] Chart/dashboard design guidance: picks chart form, assigns color by role, validates colorblind-safe palette Bundled skill Bundled skill
/debug [description] Enable debug logging and investigate session issues Bundled skill Bundled skill
/deep-research <question> Fan out web searches, cross-check sources, and synthesize a cited report (bundled workflow) Bundled workflow Bundled workflow
/design-login Authorize design-system access for /design-sync with your claude.ai account Design Plan / platform specific
/design-sync [hint] Convert your repo's React design system and upload it to Claude Design so generated designs use your real components Bundled skill Bundled skill
/fewer-permission-prompts Scan transcripts for common read-only calls and add a prioritized allowlist to project settings (bundled skill) Bundled skill Bundled skill
/hooks Inspect configured hooks Hooks Built-in
/ide Manage IDE integration and status IDE Built-in
/import [codex|gemini] [--dry-run] [--yes] Bring configuration from other coding agents on your machine (OpenAI Codex, Google Gemini CLI) into Claude Code, including instruction files, MCP servers, commands, subagents, and skills. --dry-run previews without writing, --yes skips the picker. Not available on Bedrock, Google Cloud's Agent Platform, Microsoft Foundry, or Claude Platform on AWS. Requires v2.1.213+ Setup Built-in
/init Generate CLAUDE.md guidance for the project Setup Built-in
/loop [interval] [prompt] Run a prompt repeatedly while the session stays open Bundled skill Bundled skill
/mcp [reconnect <server>|enable|disable [<server>|all]] Manage MCP connections and OAuth; reconnect, enable, or disable servers. Also available in non-interactive mode (-p) since v2.1.205, printing a text summary instead of the list MCP Built-in
/plugin Manage Claude Code plugins Plugins Built-in
/reload-plugins [--force] Reload active plugins without restart Plugins Built-in
/reload-skills Re-scan skill and command directories so on-disk changes apply without restarting Skills Built-in
/run Launch and drive your project's app to see a change working, not only passing tests (bundled skill) Bundled skill Bundled skill
/run-skill-generator Teach /run and /verify how to build, launch, and drive your project's app (bundled skill) Bundled skill Bundled skill
/setup-bedrock Configure Amazon Bedrock auth and model pins Providers Plan / platform specific
/setup-vertex Configure Google Vertex AI auth and model pins Providers Plan / platform specific
/simplify [focus] Review recent changes for reuse and quality, then fix them Bundled skill Bundled skill
/skills List available skills Skills Built-in
/verify Confirm a change works by building and running the app and observing the result, not just tests (bundled skill) Bundled skill Bundled skill
/mcp__<server>__<prompt> Prompt exposed dynamically by an MCP server MCP Built-in
Interface, access & account 25
CommandPurposeCategoryAvailability
/chrome Configure Claude in Chrome settings Platform Plan / platform specific
/color [color|default] Set the prompt bar color for this session; syncs to claude.ai/code when Remote Control is connected. Also available in non-interactive mode (-p) since v2.1.205 UI Built-in
/cost Show token usage statistics Usage Built-in
/desktop Continue the session in the desktop app Platform Plan / platform specific
/usage-credits Configure usage credits to keep working past a limit (renamed from /extra-usage) Usage Plan / platform specific
/focus Toggle focus view: only your last prompt, a one-line tool-call summary with diffstats, and the final response (fullscreen only) UI Built-in
/insights Analyze Claude Code session patterns Usage Built-in
/login Sign in to your Anthropic account. Since v2.1.243, Anthropic Console also offers a keyless "Sign in with your Console account" option alongside creating an API key, for organizations that don't allow API keys Auth Built-in
/logout Sign out from Anthropic account Auth Built-in
/mobile Show QR code for the Claude mobile app Platform Plan / platform specific
/passes Share a free week of Claude Code Plan Plan / platform specific
/powerup Interactive feature lessons Learning Built-in
/remote-control Expose this session to claude.ai remote control Remote Plan / platform specific
/remote-env Choose the default remote web environment Remote Plan / platform specific
/scroll-speed [value] Tune mouse wheel scroll speed with a live preview UI Built-in
/stats Show usage history, streaks, and model preferences Usage Built-in
/statusline Configure the status line UI Built-in
/team-onboarding Generate a teammate onboarding guide from usage history Team Built-in
/teleport Pull a Claude Code web session into the terminal Remote Plan / platform specific
/theme Change theme, including daltonized and ANSI themes UI Built-in
/tui [default|fullscreen] Switch between default and fullscreen terminal rendering UI Built-in
/upgrade Open the upgrade page for a higher plan Plan Plan / platform specific
/usage Show plan usage and rate limits. Since v2.1.243 includes a Loops breakdown (per-loop run count, tokens, tokens per run, last run); since v2.1.251 adds a Spend limit bar for accounts behind a Claude apps gateway with spend limits Usage Built-in
/voice [hold|tap|off] Toggle push-to-talk voice dictation Input Plan / platform specific
/web-setup Connect GitHub account for Claude Code on the web Remote Plan / platform specific
Utility & maintenance 16
CommandPurposeCategoryAvailability
/bug [report] Report a bug or share your conversation; you choose how much history to include and confirm before sending. Alias: /share. Canonical name since v2.1.212 (was an alias of /feedback before) Utility Built-in
/checkup Alias for /doctor (v2.1.205+) Maintenance Bundled skill
/config [key=value ...] Open the settings UI, or set keys directly (v2.1.181+), e.g. /config theme=dark model=sonnet. Alias: /settings Settings Built-in
/copy [N] Copy the latest assistant response or code block Utility Built-in
/doctor Full setup checkup: diagnoses installation, settings, unused skills/MCP/plugins, and offers to fix what it finds (asks first). Alias: /checkup. Before v2.1.205 it only opened read-only diagnostics Maintenance Bundled skill
/export [filename] Export the current conversation Utility Built-in
/feedback [report] Send product feedback; opens the same dialog as /bug with the same consent step and sending rules Utility Built-in
/heapdump Write a JS heap snapshot and memory breakdown to ~/Desktop for diagnosing high memory usage Maintenance Built-in
/help Show help and available commands Utility Built-in
/keybindings Open keybindings configuration Settings Built-in
/radio Open Claude FM lo-fi radio in the browser; prints the stream URL when no browser is available Fun Plan / platform specific
/release-notes Open the changelog picker Maintenance Built-in
/skill-doctor Show which loaded skills go unused and what they cost in context, so you can prune them (v2.1.261+) Maintenance Bundled skill
/stickers Order Claude Code stickers Fun Plan / platform specific
/terminal-setup Configure Shift+Enter and terminal-specific shortcuts Settings Built-in
/vim Removed in v2.1.92; use /config → Editor mode Settings Deprecated / removed

Keyboard & input

Interactive controls from current Interactive mode docs:

ShortcutAction
Ctrl+C Cancel current input or generation
Ctrl+D Exit Claude Code
Ctrl+L Clear terminal screen, keep history
Esc Esc Edit the previous message
Shift+Tab Cycle permission modes
\ + Enter Multiline input in every terminal
Option+Enter Default multiline on macOS
Shift+Enter Works out of the box in iTerm2, WezTerm, Ghostty, Kitty
Ctrl+O Open transcript viewer
/, !, @ Slash command, Bash mode, file / resource mention

Memory, paths & scope

Where Claude stores memory and project-level behavior:

Path / keyScopeDetails
./CLAUDE.md or ./.claude/CLAUDE.md Shared project instructions Project architecture, coding standards, common workflows; committed with the repo.
./CLAUDE.local.md Personal project instructions Project-specific but not team-shared; add to .gitignore.
~/.claude/CLAUDE.md User instructions Your personal defaults across every project.
.claude/settings.json Shared project settings Team-visible settings such as permissions, hooks, shared MCP behavior.
.claude/settings.local.json Local project settings Personal preferences for this repo; automatically gitignored when Claude creates it.
~/.claude/settings.json User settings Applies to all projects on your machine.
~/.claude.json Global config Holds global keys like editor mode and IDE auto-connect, not regular settings.json keys.
.mcp.json Shared MCP project config Project-scoped MCP server definitions shared with the repo.
~/.claude/skills/<name>/SKILL.md Personal skills Reusable skills across all projects.
.claude/skills/<name>/SKILL.md Project skills Repo-local skills for the current codebase.
~/.claude/agents/ and .claude/agents/ Subagent definitions User-wide and project-wide custom subagents.

Settings & permissions

The settings system is hierarchical. Put team defaults in shared files, personal tweaks in local or user scope.

Hierarchy

  • ~/.claude/settings.json User settings across every project
  • .claude/settings.json Shared project settings
  • .claude/settings.local.json Personal settings for one repo
  • managed-settings.json Enterprise policy, highest precedence

Core top-level settings

  • advisorModel v2.1.210: model for the advisor tool (fable, opus, sonnet, or a full model ID); written automatically by /advisor; fable requires Fable access; unset to disable
  • askUserQuestionTimeout v2.1.200: idle time before an unanswered dialog auto-continues ("60s", "5m", "never")
  • autoMode Classifier rules for auto mode. Not read from shared project settings; since v2.1.207 also no longer read from .claude/settings.local.json — set it in ~/.claude/settings.json
  • awaySummaryEnabled Show a one-line session recap when you return after being away
  • autoUpdatesChannel stable vs latest update stream
  • availableModels Restrict selectable models
  • bashOutputMaxChars v2.1.261: raise how much Bash command output Claude receives inline before it is saved to a file (up to 128K characters); see also taskOutputMaxChars
  • cleanupPeriodDays Session retention and orphan cleanup
  • crossSessionInbound v2.1.224: how this session treats inbound cross-session messages from your other Claude Code sessions — accept delivers them, hold shows a notice without delivering, refuse drops them
  • defaultShell Shell for ! commands
  • dialogExpiry v2.1.224: default 5m; deadline for dialogs Claude Code forwards to a remote client (Remote Control, SDK host) or the approval dialog for a held cross-session message
  • disableBundledSkills Disable bundled skills and workflows globally
  • disableWorkflows Disable dynamic workflows
  • effortLevel Persist low / medium / high effort
  • emojiCompletionEnabled v2.1.217: show emoji suggestions when typing : plus a shortcode, and replace a completed shortcode like :heart: with its emoji; false disables both
  • enableArtifact v2.1.196: turn off the Artifact tool from any settings file; false always wins, nothing turns it back on. Before v2.1.242, ignored in project and local settings
  • fallbackModel Up to 3 fallback models to try when the primary is unavailable
  • env Environment variables applied to each session
  • fastModePerSessionOptIn Require /fast each session
  • fileSuggestion Custom script for @ autocomplete
  • keybindingFlavor v2.1.238: choose which convention Ctrl+W follows in the prompt input — "classic" (default) deletes the previous word, "readline" deletes back to the previous whitespace like Bash
  • language Preferred response language
  • model Default model override
  • modelOverrides Map Anthropic IDs to provider-specific IDs
  • modelPicker v2.1.242: curate the /model picker with an ordered, labeled list of models (any provider format), appended to or replacing the built-in lineup. Managed/--settings/user settings only, never merged across sources
  • modelPricing v2.1.242: managed setting with contracted per-model rates (multiplier and/or overrides) used for /cost, the status line, and telemetry instead of list price
  • outputStyle Default / Explanatory / Learning / custom style
  • plansDirectory Store generated plans elsewhere
  • prefersReducedMotion Accessibility and reduced animation
  • promptCacheTtl v2.1.242: prompt-cache lifetime for the main conversation ("5m" or "1h"); see also subagentPromptCacheTtl. Overridden per-session by FORCE_PROMPT_CACHING_5M, then CLAUDE_CODE_PROMPT_CACHE_TTL, then ENABLE_PROMPT_CACHING_1H
  • respectGitignore Apply .gitignore to @ picker suggestions
  • spellcheck v2.1.235: underline misspelled words in the prompt input using an installed checker (aspell, hunspell, ispell, or auto); object with enabled, checker, language, color
  • statusLine Custom status line command / renderer
  • subagentPromptCacheTtl v2.1.242: prompt-cache lifetime ("5m" or "1h") for subagents, workflows, and other requests outside the main conversation — covers what promptCacheTtl doesn't
  • taskOutputMaxChars v2.1.261: raise how much background-task output Claude receives inline before it is saved to a file (up to 128K characters); see also bashOutputMaxChars
  • timeFormat v2.1.257: 12-hour, 24-hour, 24-hour UTC, or a strftime pattern for the turn-end clock and transcript-view timestamps
  • timeZone v2.1.257: time zone used for the turn-end clock and transcript-view timestamps; pairs with timeFormat
  • useAutoModeDuringPlan Auto-mode semantics during plan mode
  • viewMode Default transcript view mode
  • vimInsertModeRemaps v2.1.208: map two-key insert-mode sequences like jj to Escape in vim mode
  • voiceEnabled Push-to-talk voice dictation
  • workflowSizeGuideline v2.1.219: set the advisory Dynamic workflow size guideline (small / medium / large / unrestricted) from any settings file; hides the /config row while set

Security / policy settings

  • permissions allow / ask / deny / additionalDirectories / defaultMode. Since v2.1.257, defaultMode auto and bypassPermissions no longer take effect from project or local settings (bypassPermissions used to) — set them in user or managed settings, or pass --permission-mode for one session
  • permissions.blockReadsOutsideWorkingDirectories v2.1.257: block Read from going outside the working directories; auto mode otherwise shows a one-time prompt before the first such read
  • disableAllHooks Turn off hooks and custom status line
  • disableAutoMode Forbid auto mode entirely. Since v2.1.207, auto mode is on by default (no opt-in env var needed) on Bedrock, Vertex AI, and Foundry — use this setting to turn it off there
  • disableClaudeAiConnectors v2.1.182: block claude.ai MCP connectors
  • disableSideloadFlags v2.1.193: reject sideload CLI flags like --plugin-dir / --plugin-url at startup
  • disableDeepLinkRegistration Disable claude-cli:// handler registration
  • autoMode.classifyAllShell v2.1.193: route all shell commands through the auto-mode classifier
  • enforceAvailableModels v2.1.175: extend the model allowlist to the Default selection too
  • policyHelper v2.1.136: admin-deployed executable that computes managed settings dynamically
  • axScreenReader v2.1.208: opt-in screen reader mode — replaces the visual terminal UI with plain, linear text (VoiceOver, NVDA); also claude --ax-screen-reader or CLAUDE_AX_SCREEN_READER=1
  • disableSkillShellExecution Block inline shell execution inside skills
  • managedMcpServers v2.1.259: managed setting providing HTTP/SSE MCP servers to every user (same shape as .mcp.json); entries that name a command to run are skipped
  • enableAllProjectMcpServers Auto-approve all project MCP servers
  • enabledMcpjsonServers Allowlist specific .mcp.json servers
  • disabledMcpjsonServers Reject specific .mcp.json servers
  • deniedMcpServers Managed denylist for MCP servers
  • forceLoginMethod Restrict to claude.ai, console, or gateway login; since v2.1.212 enforced on every first-party login path (VS Code extension, SDK, setup-token, install-github-app), not just the terminal
  • forceLoginOrgUUID Require membership in specific org UUIDs
  • httpHookAllowedEnvVars Allowlist vars interpolated into HTTP hooks

Global config in ~/.claude.json

  • autoConnectIde Auto-connect to IDE from external terminal
  • autoInstallIdeExtension Install IDE extension automatically
  • editorMode normal or vim input mode
  • showTurnDuration Display per-turn duration
  • terminalProgressBarEnabled Enable progress bar in supported terminals
  • teammateMode How agent-team teammates are displayed

Worktree settings

  • worktree.symlinkDirectories Symlink large directories like node_modules
  • worktree.sparsePaths Use sparse checkout for large monorepos
  • .worktreeinclude Copy gitignored files such as .env into new worktrees

Sandbox settings

  • sandbox.filesystem.disabled v2.1.216: skip filesystem isolation while keeping network egress control
  • sandbox.network.strictAllowlist v2.1.219: deny non-allowlisted hosts for sandboxed commands without prompting

Permission modes

  • default Normal interactive behavior
  • acceptEdits Good for trusted repo edits with fewer interruptions
  • plan Planning first, implementation only after approval
  • auto Classifier-driven auto mode where available
  • dontAsk Aggressive reduced prompting
  • bypassPermissions Highest-risk mode; can be disabled by policy

Permission rule patterns

  • Bash(git diff *) Allow safe inspection commands
  • Bash(git push *) Ask before pushes
  • Read(./.env) Hide secrets completely
  • Read(./secrets/**) Hide secret directories completely
  • Skill(deploy *) Deny risky deployment skills unless explicit
  • WebFetch Disable outbound fetches if policy requires it

Useful environment variables

  • ANTHROPIC_API_KEY Override subscription auth with API usage
  • ANTHROPIC_BASE_URL Route requests through a proxy or gateway
  • ANTHROPIC_DEFAULT_MODEL v2.1.236: model that new sessions start on by default
  • CLAUDE_CODE_NEW_INIT Interactive /init flow with memory, skills, and hooks
  • CLAUDE_CONFIG_DIR Use a different config directory for a second account / profile
  • CLAUDE_ENV_FILE Persist environment variables across Bash commands
  • CLAUDE_CODE_USE_BEDROCK Use Amazon Bedrock provider
  • CLAUDE_CODE_USE_VERTEX Use Google Vertex AI provider
  • CLAUDE_CODE_USE_POWERSHELL_TOOL Enable native PowerShell tool on Windows
  • CLAUDE_CODE_TASK_LIST_ID Share a task list across sessions
  • CLAUDE_CODE_NO_FLICKER Enable fullscreen rendering research preview
  • CLAUDE_CODE_SYNTAX_HIGHLIGHT Disable syntax highlighting if terminal colors misbehave
  • CLAUDE_ENABLE_STREAM_WATCHDOG Abort hung streams in automation scenarios
  • CLAUDE_CODE_PROCESS_WRAPPER v2.1.208: run every Claude Code self-spawn through a required corporate launcher wrapper
  • CLAUDE_CODE_FORWARD_SUBAGENT_TEXT v2.1.211: include subagent text and thinking in stream-json output
  • CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION v2.1.212: session-wide WebSearch call limit to stop runaway search loops (default 200)
  • CLAUDE_CODE_MAX_SUBAGENTS_PER_SESSION v2.1.212: per-session cap on subagent spawns (default 200); /clear reset the budget. Removed in v2.1.224 — long-running sessions no longer refuse new agents (concurrency/depth limits still apply)
  • CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS v2.1.217: cap on concurrently-running subagents so one message can't fan out unbounded background agents (default 20)
  • CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH v2.1.217: how deep subagents can spawn nested subagents; default raised to 3 in v2.1.219 (was 1/disabled)
  • CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS v2.1.212: threshold for auto-backgrounding MCP tool calls running longer than 2 minutes (0 disables)
  • CLAUDE_CODE_RESTRICTED v2.1.248: start in restricted mode, same as --restricted; ignored inside a settings file's env block
  • CLAUDE_CODE_OTEL_CONTENT_MAX_LENGTH v2.1.214: truncation limit for OpenTelemetry content attributes (default 60 KB)
  • ANTHROPIC_BEDROCK_REGION_PREFIX v2.1.224: for Bedrock, prefer a specific cross-region inference profile over the one derived from AWS_REGION
  • CLAUDE_CODE_DISABLE_UNKNOWN_MODEL_WINDOW_ENFORCEMENT v2.1.223: restore pre-v2.1.223 behavior of letting sessions on unrecognized model IDs grow past the assumed context window instead of auto-compacting to stay within it
  • CLAUDE_CODE_SUBAGENT_MODEL Default model for subagents, agent-team teammates, and workflow agents not assigned one another way (alias or full model name); an explicit per-spawn model or the agent definition's model take precedence unless CLAUDE_CODE_SUBAGENT_MODEL_FORCE is set
  • CLAUDE_CODE_SUBAGENT_MODEL_FORCE v2.1.257: set to 1 to force CLAUDE_CODE_SUBAGENT_MODEL (or the main model) onto every subagent, teammate, and workflow agent, overriding per-spawn and agent-definition model choices
  • CLAUDE_CODE_DISABLE_CFC_PROMPT v2.1.257: keep Claude in Chrome browser tools available while omitting the Chrome section of the system prompt and the /claude-in-chrome bundled skill; for hosts that embed Claude Code and supply their own browser guidance

Hooks

Hooks are lifecycle automations defined in settings JSON. Keep SessionStart fast; use CLAUDE.md for static context.

EventWhen it firesNotes
SessionStart On new or resumed session Only command hooks; keep it fast. Prefer CLAUDE.md for static context.
PreToolUse Before a tool runs Can allow, deny, escalate, mutate input, or inject extra context.
PermissionRequest When Claude needs approval Can answer permission prompts programmatically.
PostToolUse After a tool finishes Good for logging, async checks, and post-processing.
PreCompact Before compaction Can block manual or auto compaction.
PostCompact After compaction React to the new compacted state.
SessionEnd When the session closes Cleanup, logging, finalization; cannot block exit.
Elicitation When an MCP server asks for user input Hooks can answer instead of showing a dialog.
SubagentStop When a subagent finishes Stop hooks inside subagents are converted here.
DirectoryAdded v2.1.219: after /add-dir or the SDK register_repo_root control request registers a new working directory mid-session Fires mid-session, not just at startup.
PreModelSwitch v2.1.251: before a model switch (/model, /fast, /effort, or automatic fallback) takes effect Can block, confirm, or annotate the switch.
PostModelSwitch v2.1.251: after a model switch (/model, /fast, /effort, or automatic fallback) takes effect React to or log the new model.

Skills

Skills are reusable prompt workflows. Claude can invoke them automatically unless you disable model invocation.

Bundled skills

  • /batch Parallel large codebase migration / transformation with isolated worktrees and PRs
  • /claude-api Bring in Claude API reference material for SDK work
  • /code-review Review the current diff (or a PR/branch/path) at an effort level low..max; no level given reuses the level you last typed; --fix, --comment, or ultra for cloud review. Since v2.1.218 a local review runs as a background subagent so it doesn't fill your conversation; since v2.1.215 Claude no longer runs it on its own. Alias: /review
  • /dataviz Design guidance for charts and dashboards with colorblind-safe palette validation
  • /debug Turn on debug logging and inspect session logs
  • /deep-research Bundled workflow: fan out web searches, cross-check sources, synthesize a cited report
  • /design-sync Convert your React design system and upload it to Claude Design
  • /fewer-permission-prompts Scan transcripts and add a prioritized allowlist to project settings
  • /loop Repeat a prompt while the current session stays open
  • /run Launch and drive your project's app to see a change working
  • /run-skill-generator Teach /run and /verify how to build, launch, and drive your app
  • /simplify Parallel quality pass over recently changed files
  • /skill-doctor Show which loaded skills go unused and what they cost in context, so you can prune them
  • /verify Confirm a change works by running the app and observing the result. Since v2.1.215 Claude no longer runs it on its own — invoke it explicitly
  • /workflow-authoring Reference for writing dynamic workflow scripts: script API, resume behavior, quality patterns, worked examples (v2.1.248+)

SKILL.md frontmatter

  • name Slash-command name; defaults to directory name
  • description When the skill should be used; most important field
  • disable-model-invocation Prevents Claude from auto-loading the skill
  • allowed-tools Restrict tools available while the skill runs
  • context Can run inline or in a fork / subagent context
  • shell Use PowerShell for ! blocks on Windows where supported
  • $ARGUMENTS / $0 / $1 Argument substitution inside SKILL.md

Tools, subagents & tasks

Claude’s built-in tools matter because permissions, hooks, and subagents all reference these exact names.

Built-in tools

Tool / conceptPurposePermission
Agent Spawn a subagent with separate context No
AskUserQuestion Gather requirements or disambiguate No
Bash Run shell commands Yes
Edit Make targeted file edits Yes
Write Create or overwrite files Yes
Read / Glob / Grep Read and search the repo No
LSP Definitions, references, warnings, type errors No
Monitor Watch logs or status in background Yes
EnterWorktree Create or enter isolated git worktree No
TaskCreate / TaskList / TaskUpdate Interactive task system No
Skill Run a reusable prompt workflow Yes
WebFetch / WebSearch Use the web where enabled Yes
EndConversation v2.1.214: end sessions with highly abusive users or jailbreak attempts No

Subagents & agent teams

  • Explore Fast, read-only codebase search and analysis agent (Haiku)
  • Plan Research agent used while Claude is in plan mode
  • General-purpose Multi-step agent for exploration plus modification
  • Custom subagents Markdown definitions with frontmatter and tool restrictions
  • Agent teams Experimental teammate mode controlled via settings and env flags

MCP

MCP gives Claude extra tools, prompts, and resources. Use scopes deliberately:

MCP commands

  • claude mcp add --transport stdio myserver -- npx my-mcp Add a stdio server
  • claude mcp add --transport http hubspot https://mcp.example.com Add an HTTP server
  • claude mcp list List configured servers
  • claude mcp get github Inspect one server
  • claude mcp remove github Remove a server
  • /mcp Inspect server status and handle OAuth inside Claude

MCP patterns to remember

  • --scope local Only you in the current project (new name for older project-local scope)
  • --scope project Shared with the repo via .mcp.json
  • --scope user Available across all projects for the current user
  • @server:protocol://resource/path Reference an MCP resource with @ mention
  • /mcp__<server>__<prompt> Run an MCP prompt exposed as a command
  • MCP_TIMEOUT Startup timeout for MCP servers
  • MAX_MCP_OUTPUT_TOKENS Default cap is 25,000 tokens; warning starts above 10,000
  • list_changed Dynamic refresh of tools, prompts, and resources without reconnecting
  • --channels Enable pushed channel messages from MCP servers that support them

Worktrees, remote & routines

Use isolation when parallel work would otherwise collide:

  • claude --worktree feature/name Launch Claude inside a new isolated git worktree
  • -w #<number> Fetch a GitHub PR or GitLab MR (v2.1.233+) from origin and branch the worktree from it
  • /schedule Create, update, list, or run routines conversationally
  • /tasks Inspect long-running background tasks
  • --remote Start or connect to Claude Code on the web / remote environments
  • /remote-env Choose the default web environment used by --remote
  • /teleport Pull a web session back into the terminal
  • /remote-control Let claude.ai remote-control this terminal session

IDE, web & desktop

Current official surfaces beyond the plain terminal:

  • VS Code / Cursor / Windsurf / VSCodium Run claude in the integrated terminal and the extension auto-installs
  • JetBrains IDEs Dedicated plugin with quick launch, diff viewer, selection context, diagnostics
  • Cmd+Esc / Ctrl+Esc Quick launch from supported IDE integrations
  • Cmd+Option+K / Alt+Ctrl+K Insert file reference shortcuts from the IDE
  • /ide Connect an external terminal session back to the IDE
  • Desktop app Use /desktop to continue the session in the desktop UI
  • Claude Code on the web Remote browser-based sessions with environment selection and setup scripts

What's new by month

Commands, flags, and settings added since April 2026 — grouped by release month. Version numbers are approximate first appearances.

September 2026

  • /skill-doctor /skill-doctor shows which loaded skills go unused and what they cost in context, so you can prune them command v2.1.261
  • bashOutputMaxChars + taskOutputMaxChars New settings raise how much Bash command and background-task output Claude receives inline before it is saved to a file, up to 128K characters setting v2.1.261
  • --append-subagent-system-prompt-file New flag reads the subagent system-prompt addition from a file, for prompts too large to pass on the command line flag v2.1.261
  • /diff fullscreen side panel In fullscreen mode, /diff now opens a persistent side panel that updates live as Claude edits files, instead of a full-screen viewer command v2.1.260
  • managedMcpServers New managed setting lets organizations provide HTTP/SSE MCP servers to every user, in the same shape as .mcp.json setting v2.1.259
  • --permission-prompts none New flag for unattended headless hosts: none denies anything that would prompt instead of asking, while the active permission mode keeps deciding everything else flag v2.1.259
  • Claude Fable 5.1 default Claude Fable 5.1 (claude-fable-5-1) shipped and is now the default Fable model — 1M context, $10/$50 per Mtok with $0.25/Mtok cache reads feature v2.1.257
  • timeFormat + timeZone New settings choose 12-hour, 24-hour, 24-hour UTC, or a strftime pattern and time zone for the turn-end clock and transcript-view timestamps setting v2.1.257
  • CLAUDE_CODE_SUBAGENT_MODEL_FORCE New environment variable forces CLAUDE_CODE_SUBAGENT_MODEL (or the main model) onto every subagent, teammate, and workflow agent, overriding per-spawn and agent-definition model choices env var v2.1.257
  • permissions.blockReadsOutsideWorkingDirectories New setting blocks Read from going outside the working directories; auto mode otherwise shows a one-time prompt before the first such read setting v2.1.257
  • defaultMode bypassPermissions restricted defaultMode: "bypassPermissions" in project or local settings is now ignored, like "auto"; set it in user or managed settings, or pass --permission-mode for one session setting v2.1.257
  • CLAUDE_CODE_DISABLE_CFC_PROMPT New environment variable keeps Claude in Chrome browser tools available while omitting the Chrome section of the system prompt and the /claude-in-chrome bundled skill, for hosts that embed Claude Code and supply their own browser guidance env var v2.1.257

August 2026

  • PreModelSwitch / PostModelSwitch hooks New hook events fire before and after a model switch (/model, /fast, /effort, or automatic fallback) so a hook can block, confirm, or annotate it feature v2.1.251
  • --restricted / CLAUDE_CODE_RESTRICTED New restricted mode for shared or eval-harness machines: removes tools that run commands or code and WebFetch (unless named in --tools), confines file tools to the working directories, loads only managed settings and --settings, and refuses bypassPermissions flag v2.1.248
  • /workflow-authoring New bundled skill: loads the dynamic workflow scripting reference (script API, resume behavior, quality patterns, worked examples); Claude loads it on its own before writing a script, run it yourself before hand-editing a saved one command v2.1.248
  • /claude-api cost-optimize New /claude-api subcommand profiles the project's Claude API spend and works through savings (caching, token hygiene, batch, effort, model choice) one measured change at a time command v2.1.247
  • /permissions Auto mode tab /permissions gained an Auto mode tab for viewing and editing auto-mode classifier rules command v2.1.246
  • modelPicker + modelPricing New settings: modelPicker curates the /model picker with an ordered, labeled model list; modelPricing lets organizations report /cost, status line, and telemetry figures at contracted per-model rates instead of list price setting v2.1.242
  • promptCacheTtl + subagentPromptCacheTtl New settings choose the prompt-cache lifetime ("5m" or "1h") separately for the main conversation and for subagents/workflows/background requests setting v2.1.242
  • /login keyless Console sign-in Anthropic Console sign-in in /login gained a keyless "Sign in with your Console account" option alongside creating an API key, for organizations that don't allow API keys command v2.1.243
  • /usage Loops breakdown /usage gained a Loops breakdown: per-loop run count, total tokens, tokens per run, and last run, so runaway or chatty /loop tasks are easy to spot command v2.1.243
  • /list-agents teammate rows Teammate rows and the first line showing this session's own name in /list-agents (also /peers) require v2.1.239 or later command v2.1.239
  • keybindingFlavor New setting to make Ctrl+W follow readline conventions (delete back to previous whitespace, like Bash) instead of the classic previous-word delete setting v2.1.238
  • spellcheck New setting to underline misspelled words in the prompt input as you type, using an installed checker (aspell, hunspell, or ispell) setting v2.1.235
  • ANTHROPIC_DEFAULT_MODEL New environment variable to set the model that new sessions start on by default env var v2.1.236
  • claude ultrareview --json/--timeout/--post Non-interactive claude ultrareview gains --json for the raw payload, --timeout <minutes> to override the 30-minute default, and --post to post findings to a GitHub PR as a comment flag v2.1.227
  • GitLab merge request support /batch and claude --worktree can branch a worktree from a GitLab MR URL or number (v2.1.233+), and a clickable MR badge appears in the footer when glab is authenticated (v2.1.234+) feature v2.1.233
  • /list-agents + cross-session SendMessage Claude Code sessions can now message each other across your machines via SendMessage; /list-agents (also /peers) lists reachable subagents and sessions feature v2.1.224
  • /autocompact New command to set the auto-compact window (how full context gets before auto-compaction), e.g. /autocompact 500k or /autocompact auto command v2.1.221
  • /ultraplan removed The browser-backed /ultraplan feature was removed; use /plan for planning instead feature v2.1.222
  • /code-review runs in the background Local /code-review reviews now run as a background subagent in interactive sessions instead of filling your conversation command v2.1.218
  • /verify and /code-review no longer auto-run Claude no longer runs the /verify and /code-review skills on its own — invoke them explicitly when you want them command v2.1.215
  • /review = /code-review alias, effort level reuse /review is now an alias of /code-review; with no effort level given, /code-review reuses the level you last typed command v2.1.223
  • Claude Opus 5 default Claude Opus 5 (claude-opus-5) shipped and is now the default Opus model, with 1M context and fast mode feature v2.1.219
  • Subagent nesting depth 3 by default Subagents can now spawn nested subagents up to depth 3 by default (was disabled); set CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH to change it env var v2.1.219
  • Per-session subagent spawn cap removed The 200-subagent-per-session spawn cap was removed; long sessions no longer refuse new agents (concurrency / depth limits still apply) env var v2.1.224
  • emojiCompletionEnabled Emoji shortcode autocomplete in the prompt input (type :heart: for ❤️); disable with emojiCompletionEnabled setting v2.1.217
  • crossSessionInbound + dialogExpiry New cross-session messaging settings: crossSessionInbound controls how inbound messages from other sessions are handled (accept / hold / refuse), dialogExpiry sets the deadline for remote-forwarded dialogs setting v2.1.224
  • sandbox.filesystem.disabled / sandbox.network.strictAllowlist New sandbox settings: filesystem.disabled skips filesystem isolation while keeping network egress control, network.strictAllowlist denies non-allowlisted hosts without prompting setting v2.1.216
  • /fork creates its own worktree by default /fork now creates its own worktree by default before editing files, instead of working in the original session's checkout command v2.1.221
  • claude self-hosted-runner New claude self-hosted-runner CLI subcommand turns your own machines or containers into a place Claude Code web, mobile, and desktop sessions can run (Team / Enterprise) feature v2.1.224
  • /context exceeded-window warning /context now shows an explicit warning when the conversation exceeds the context window, naming the command that frees space command v2.1.216

July 2026

  • /fork creates a background session, /subtask renamed /fork now copies your conversation into a new background session (its own row in agent view) while you keep working; the old in-session subagent behavior is now /subtask command v2.1.212
  • /bug canonical name (was alias of /feedback) /bug is now the primary command for bug reports and conversation sharing, with /share as its alias; before v2.1.212 both were aliases of /feedback command v2.1.212
  • WebSearch and subagent spawn session limits New session-wide caps to stop runaway loops: WebSearch calls (default 200, CLAUDE_CODE_MAX_WEB_SEARCHES_PER_SESSION) and subagent spawns (default 200, CLAUDE_CODE_MAX_SUBAGENTS_PER_SESSION); /clear resets both budgets env var v2.1.212
  • MCP tool calls auto-background after 2 minutes Long-running MCP tool calls now move to the background automatically so the session stays usable; configure the threshold or disable with CLAUDE_CODE_MCP_AUTO_BACKGROUND_MS env var v2.1.212
  • Screen reader mode Opt-in plain-text rendering that replaces the visual terminal UI for screen readers like VoiceOver and NVDA; run claude --ax-screen-reader, set CLAUDE_AX_SCREEN_READER=1, or axScreenReader: true setting v2.1.208
  • EndConversation tool Claude can now end sessions with highly abusive users or jailbreak attempts, as on claude.ai since 2025 feature v2.1.214
  • advisorModel setting Configure the advisor model (opus, sonnet, or a full model ID) via settings, written automatically by /advisor; Fable 5 is not offered as advisor setting v2.1.210
  • autoMode not read from settings.local.json Auto mode config no longer read from project .claude/settings.local.json; set it in ~/.claude/settings.json instead setting v2.1.207
  • disableAutoMode (Bedrock/Vertex/Foundry default-on) Auto mode now runs by default without an opt-in env var on Bedrock, Vertex AI, and Foundry; disable with disableAutoMode setting v2.1.207
  • /doctor full checkup + /checkup alias /doctor is now a full setup checkup that diagnoses and can fix issues (asks first); /checkup is its alias command v2.1.205
  • /review back to single-pass /review <pr> reverted to a fast single-pass, read-only review; use /code-review <level> <pr#> for the multi-agent review at a chosen effort level command v2.1.202
  • /color, /effort, /fast, /mcp, /model, /rename in -p mode These commands now also work in non-interactive mode (-p), applying to the current session only command v2.1.205
  • Dynamic workflow size /config gets an advisory "Dynamic workflow size" setting (small/medium/large) guiding how many agents workflows tend to spawn setting v2.1.202
  • askUserQuestionTimeout Idle time before an unanswered dialog auto-continues ("60s", "5m", "never") setting v2.1.200
  • /agents (reminder mode) /agents now prints a reminder to ask Claude to manage subagents instead of opening a UI command v2.1.198
  • /dataviz Chart/dashboard design guidance with colorblind-safe palette validation (bundled skill) command v2.1.198
  • /config key=value Set settings directly from /config without opening the UI, e.g. /config theme=dark model=sonnet command v2.1.181
  • /design-sync + /design-login Upload your repo's React design system to Claude Design so generated designs use real components command v2.1.190
  • /usage-credits Configure usage credits to keep working past a limit (renamed from /extra-usage) command v2.1.185
  • disableClaudeAiConnectors Block claude.ai MCP connectors setting v2.1.182
  • axScreenReader Render screen-reader friendly output setting v2.1.181
  • footerLinksRegexes Render clickable badges in the footer from regex matches on output setting v2.1.176
  • enforceAvailableModels Extend the model allowlist to the Default selection too setting v2.1.175

June 2026

  • --safe-mode Disable all customizations for debugging flag v2.1.169
  • /cd <path> Change working directory without breaking prompt cache command v2.1.169
  • disableBundledSkills Disable all bundled skills globally setting v2.1.169
  • fallbackModel Up to 3 fallback models when primary is unavailable setting v2.1.166
  • /fork <directive> Spawn background subagent inheriting full conversation context command v2.1.161
  • /reload-skills Re-scan skill directories without restarting command v2.1.152

May 2026

  • /background [prompt] Detach session to run as background agent, free the terminal command v2.1.154
  • /run Launch app and verify a change in the running app (bundled skill) command v2.1.145
  • /verify Confirm code change works by running the app, not just tests (bundled skill) command v2.1.145
  • /run-skill-generator Teach /run and /verify how to build and launch your project (bundled skill) command v2.1.145
  • /scroll-speed Adjust mouse wheel scroll speed interactively (fullscreen only) command v2.1.154
  • /deep-research <question> Fan-out web searches, fetch sources, synthesize cited report (bundled workflow) command v2.1.154
  • /workflows Watch, pause, resume, or save running dynamic workflows command v2.1.154
  • /stop Stop the current background session (transcript and worktree kept) command v2.1.154
  • claude --bg --exec Start a background session from the CLI flag v2.1.154
  • EnterWorktree Switch between worktrees from inside a session feature v2.1.157
  • claude plugin init Initialize a new plugin from the CLI flag v2.1.157
  • worktree.bgIsolation: "none" Disable background isolation for worktrees setting v2.1.143

April 2026

  • claude agents List all sessions — running, blocked, or done (Agent View) flag v2.1.139
  • /goal <condition> Set a completion condition — Claude keeps working until met command v2.1.139
  • claude project purge [path] Delete all Claude Code state for a project (supports --dry-run) flag v2.1.126
  • --plugin-url <url> Load a plugin from a URL at launch time flag v2.1.129
  • skillOverrides Override skill visibility or behavior per project setting v2.1.129
  • worktree.baseRef: "fresh" | "head" Control whether worktrees start from a clean base or HEAD setting v2.1.133
  • settings.autoMode.hard_deny Hard deny rule for auto mode — blocks even without confirmation setting v2.1.136
  • CLAUDE_CODE_SESSION_ID Access current session ID from hooks or scripts env var v2.1.132
  • CLAUDE_CODE_DISABLE_ALTERNATE_SCREEN Disable alternate screen buffer (for terminal compatibility) env var v2.1.132
  • $CLAUDE_EFFORT in hooks Hooks receive current effort level as environment variable feature v2.1.133

Sources

Primary sources used for this sheet:

SourceFreshness
Commands Crawled today
Changelog Crawled today
What's new Crawled today
CLI reference Crawled today
Settings Crawled today
Interactive mode Crawled today
Hooks reference Crawled today
Skills Crawled today
Memory Crawled 7 months ago
MCP Crawled 2 weeks ago
Subagents Crawled 7 months ago
Tools reference Crawled today
Environment variables Crawled today
IDE integrations Crawled 7 months ago
Getting started Crawled 8 months ago