40+ Hidden Commands in Claude Code You Can Use Right Now

Claude Code v2.1.88 contains 37 documented slash commands, 18 semi-public commands, and at least 15 hidden commands discovered through source code analysis. This guide catalogs every confirmed and speculative command with copy-paste examples, 60+ environment variables that unlock hidden features, and CLI flags that change how Claude Code operates. Updated April 2026 based on verified community testing.

What Are Claude Code Hidden Commands?

Claude Code is Anthropic's official CLI for AI-assisted development. While /help lists 37 official commands, the leaked source code (v2.1.88, March 31 2026) reveals dozens of additional slash commands, environment variables, and CLI flags that are functional but undocumented. Some are semi-public (mentioned in blog posts but not in /help), others are fully hidden, and a few are internal-only (gated by employee credentials).

This guide separates confirmed commands (tested by the community) from speculative ones (found in source code but unverified). Every entry includes the exact command to type, what it does, and its current status.

All 37 Public Commands (Official)

These commands appear in /help and work for all users on any plan. They form the baseline Claude Code experience.

Semi-Public Commands: Functional but Undocumented

These 18 commands work but do not appear in /help. Some are mentioned in blog posts, Anthropic changelogs, or community guides. All are CONFIRMED functional as of v2.1.88.

Bundled Skills (Invoke Like Slash Commands)

These skills ship with Claude Code and are invoked as slash commands. They are maintained by Anthropic and updated with each release.

Hidden Commands to Test

These commands exist in the source code but are not in /help. Some require specific plans (Max/Pro) or feature flags. We categorize them by confidence level.

High Confidence (Likely Functional)

Medium Confidence (Speculative)

Internal Only (Employee-Gated)

These commands are gated by USER_TYPE=ant (Anthropic employee credentials). Setting this variable manually does not work because verification happens server-side. Listed here for completeness only.

Easter Eggs Found in the Source Code

/stickers -- CONFIRMED

Type /stickers to request physical Anthropic stickers by mail. Confirmed by Boris Cherny (Claude Code dev lead). The first 1,000 people to discover this easter egg received physical stickers. Stock may be exhausted, but the command still exists in the codebase.

Buddy / Crumbshaker -- PARTIALLY RELEASED

A virtual ASCII pet companion that watches you code and occasionally comments in a speech bubble. Details from the source:

• Activation: Gated behind the BUDDY compile-time flag. Not user-activatable.
• Species: 18 types including duck, goose, blob, cat, dragon, octopus, owl, penguin, turtle, snail, ghost, axolotl, capybara, cactus, robot, rabbit, mushroom, and chonk
• Rarity: Common (60%), Uncommon (25%), Rare (10%), Epic (4%), Legendary (1%)
• Shiny variant: Independent 1% chance on any species
• Stats: DEBUGGING, PATIENCE, CHAOS, WISDOM, SNARK (0-100 each)
• Cosmetics: 6 eye styles, 8 hats (none, crown, tophat, propeller, halo, wizard, beanie, tinyduck)
• Your buddy is deterministic: Always the same species for the same account, seeded on your UUID
• Timeline: Teaser April 1-7 2026, full launch May 2026

/good-claude -- SPECULATIVE

A quick positive feedback command or fun response. Found as index.js in the source. Likely an internal test pattern. Try it -- worst case, nothing happens.

Environment Variables That Unlock Features

Claude Code reads over 60 environment variables. Set them in your ~/.bashrc, ~/.zshrc, or in ~/.claude/settings.json under the "env" key for persistence across sessions.

Model and Performance (All CONFIRMED)

Behavior and UX (All CONFIRMED)

Agent Teams and Multi-Agent

Stealth and Privacy

Debug and Profiling

Tools and Limits

Context and Compaction

Provider Routing

Danger Zone (Security Bypass Variables)

Warning: These variables disable security protections. Use only in isolated test environments.

CLI Flags for Power Users

Confirmed Functional Flags

Speculative Flags

Upcoming Features Confirmed in the Code

The source code reveals several major features in development. These are not yet available to users but have significant code already written.

KAIROS -- Always-On Persistent Agent

A background daemon agent that never stops. It receives heartbeat ticks and decides whether to act or sleep. Budget: 15 seconds max blocking the workflow. Has exclusive tools: SleepTool, SendUserFile, PushNotification, SubscribePR. Managed via daemon ps, daemon logs, daemon attach, daemon kill. Over 150 mentions in source. Expected H2 2026.

ULTRAPLAN -- 30-Minute Cloud Planning

Offloads planning to a cloud container (CCR) running Opus 4.6 for up to 30 minutes. Requires browser-based approval (approve/reject). Uses a sentinel token __ULTRAPLAN_TELEPORT_LOCAL__ to teleport results back to the local machine. Expected H2 2026.

Voice Mode -- Rolling Out

Push-to-talk dictation using the Space bar (speech-to-text only, no voice synthesis). Feature flag tengu_amber_quartz in GrowthBook with a kill-switch tengu_amber_quartz_disabled. Currently rolled out to approximately 5% of users. Official docs page exists at code.claude.com/docs/en/voice-dictation. Expected for all users by mid-2026.

AutoDream -- Memory Consolidation

Like REM sleep for AI memory. Triggers automatically after 24 hours plus 5 sessions. Consolidates, deduplicates, and organizes notes between sessions. Enable with "autoDreamEnabled": true in settings.json. Manual trigger: /dream. Feature flag tengu_onyx_plover in GrowthBook.

Claude Mythos (Capybara)

A new model tier ABOVE Opus. Internal codename "Capybara." Referenced as "The most advanced model on the market." Has fast/slow variants including capybara-v2-fast with a 1M token context window. Multiple versions (v2 through v8) have been tested internally.

UDS Inbox -- Inter-Session Communication

Multiple Claude sessions on the same machine can send messages to each other. Addressing via to: "researcher", to: "uds:/.../sock", or to: "bridge:...". Uses ListPeersTool reading ~/.claude/sessions/ for peer discovery.

OpenTelemetry: Monitor Claude Code with Your Own Stack

Claude Code supports full OpenTelemetry integration, letting you export traces, metrics, and logs to Grafana, Datadog, or any OTEL-compatible backend. This is particularly useful for teams tracking AI usage patterns and costs across developers.

The source code reveals 22 OTEL variables are supported. This means Claude Code has a production-grade observability layer that most users never touch. If you run a team and want to track which developers use which tools, how much each session costs, and what patterns of usage emerge, OTEL integration is the way to do it.

Anti-Distillation: What It Is and How to Disable It

The source code reveals an anti-distillation system (ANTI_DISTILLATION_CC flag combined with tengu_anti_distill_fake_tool_injection) that injects fake tools into the system prompt. The purpose is to pollute the training data if a competitor scrapes Claude Code output to train their own models. This feature is transparent to users and does not affect response quality. However, if you want to disable it along with other experimental betas, set:

export CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1

This removes the fake tool injections and disables other beta features that may be running in the background.

settings.json Quick Wins

These three settings in ~/.claude/settings.json deliver the highest impact for the least effort:

{
  "preferences": {
    "autoDreamEnabled": true,
    "alwaysThinkingEnabled": true,
    "autoMemoryEnabled": true
  }
}

• autoDreamEnabled: Claude consolidates its memory files every 24 hours (after 5+ sessions), removing duplicates and organizing notes. Enable this and never manually clean your CLAUDE.md again.
• alwaysThinkingEnabled: The reasoning/thinking step runs on every request, even when you do not explicitly ask for it. This improves code quality on complex tasks at the cost of slightly more tokens.
• autoMemoryEnabled: Claude automatically saves important learnings, decisions, and patterns to MEMORY.md. This builds an institutional memory for your project that persists across sessions.

Top 20 Actions You Can Take Right Now

Ordered by impact. Each can be done in under 2 minutes.

1. /btw + question -- Ask a side question without polluting your main context
2. /insights -- Generate a usage report to understand your patterns
3. /stickers -- Try the easter egg (physical stickers from Anthropic)
4. /tasks -- Persistent todo-list within your session
5. /rewind -- Undo mistakes in both conversation and code
6. /thinkback -- Get your "Spotify Wrapped" style usage retrospective
7. /commit-push-pr -- Ship code in one command instead of three
8. /mobile -- Control Claude Code from your phone while away from your desk
9. /loop 5m /status -- Monitor anything on a recurring interval
10. /schedule -- Set up cron jobs for remote agents
11. ANTHROPIC_MODEL=claude-opus-4-6 -- Force the best model
12. CLAUDE_CODE_EFFORT_LEVEL=high -- Force maximum effort
13. MAX_THINKING_TOKENS=32000 -- More reasoning budget
14. DISABLE_TELEMETRY=1 -- Zero telemetry
15. CLAUDE_DEBUG=1 -- See what Claude is actually doing
16. claude --bare -- Ultra-fast startup for quick questions
17. claude -w -- Isolated Git worktree for parallel work
18. claude --bg "prompt" -- Run tasks in background
19. CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 -- Enable parallel agent teams
20. CLAUDE_CODE_UNDERCOVER=1 -- Remove AI traces from your commits

Copy-Paste Power User Shell Profile

Add this to your ~/.bashrc or ~/.zshrc to unlock the most impactful features permanently:

# === CLAUDE CODE POWER USER CONFIG ===

# --- Model and Performance ---
export ANTHROPIC_MODEL=claude-opus-4-6
export CLAUDE_CODE_SUBAGENT_MODEL=claude-sonnet-4-6
export MAX_THINKING_TOKENS=32000
export CLAUDE_CODE_MAX_OUTPUT_TOKENS=64000
export CLAUDE_CODE_EFFORT_LEVEL=high
export FALLBACK_FOR_ALL_PRIMARY_MODELS=claude-sonnet-4-6

# --- Agent Teams ---
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
export CLAUDE_CODE_PLAN_V2_AGENT_COUNT=3

# --- Privacy ---
export DISABLE_TELEMETRY=1
export DISABLE_ERROR_REPORTING=1
export CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1
export CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY=1

# --- UX ---
export DISABLE_COST_WARNINGS=1
export CLAUDE_CODE_DISABLE_TERMINAL_TITLE=1
export IDLE_THRESHOLD_MINUTES=120

# --- Tools ---
export CLAUDE_CODE_MAX_TOOL_USE_CONCURRENCY=5
export BASH_MAX_OUTPUT_LENGTH=50000
export MCP_TIMEOUT=60000
export MCP_TOOL_TIMEOUT=120000

# --- Aliases ---
alias cc="claude"
alias cco="ANTHROPIC_MODEL=claude-opus-4-6 claude"
alias ccs="ANTHROPIC_MODEL=claude-sonnet-4-6 claude"
alias cch="ANTHROPIC_MODEL=claude-haiku-4-5 claude"
alias ccbare="claude --bare"
alias ccbg="claude --bg"
alias ccw="claude -w"
alias ccdebug="CLAUDE_DEBUG=1 claude"
alias ccstealth="CLAUDE_CODE_UNDERCOVER=1 claude"

Frequently Asked Questions

Can I unlock employee-only features by setting USER_TYPE=ant?

No. While the variable exists in the source code, Anthropic verifies employee status server-side via OAuth credentials. Setting USER_TYPE=ant locally has no effect because the API endpoints check authentication independently. The internal tools (ConfigTool, TungstenTool) and debug commands (/heapdump, /ant-trace) are completely inaccessible to external users.

Are hidden commands safe to try?

Yes, for all commands in our CONFIRMED and SPECULATIVE categories. The worst outcome is the command doing nothing or showing an error message. No hidden command can damage your system or data. However, the "Danger Zone" environment variables (like CLAUDE_CODE_DISABLE_COMMAND_INJECTION_CHECK) genuinely disable security protections and should only be used in isolated environments.

How do I enable AutoDream memory consolidation?

Add "autoDreamEnabled": true to ~/.claude/settings.json under the "preferences" key. AutoDream triggers automatically after 24 hours and at least 5 sessions. You can also trigger it manually with /dream. It consolidates duplicate notes, organizes learnings, and removes stale information from your memory files.

What is the Undercover Mode and why would I use it?

Setting CLAUDE_CODE_UNDERCOVER=1 removes all references to Claude, Anthropic, and AI from commits, PRs, and comments. Originally designed for Anthropic employees working on external repos, it is available to any user. The system prompt becomes "Do not blow your cover." Use it when you want clean commits without "Co-Authored-By: Claude" signatures or when contributing to projects where AI attribution is not required.

Will these commands work in future versions?

This analysis is based on the v2.1.88 source code leaked on March 31, 2026. Anthropic may remove, rename, or gate any of these features in future releases. Commands in the CONFIRMED category have been verified by multiple community members and are most likely to persist. SPECULATIVE commands may disappear or become gated behind new feature flags. Check code.claude.com/docs for official documentation updates.

What is the difference between /compact and DISABLE_AUTO_COMPACT?

The /compact command manually compresses your conversation to free up context window space. By default, Claude Code also auto-compacts when the context reaches a certain percentage of its maximum. Setting DISABLE_AUTO_COMPACT=1 turns off the automatic compaction, giving you full control over when to compact. You can fine-tune the threshold with CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=80 (compact at 80% usage). For long sessions where you need maximum context, disable auto-compaction and compact manually when you notice Claude starting to forget earlier context.

How do I use /schedule to create recurring agents?

The /schedule command creates remote agents that run on a cron schedule. Example: /schedule create --cron "0 3 * * *" --prompt "Consolidate memory and clean obsolete notes" creates an agent that runs daily at 3 AM. You can list scheduled agents with /schedule list, view logs with /schedule logs, and remove them with /schedule delete. This is particularly powerful for automated code reviews, dependency updates, or memory maintenance. Scheduled agents run in Anthropic's cloud infrastructure.

What is Bridge Mode and when should I use it?

Bridge mode (/bridge) creates a connection between your local Claude Code terminal and claude.ai in the browser. This lets you or a team member control a terminal Claude Code session from the web interface. Use /bridge-kick to disconnect any connected client. This is useful for remote pair programming, sharing a Claude session with a colleague, or controlling your development environment from a tablet when you are away from your desk.

Can I use multiple models in the same session?

Yes. Set ANTHROPIC_MODEL=claude-opus-4-6 for your main agent and CLAUDE_CODE_SUBAGENT_MODEL=claude-sonnet-4-6 for sub-agents. This means your primary Claude runs on the most capable model while sub-agents (used for search, file reading, and other lightweight tasks) run on the faster, cheaper Sonnet. You can also use FALLBACK_FOR_ALL_PRIMARY_MODELS=claude-sonnet-4-6 so that if Opus is rate-limited, the session automatically falls back to Sonnet instead of pausing.