Command-line reference

This page is the canonical source-visible inventory of root flags, command families, aliases, gates, and mode-specific CLI surfaces. Process routing and mode dispatch remain in CLI main paths, so this reference does not duplicate the bootstrap narrative.

Scope and caveats

• This is a source-anchored reference for the analyzed cli.renamed.js bundle, not a replacement for live claude --help output.
• Hidden and feature-gated options are included when existing docs have direct anchors.
• The bundle is minified, so approximate line numbers and byte offsets are search handles rather than stable public API names.

Source anchors

Semantic alias	Anchor	Meaning
CommanderRoot	Usage: claude [options] [command] [prompt]	Commander root construction and invocation shape.
ClaudeRootCommand	Claude Code - starts an interactive session by default	Root command name and description.
PrintModeFlag	-p, --print	Selects non-interactive print/headless mode.
ToolAllowListFlag	--allowedTools, --allowed-tools <tools...>	Tool allow-list flag.
PermissionModeFlag	--permission-mode <mode>	Session permission mode selector.
ContinueSessionFlag	-c, --continue	Continue the most recent conversation in the current directory.
ResumeSessionFlag	-r, --resume [value]	Resume by session ID or open picker/search.
CloudSessionFlag	--cloud [description|session_id|url]	Hidden hosted-session creation/attach flag.
RemoteSessionAlias	--remote [description|session_id|url]	Hidden deprecated alias for --cloud.
RemoteControlFlag	--remote-control [name]	Remote Control flag.
DoctorCommand	doctor — Check the health of your Claude Code installation	Non-interactive installation/settings health check.
UpdateCommandFamily	`update	upgrade`
CoreInteractiveRegistry	Hur() / getBuiltinCommands() at ~567,141–567,875	Memoized core local, local-jsx, and built-in prompt command array.
CommandAggregator	_xo() / getCommands() at ~567,226–567,327	Merges skill directories, workflows, plugin/MCP/bundled skills, and core commands, then resolves collisions and gates.
BundledSkillRegistrar	Lu() / fhs() at ~421,026–421,095	Turns bundled skill metadata and assets into prompt commands.
SurfaceFilters	filterCommandsForHeadless, filterCommandsForRemoteMode, isThinClientSafe	Selects commands that can cross non-interactive, Remote Control, and thin-client boundaries.
AgentTeamsRawArg	process.argv.includes("--agent-teams")	Experimental team gate inspects raw argv; this spelling is not registered in the Commander root option table ~333,255–333,264.
HiddenTeammateModeFlag	--teammate-mode <mode>	Hidden root option selecting auto, tmux, iterm2, or in-process ~958,950.

Root invocation modes

Mode	Representative surface	Behavior
Interactive TUI	default TTY invocation	Starts the interactive session loop after trust/auth/settings/context setup.
Headless/print	-p, --print	Runs the scriptable path with stdout result/stream framing.
SDK-style headless	--sdk-url, stream/control frames	Exposes structured frames and control requests for external hosts.
Resume/continue	--continue, --resume, --session-id	Restores a local JSONL session into the live envelope.
Cloud/teleport/control	--cloud (deprecated --remote), --teleport, --remote-control, --rc	Adds hosted-session, import, or bridge control around related session models.
Recovery/accessibility	--safe-mode, --ax-screen-reader	Safe mode isolates broken customizations; screen-reader mode selects the classic flat renderer.

Root flag families

Family	Flags	Primary owner
Diagnostics/recovery	-d, --debug [filter], --debug-file, --verbose, --safe-mode	Diagnostics and debug logs, Safe mode and recovery
Headless/SDK output	-p, --print, --output-format, --input-format, --json-schema, --include-partial-messages, --include-hook-events, --replay-user-messages, --forward-subagent-text, --prompt-suggestions	Headless streaming and resilience
Thinking and budget	--effort, --max-budget-usd, --fallback-model	Model selection, calls, usage, quota, and billing
Tools and permissions	--tools, --allowedTools, --allowed-tools, --disallowedTools, --disallowed-tools, --permission-mode, --dangerously-skip-permissions, --allow-dangerously-skip-permissions	Tool inventory and schemas
Prompt and context	--system-prompt, --append-system-prompt, --add-dir, --exclude-dynamic-system-prompt-sections, --disable-slash-commands	Prompt assembly scenarios, Slash commands and automation
Sessions	-c, --continue, -r, --resume, --fork-session, --from-pr, --no-session-persistence, --session-id, --name, --bg	Session resume and transcripts
Models and auth	--model, --fallback-model, --betas	Models, providers, and auth
Settings and integrations	--settings, --setting-sources, --mcp-config, --strict-mcp-config, --plugin-dir, --plugin-url, --agents, --agent, --ide, --chrome, --file	Settings schema reference
Experimental Agent Teams	raw argv --agent-teams; hidden --teammate-mode; child-only --agent-id, --agent-name, --team-name, --agent-color, --parent-session-id, --plan-mode-required, --agent-type	Agent Teams
Accessibility	--ax-screen-reader	Accessibility and screen-reader mode
Cloud/remote and deep links	--cloud, deprecated --remote, --teleport, --remote-control, --rc, --remote-control-session-name-prefix, --prefill, --deep-link-origin	Remote control and teleport

--agent-teams is unusual: isAgentSwarmsEnabled() checks for the exact token in raw process.argv, but no matching Commander .option(...) registration appears in this retained build. It should therefore be treated as an internal/experimental argv signal rather than an advertised help option; CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS is the explicit parsed environment opt-in. --teammate-mode and the identity flags are registered but hidden, primarily for parent/child teammate launches.

Command families

Command	Anchor	Runtime role
auth	H.command("auth")	Login, logout, and status flows for account credentials.
mcp	McpCommandRegistrar	Manages MCP servers and can start the Claude Code MCP server.
mcp serve	command("serve"), line ~9173, byte 0xbf3ccb	Starts the Claude Code MCP server.
mcp add-from-claude-desktop	line ~9173, byte 0xbf44e2	Imports MCP servers from Claude Desktop config.
mcp add OAuth/XAA options	--client-id, --client-secret, --callback-port, --xaa	Adds HTTP/SSE OAuth metadata or gated XAA; see MCP, plugins, and hooks.
mcp xaa setup|login|show|clear	command("xaa")	Manages the gated shared SEP-990 IdP connection and cached token.
plugin / plugins	PluginCommandRegistrar	Manages plugins, marketplaces, local/session plugins, and plugin capabilities.
plugin eval [target] / plugin eval init [name]	command("eval [target]")	Early-access evaluation and suite-authoring harness; see Plugin lifecycle and configuration.
plugin prune / plugin autoremove	command("prune").alias("autoremove")	Removes unneeded auto-installed dependencies, with dry-run/confirmation controls.
project purge [path]	H.command("project")	Deletes project-scoped Claude Code state.
setup-token	H.command("setup-token")	Sets up a long-lived authentication token flow.
agents	H.command("agents")	Opens/manages background agents and dispatched sessions.
ultrareview [target]	H.command("ultrareview [target]")	Runs cloud-hosted multi-agent code review.
auto-mode	config, defaults, critique, reset	Inspects, critiques, or resets auto-mode classifier configuration.
remote-control / rc	H.command("remote-control", {hidden: true}).alias("rc")	Starts local-session Remote Control.
gateway --config <path>	command("gateway")	Starts the native-only enterprise auth, inference, managed-settings, spend, and telemetry gateway. It is registered after the print fast path.
import-conversations <exportPath>	hidden registration near ~978,660	Gated archive import requiring --cwd; --dry-run plans and checks without writes. See Session resume and transcripts.
doctor	H.command("doctor")	Checks auto-updater and related health state.
update / upgrade	H.command("update").alias("upgrade")	Checks for and installs updates.
install [target]	H.command("install [target]")	Installs a stable/latest/specific native build.

agents options

claude agents accepts --json for a non-TTY JSON array and --all to include completed sessions. Dispatch defaults include --agent, --model, --effort, --permission-mode, --settings, --setting-sources, --add-dir, --mcp-config, --strict-mcp-config, and --plugin-dir. The JSON state can expose what a session is waiting for; sandbox, MCP-input, and managed-settings prompts are classified as needing input.

Built-in interactive-command registry

The retained build has two static registration layers plus dynamic additions:

1. Hur() references 105 distinct core names across its conditional branches. A core entry is either local-jsx (TUI component), local (text/control implementation), or a small number of built-in prompt commands.
2. Twenty-nine Lu({...}) registration sites expand to 32 bundled-skill names because one site loops over four Artifact templates. /design exists in both layers, yielding 136 distinct static names across core plus bundled layers.
3. _xo() prepends filesystem skills, dynamic workflows, builtin-plugin skills, plugin skills, and bundled skills before the core array. getCommands() then adds MCP/fallback skills, scopes colliding nested skills, removes shadowed fallbacks, and applies availability plus isEnabled gates.

The counts are an artifact inventory, not a promise that 136 commands appear in one /help menu. Conditional branches do not all coexist, and the inventory deliberately includes account/policy/platform/feature-gated entries, hidden compatibility helpers, and entries whose isEnabled() is false in 2.1.215. Live /help remains authoritative for one running account and surface.

Command kinds and surface rules

Kind	Execution owner	Surface behavior
local-jsx	Interactive Ink/TUI component	Requires a workspace and Ink by default. It cannot execute directly in headless mode; thin clients need an explicit dispatch route or a local twin.
local	JavaScript handler returning text/control events	Enters headless mode only when supportsNonInteractive is true. A same-name local entry often acts as the bridge/headless twin of local-jsx.
prompt with source: "builtin"	Built-in prompt workflow	Can enter headless mode unless disableNonInteractive; model invocation is separately controlled.
prompt with source: "bundled"	Lu() skill registration	Direct slash use requires userInvocable !== false; disableModelInvocation blocks the model/Skill caller but not a direct user invocation. Bundled-skill kill switches can reduce these to user-only or remove them.

Remote mode accepts only thin-client-safe built-in/bundled prompts plus objects in REMOTE_SAFE_COMMANDS. Bridge dispatch similarly uses BRIDGE_SAFE_COMMANDS or a same-name local fallback. This is why a command can exist in the registry yet be unavailable on a particular remote/headless surface.

Core session, context, and interface commands

Command (aliases)	State in 2.1.215	Current role / mechanism owner
/add-dir <path>	Core TUI	Adds a working root; MCP root-change behavior is in MCP, plugins, and hooks.
/autocompact [auto|tokens]	Model/feature gated; TUI + text twin	Reads or writes the effective compaction window; see Context, memory, compaction, checkpoints, and rewind.
/branch [name]	Core TUI	Copies the current chain into a new resumable local session; see Session resume and transcripts.
/btw [question]	Core TUI	Runs a disposable one-turn, no-tool side fork without interrupting the main agent; see Slash commands and automation.
/cd <path>	Core TUI	Relocates the live cwd and refreshes cwd-sensitive roots/state.
/clear [name] (/reset, /new)	Core local; non-interactive capable	Runs the clear lifecycle, rotates the session identity, and leaves the prior transcript resumable; see Session resume and transcripts.
/color [name|default]	TUI + non-interactive twin	Persists a session color and best-effort mirrors its bridge tag.
/compact [instructions]	Core local; non-interactive capable	Runs manual compaction and PreCompact hooks; see Context, memory, compaction, checkpoints, and rewind.
/context [all]	TUI + non-interactive twin	Estimates current model, prompt, tool, MCP, agent, memory, skill, and message token use.
/copy [N]	Core TUI	Copies the latest or Nth-latest assistant response, subject to terminal clipboard support.
/diff	Core TUI	Opens uncommitted and per-turn file-diff views.
/effort [level|auto|ultracode]	Model-capability gated; TUI + text twin	Applies a supported effort level; ultracode combines xhigh with workflow orchestration. See Model selection, calls, usage, quota, and billing.
/exit (/quit)	TUI + local twin	Exits normally, or detaches while a background session keeps running.
/export [filename]	Core TUI	Exports the current conversation to a file or clipboard.
/fast [on|off]	Model/account gated; TUI + text twin	Toggles fast-mode state and its required model; see Model selection, calls, usage, quota, and billing.
/focus	Core TUI; fullscreen-sensitive	Toggles the reduced transcript view and synchronizes the view-mode flag when the transport supports it.
/help	Core TUI	Shows the enabled, advertised command set for the current runtime.
/keybindings	Keybinding-feature gated	Creates or opens ~/.claude/keybindings.json; safe mode delays custom binding activation.
/memory	Core TUI	Opens a memory file in the configured editor.
/model [name]	TUI + non-interactive twin	Validates entitlement, policy, account availability, extended context, and explicit IDs before switching.
/pause-memory (/memory-pause, /toggle-memory)	Disabled (isEnabled: () => false)	Dormant session metadata toggle; do not advertise it as available in this build.
/plan [open|share|description]	Core TUI	Enables plan mode or opens/shares the current plan.
/recap	Feature gated, normally on; non-interactive capable	Runs a one-turn, no-tool fork to generate a one-line session recap.
/release-notes	Core TUI	Opens the packaged release-note view.
/rename [name] (/name)	TUI + non-interactive twin	Persists an explicit name or generates a short name from conversation context.
/resume [id|search] (/continue)	Core TUI	Opens or resolves a resumable conversation; see Session resume and transcripts.
/rewind (/checkpoint, /undo)	Core local interactive handler	Opens the message selector for code and/or conversation restoration; it is not a separate arbitrary undo engine.
/scroll-speed	Terminal/renderer gated	Adjusts mouse-wheel scroll speed where the terminal does not use a known native path.
/status	Core TUI	Shows version, model, account, API connectivity, and tool/integration status.
/tasks (/bashes)	Core TUI	Lists and manages running background tasks.
/theme	Core TUI	Opens the theme picker, including eligible custom themes.
/tui [default|fullscreen]	Core TUI	Chooses the terminal renderer; switching may relaunch and resume the session.
/usage (/cost, /stats)	TUI + non-interactive twin	Shows session totals, plan utilization, and local behavior attribution; see Model selection, calls, usage, quota, and billing.
/wellbeing (/breaks, /break-reminder, /downtime)	Disabled (isEnabled: () => false)	Dormant break/quiet-hours settings UI.
/brief	Feature/account gated	Toggles brief-only output and injects a mode-change reminder into model context.

Core configuration, account, and integration commands

Command (aliases)	State in 2.1.215	Current role / mechanism owner
/advisor [model|off]	Model/account gated	Configures the stronger-model advisor surface.
/artifacts	Artifact-account gated	Browses published/shared artifacts; see Artifact publishing and live pages.
/auto-mode-setup	Auto-mode feature gated; TUI + non-interactive twin	Guided environment/rule proposal flow with reviewed-file application; see Settings, policy, and integrations.
/bug [report] (/share)	Core TUI	Opens bug-report/conversation-sharing UI.
/chrome	Claude.ai/TUI gated	Opens Claude-in-Chrome settings; see Browser automation and Claude in Chrome.
/config [key=value ...] (/settings)	TUI + non-interactive twin	Opens settings or validates and applies exposed shorthand keys; see Settings, policy, and integrations.
/design [subcommand|prompt]	First-party auth/policy gated	Bundled hub plus local consent helpers; see Claude Design and design-system sync.
/design-login	Gated support surface	Acquires design-specific authorization for /design-sync.
/design-consent, /design-revoke	Hidden non-interactive helpers	Record or revoke durable agent_design_projects access.
/feedback [report]	Core TUI	Sends feedback or opens the reporting flow; provider availability can narrow it.
/heapdump	Hidden diagnostic	Writes a V8 heap snapshot and a diagnostics JSON file; see Diagnostics and debug logs.
/hooks	Core TUI	Opens the effective hook-configuration view.
/ide [open]	Core TUI	Shows and manages IDE integration state.
/import [codex|gemini]	Feature gated; TUI + non-interactive twin	Deterministically scans and selectively imports another coding agent’s configuration; see Settings, policy, and integrations.
/init	Built-in prompt workflow	Creates/improves CLAUDE.md and, on the new flow, can propose skills and hooks.
/install-github-app	Account/provider gated	Starts Claude GitHub Actions setup.
/install-slack-app	Claude.ai gated	Opens the Claude Slack Marketplace listing.
/login, /logout	Auth/env gated TUI	Starts account authentication or revokes/clears cached account, design, gateway, and model bootstrap state.
/loops	Disabled (isEnabled: () => false)	Dormant TUI loop manager; /loop is the active bundled skill.
/mcp [reconnect|enable|disable ...]	TUI + non-interactive twin	Manages live MCP clients; see MCP, plugins, and hooks.
/mobile (/ios, /android)	Core TUI	Shows a QR code for the Claude mobile app.
/vim, /output-style	Hidden compatibility shims	Tell the user those settings moved to /config; they are not independent configuration flows.
/passes	Subscriber/referral-cache gated	Opens guest-pass/referral UI only after eligibility has been cached.
/permissions (/allowed-tools)	Core TUI	Manages effective allow/ask/deny rules.
/plugin (/plugins, /marketplace)	Core TUI	Manages plugins and marketplaces; see MCP, plugins, and hooks.
/powerup	Core TUI	Opens interactive product lessons.
/privacy-settings	Consumer-subscriber gated	Views or updates the account privacy setting and can complete a pending terms decision; see Consumer terms and privacy policy.
/radio	Feature gated	Opens Claude FM in the browser.
/reload-plugins [--force]	Core local interactive command	Rebuilds plugin/agent/hook/MCP/LSP runtime state and warns before cache-invalidating MCP changes.
/reload-skills	Core local; non-interactive capable	Clears command caches and reports added/removed filesystem skills.
/sandbox	Platform/policy/dependency gated	Configures command sandboxing; see Sandbox and isolation.
/setup-bedrock, /setup-vertex	Provider-env gated	Reopens provider authentication, region, and model-pin setup only for the active provider.
/skill-doctor	Core local; non-interactive capable	Reports unused loaded skills and stale plugins that consume command-list context.
/skills	Core TUI	Lists available skills.
/statusline	User-only built-in prompt; interactive only	Launches the status-line setup agent; safe mode refuses because user status lines cannot render there. The later runtime is a separate JSON/local-command/render loop.
/stickers	Core local interactive command	Opens the Claude Code sticker page.
/terminal-setup	Terminal-specific TUI	Installs/checks newline, bell, or clipboard integration for the detected terminal.
/upgrade	Claude.ai/subscription gated	Opens the eligible plan-upgrade flow. It is distinct from root claude update.
/usage-credits	Account/billing gated; TUI + text twin	Opens billing management or an admin request flow.
/extra-usage	Hidden rename shim	Reports that the command is now /usage-credits, then delegates.
/voice [hold|tap|off]	OAuth, policy, platform, and audio gated	Configures dictation; see Audio capture and voice.
/web-setup	Claude.ai, remote-policy, and rollout gated	Confirms and imports the local GitHub CLI credential for hosted clone/push access, then bootstraps a default environment when needed; see Remote control and teleport.

Core agents, automation, remote, and internal commands

Command (aliases)	State in 2.1.215	Current role / mechanism owner
/agents	Removed compatibility handler	Explains that the wizard was removed and points to .claude/agents/ or natural-language creation.
/autofix-pr	Subscriber/remote-policy gated	Resolves/subscribes an open PR and creates or reuses a long-running cloud repair session; see Slash commands and automation.
/background [prompt] (/bg)	Core TUI; operational fleet/persistence/seed gates	Checkpoints and hands the current conversation to a daemon worker, transfers eligible task/worktree state, and exits the foreground terminal; see Daemon and background service.
/stop	Background-session only	Persists a stopped terminal state and gracefully detaches/exits while retaining transcript/worktree; see Daemon and background service.
/daemon	Daemon-registry gated	Opens local background-service and routine controls.
/desktop (/app)	Disabled (GTy() === false)	Dormant Desktop handoff surface despite its retained registration.
/fork <directive>	Agent-fleet/coordinator gated	Depending on the fleet branch, creates a background conversation copy or a full-context background fork; see Agents, tasks, and subagents.
/subtask <task>	Agent-fleet/coordinator gated	Runs an inherited-context subagent and returns its result to the parent.
/goal [condition|clear]	Core TUI + text twin	Installs/inspects/removes the session Stop condition; see Slash commands and automation.
/insights	User-only built-in prompt	Generates a local-session usage report; model invocation is disabled.
/pro-trial-expired, /rate-limit-options	Hidden UI entry points	Support account/rate-limit dialogs; not ordinary menu commands.
/remote-control [name] (/rc)	Bridge/auth/policy gated	Starts or disconnects Remote Control; see Remote control and teleport.
/remote-env	Subscriber/remote-policy gated	Lists hosted targets, clears a local override, and writes the chosen default environment to user settings; see Remote control and teleport.
/review [PR]	Built-in prompt	Reviews a GitHub PR; working-tree review belongs to /code-review.
/security-review	Built-in prompt fallback	Runs the bundled security-review instructions when the marketplace implementation is unavailable.
/session (/remote)	Remote-session gated; hidden without fanout	Read-only display of the cloud session URL and a screen-reader-aware QR code; see Remote control and teleport.
/team-onboarding	Organization-policy gated, user-only prompt	Generates ONBOARDING.md from bounded local usage; see Team onboarding and share flows.
/teleport (/tp)	Subscriber/remote-policy gated	Imports/resumes a claude.ai session locally; see Remote control and teleport.
/ultraplan	Bridge entitlement/feature gated	Launches editable remote planning and polls for approval, remote execution, or teleport-back; see Slash commands and automation.
/ultrareview	Cloud-review entitlement gated	Starts the hosted review flow; /code-review ultra is the newer routed surface.
/update (/restart)	Hidden and disabled	Retained relaunch implementation, but isEnabled: () => false in this command registry. Root claude update remains separate.
/workflows	Workflow-feature gated	Opens live/completed dynamic workflow runs; see Dynamic workflows.
/__remote-workflow	Hidden internal transport command	Deterministically compiles and runs an environment-delivered workflow only inside a remote CCR session.

Bundled skill commands

These names come from Lu() rather than Hur(). Lu() can extract bundled files into a skill root, prepend that base directory to the prompt, attach allowed/disallowed tools and hooks, and preserve user/model invocation boundaries. The generic mechanism is documented in Slash commands and automation.

Command(s)	Availability / caller boundary	Mechanism family
/artifact-capabilities, /artifact-design, /artifact-dashboard, /artifact-data-table, /artifact-explainer, /artifact-report, /artifact-pr-review, /plan-artifact	Artifact schema/account gates; user-invocable	Live capability/type references, design guidance, four extracted templates, PR-review template, and plan publishing. See Artifact publishing and live pages.
/dataviz	User- and model-invocable	Extracts chart guidance plus JS/Python palette validators.
/batch <instruction>	User-only	Plans 5–30 independent worktree units, then launches background agents that each verify and open a PR.
/claude-in-chrome	Browser-extension/session gated	Loads browser-tool guidance or a guarded setup/reconnect path before Chrome MCP calls.
/claude-api	Enabled unless explicitly disabled	Detects project language, extracts matching SDK/API references, and exposes migrate / managed-agents-onboard prompt branches.
/claude-code-docs [question]	Rollout gated	Builds a live command/config snapshot and combines it with extracted current-doc references.
/code-review [effort] [--fix] [--comment] [target]	User-only	Routes by model/effort to inline or workflow-backed review; ultra falls back locally when cloud review is unavailable.
/code-walkthrough	Disabled (isEnabled_10() === false)	Dormant explainer-Artifact workflow.
/cowork-plugin	Non-user-invocable; remote Cowork only	Model-only Cowork plugin authoring/customization reference.
/debug [issue]	User-only	Enables debug logging, flushes it, and injects bounded session/daemon evidence for diagnosis.
/design	First-party Design gated; user-only	Bundled hub registered before the core helper, so name resolution reaches its subcommand router first.
/design-sync	Design auth/policy gated; user-only	Extracts converter/validator assets and drives design-system synchronization.
/doctor (/checkup)	User-only; survives bundled-skill kill switch	Runs the repair-capable setup/context/permission audit; root claude doctor remains read-only.
/fewer-permission-prompts	User- and model-invocable	Scans bounded recent transcripts and proposes project permission allows.
/keybindings-help	Non-user-invocable	Model-only keybinding schema/action/reserved-key reference.
/loop (/proactive)	Local scheduling feature gated	Parses fixed intervals into cron or uses self-paced wakeups; see Cron and scheduled tasks.
/memory-types	Non-user-invocable; automemory gated	Model-only memory taxonomy reference.
/pr-explainer	Disabled (isEnabled_13() === false)	Dormant PR walkthrough Artifact workflow.
/run	User- and model-invocable	Finds a project run skill first, otherwise selects a CLI/server/TUI/Electron/browser/library launch pattern.
/run-skill-generator	User-only	Requires an actual launch/interaction, then writes a reusable project run-<unit> skill and driver.
/schedule (/routines)	First-party subscriber/remote-policy/rollout gated	Creates, updates, lists, or runs cloud-agent routines; see Cron and scheduled tasks.
/setup-cowork	Remote Cowork only	Guided role/plugin/connector/skill/writing-voice onboarding.
/simplify [target]	User- and model-invocable	Runs four cleanup angles (parallel when agent tools exist) and applies behavior-preserving fixes.
/update-config	User- and model-invocable	Injects the current settings schema and a verified hook-construction flow.
/verify [target]	User-only	Drives the real runtime surface and captures evidence; it explicitly distinguishes verification from tests/typecheck.

Non-advertised command-shaped definitions

INTERNAL_ONLY_COMMANDS also retains command objects that are deliberately outside Hur(). The notable real names are /commit-push-pr (an internal prompt with narrowly pre-authorized Git/gh patterns) and /version (both TUI and text registrations have isEnabled: () => false). They are implementation support objects, not ordinary shipped slash commands. Stub objects and optional null feature arms are likewise excluded from the catalog above.

Aliases and spelling variants

Alias or variant	Canonical surface
root claude upgrade	root claude update
root claude rc	root claude remote-control
hidden root --remote [description|session_id|url]	hidden root --cloud [description|session_id|url] (deprecated alias)
interactive /rc	interactive /remote-control
--rc [name]	--remote-control [name]
--allowed-tools	--allowedTools
--disallowed-tools	--disallowedTools
plugin autoremove	plugin prune

Interactive /upgrade is not an alias for /update: it is the gated subscription-upgrade UI. The disabled interactive /update object is a separate relaunch implementation. The alias above applies only to the Commander root command family.

--permission-mode manual is accepted as the user-facing spelling of the historical default mode; 2.1.215 --help lists manual and the settings schema continues to accept default.

--disable-slash-commands is registered with the help text Disable all skills. During setup it sets the shared disableSlashCommands state; command/skill aggregators then return no slash-command entries, prompt suggestions are suppressed, and relaunch/remote setup propagates the flag. It does not silently reinterpret /name text as an available command—the command surface is removed for that run.

Parse-time optimization

CommanderRoot has a fast path for -p/--print: when print mode is selected and no cc:// or cc+unix:// URI argument is present, the root options can be parsed before the heavier subcommand tree is registered. This keeps common headless runs lighter.

Related docs

• CLI main paths
• Accessibility and screen-reader mode
• Safe mode and recovery
• Headless streaming and resilience
• Tool inventory and schemas
• Remote control and teleport