New in the last 30 days. Each row also lives in its own section below.
/update. · in Session & context/plugin. · in Config & extensionscommand and http hooks: block the action if the hook fails or times out. Default "continue" lets it through, so a broken policy hook fails open. · in Hooksno matches in this section
Everything you can type before you're in a session.
--output-format json.--config is required.stable, latest, or a pinned version./doctor can also fix things.)/import to bring in config from Codex, Gemini CLI, or Cursor. Takes --dry-run and --yes.login --console to bill via API instead of subscription. status prints JSON, exit 0 = logged in; since v2.1.318 its loginExpiresAt gives the date a stored claude.ai sign-in ends.--json for scripting.--all to pick up a new binary.--keep-workers leaves sessions running.claude mcp login <name> runs a server's OAuth flow from the shell; claude mcp logout <name> clears its stored credentials.list, enable, disable, update, init, prune).--dry-run first; --all for every project. Formerly claude project purge (still works).--post posts findings to a github.com PR as a comment. Needs workspace trust: asks on a terminal, else refuses — scripts pass --trust-workspace.setup walks an operator through config; doctor diagnoses a deployed runner. --hold-idle-file and --max-hold-min keep an idle session while your host reports processes it left running.--label <prefix> filters by rule label; claude auto-mode config shows your effective config with settings applied.autoMode section from user settings. -y/--yes skips the confirmation prompt.no matches in this section
claude --help does not list every flag — absence from help ≠ unavailable.
claude -r "<name>".<repo>/.claude/worktrees/<name>. Pass a GitHub PR or GitLab MR URL to branch from it.-w: spin the worktree up in a tmux (or iTerm2) pane.claude agents.--bg: run a shell command as a PTY-backed background job (no model)..claude/skills/, not the rest of their config.--continue or --resume <id> to open that session there. macOS and x64 Windows only.ccpool_...) instead of the default.--environment: base the new session's checkout on a named ref instead of local HEAD./autocompact.ANTHROPIC_MODEL.ultracode = xhigh + automatic workflow orchestration.fable requires Fable 5 access on your org.--append-system-prompt across --continue runs.default(=manual), acceptEdits, plan, auto, dontAsk, bypassPermissions."" = none, default = all.--permission-mode bypassPermissions. Prefer auto + an allowlist for real work.--settings, refuses bypassPermissions.none denies them when nobody can answer.credential_expires_at, and text output warns on stderr from 10 days before a stored claude.ai sign-in ends.-p --output-format stream-json.assistant/user messages with parent_tool_use_id set, so you can reconstruct each subagent's transcript. Needs -p --output-format stream-json.--max-budget-usd; hitting it blocks new subagent spawns and halts ones still running (v2.1.217+).-p mode, Claude Code now waits (up to the MCP_TIMEOUT) for these to connect before the first turn (v2.1.221+).prompt).!statsig.CLAUDE_AX_SCREEN_READER=1 or axScreenReader in settings.no matches in this section
Typed inside a session. A command is only recognized at the start of a message.
/resume; a name labels it there. Aliases /reset, /new.500k, or auto for the model-tuned default. Saves to user settings and applies to this session; saved per model as of v2.1.288.bg./undo, /checkpoint.default resets. Syncs to claude.ai/code under Remote Control.CLAUDE_CODE_NEW_INIT=1 for an interactive flow covering skills + hooks./update.no matches in this section
low → medium → high → xhigh → max; max is session-only. status prints it; ultracode on|off toggles ultracode at the current level (v2.1.284+; earlier, /effort ultracode just set xhigh).fable, opus, sonnet, or a full model ID — fable requires Fable 5 access./config theme=dark, /config thinking=false. /config --help lists keys.interactive or background job (attached/unattended). Works while Claude is responding./output-style concise./cost, /stats.no matches in this section
skill = prompt-based bundled skill (Claude can also trigger it on its own). workflow = fans out across subagents in the background.
low…max (omit to reuse your last pick, v2.1.223+), or ultra for the cloud review. --comment posts findings on the GitHub PR or GitLab MR (MR needs v2.1.257+). --max-findings raises or lowers the findings cap and sticks until you pass default (v2.1.288+). Claude won't trigger this on its own — you have to invoke it./code-review, same levels and flags. Before v2.1.223 this was a separate single-pass PR review./code-review ultra, the preferred form..claude/skills/run-<name>/ so /run and /verify stop guessing. Run once per repo.no matches in this section
claude agents, not reporting back. As of v2.1.221 the copy creates its own git worktree before editing code (unless it edits in place). (Contrast /branch, which switches you into the copy instead.)/fork did before v2.1.212.)/bg. Monitor via claude agents. As of v2.1.221, background sessions commit and push their work as they go, open a draft PR only when the task calls for one, and report where the work lives when done./bashes./peers./loop 5m check if the deploy finished. Omit the interval and Claude self-paces; since v2.1.319 that loop ends when a turn doesn't reschedule (CLAUDE_CODE_LOOP_KEEPALIVE=1 keeps the old fallback). Defaults to .claude/loop.md./routines.no matches in this section
/allowed-tools.-p mode with no args it prints a status summary.--force when it would invalidate the prompt cache.skillOverrides..claude/agents/ — ask Claude to create/manage subagents, or edit the files./checkup.-diagnostics.json — the snapshot holds your conversation and credentials. Hidden from the menu; type it in full./bug to report one — same dialog, with a consent step before anything is sent.autoMode.environment entries from your project and recent sessions, review, and save to user settings. Pro/Max/Team from 2.1.228; Enterprise and Console API keys from 2.1.283.settings.json.migrate, upgrade, api-version, managed-agents-onboard, managed-agents-secure, prompt-audit, cost-optimize, latency-optimize, build-eval, hillclimb — each has its own minimum version./plugin.no matches in this section
/app./tp./rc.claude in an IDE terminal no longer installs the extension — run /ide from an outside terminal, or use the editor's extensions view.CLAUDE_CODE_USE_BEDROCK=1 / CLAUDE_CODE_USE_VERTEX=1.no matches in this section
/code-review /fix-issue 123.! npm test → explanation of the failures.@ also suggests your other live sessions to message.:heart: to insert ❤️, or 2+ characters for suggestions. Toggle with emojiCompletionEnabled.$EDITOR.[Image #N] chip. (Alt+V on Windows/WSL.)no matches in this section
Cycle with Shift+Tab, set at launch with --permission-mode, or persist as defaultMode in settings.
claude auto-mode defaults and review denials in /permissions.no matches in this section
Custom commands merged into skills: .claude/commands/deploy.md and .claude/skills/deploy/SKILL.md both produce /deploy. Skills add supporting files, invocation control, and automatic loading. Follows the agentskills.io standard.
apps/web:deploy)./plugin-name:skill-name, so it can't collide.when_to_use it's truncated at 1,536 chars./deploy, /commit). Also keeps its description out of context.AskUserQuestion for an unattended loop). Clears on your next message.context: fork, wait for the forked subagent's result in the invoking turn instead of running it in the background (the default since v2.1.218).!`cmd` blocks in the skill. powershell runs them via PowerShell.$name substitution.$ARGUMENTS is absent, args are appended as ARGUMENTS: ….allowed-tools rules.```! block. Only recognized at line start or after whitespace.--- name: summarize-changes description: Summarize uncommitted changes and flag risks. Use when the user asks what changed, wants a commit message, or asks to review their diff. allowed-tools: Bash(git *) --- ## Current changes !`git diff HEAD` ## Instructions Summarize in 2–3 bullets, then list risks: missing error handling, hardcoded values, tests that need updating.
Gotchas worth internalizing: skill content enters context once and stays for the session (write standing instructions, not one-time steps) — keep SKILL.md under ~500 lines and push reference material into sibling files. If you have many skills, descriptions get truncated to a context budget (1% of the window; raise with skillListingBudgetFraction); /doctor tells you who's eating it. Turn skills off per-project without editing them via skillOverrides in .claude/settings.local.json.
no matches in this section
skills to preload, hooks) + a body that becomes its system prompt./btw = full context, no tools (ask about what's already known)./subtask.--teammate-mode in-process|auto|tmux|iterm2./batch does this automatically per unit of work.CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH) and cap at 20 running concurrently (CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS, exempt under ultracode). No limit on the total spawned over a session.no matches in this section
Shell, HTTP, prompt-based, or agent-based handlers on lifecycle events. Configure in settings under hooks, in a plugin, or in skill/agent frontmatter. Exit code 2 blocks; JSON output can inject context or make decisions. Inspect with /hooks, debug with claude --debug hooks.
Setup supports init / maintenance matchers via CLI flags.if field (tool name + args). Auto-format on edit lives here.PreModelSwitch can block it (a timeout blocks too); PostModelSwitch fires after any change, including restores on resume. Matcher is the model name./add-dir or an SDK register_repo_root call adds a working directory mid-session — not the --add-dir startup flag..env, xcconfig, local secrets) into new worktrees.command and http hooks: block the action if the hook fails or times out. Default "continue" lets it through, so a broken policy hook fails open.no matches in this section
skillOverrides lands here.-w and /batch put isolated worktrees./loop./keybindings opens your bindings file.no matches in this section