Feature Flags Checklist - Complete Overview of 89 Feature Switches

Claude Code source code contains 89 Feature Flags - “secret switches” in the code controlling feature enable/disable. This checklist organizes all 89 Flags by functional domain for easy reference.

Flag Maturity Distribution

89 Flags by reference count maturity distribution:

Domain 1: Autonomous Agents and Background Execution (18 Flags)

Core Flags

KAIROS — 154 references Background autonomous Agent mode, flagship feature of Claude Code’s evolution toward “background intelligence.”

• Terminal focus awareness (detects if user is looking at terminal)
• Timed awakening (Tick mechanism)
• Brief communication (Brief mode)
• Multi-channel communication (Channels)
• GitHub Webhook subscription

Major Flags

PROACTIVE — 35 references Proactive work mode where Agent can autonomously execute tasks without user initiative.

COORDINATOR_MODE — 28 references Cross-Agent coordination mode supporting multi-Agent collaboration for complex projects.

BG_SESSIONS — 22 references Background session management supporting sessions running continuously in background.

AGENT_TRIGGERS — 18 references Timed triggers executing Agent based on time or events.

Other Flags

Domain 2: Remote Control and Distributed Execution (14 Flags)

Core Flags

BRIDGE_MODE — 28 references Remote control mode for remotely controlling Claude Code instances via network protocol.

• Bridge protocol: remote control communication
• UDS_INBOX: Unix Domain Socket communication
• SSH_REMOTE: SSH remote connection

Major Flags

DAEMON — 24 references Daemon mode where Claude Code runs as a background service.

UDS_INBOX — 19 references Unix Domain Socket communication for local inter-process communication.

SSH_REMOTE — 16 references SSH remote connection controlling remote instances via SSH tunnel.

Other Flags

Domain 3: Context Management and Performance Optimization (17 Flags)

Core Flags

TRANSCRIPT_CLASSIFIER — 107 references Intelligent permission classifier, core component implementing auto permission mode.

• Between plan (confirm all) and auto-accept (accept all)
• AI judges operation risk level
• Safe operations auto-permitted, dangerous operations require confirmation

Major Flags

CONTEXT_COLLAPSE — 20 references Fine-grained context collapse, selectively collapsing unimportant tool results.

CACHED_MICROCOMPACT — 18 references Cache-aware microcompaction optimizing cache hit rate.

TOKEN_BUDGET — 15 references Token budget tracking for fine-grained token usage monitoring.

PROMPT_CACHE_BREAK_DETECTION — 14 references Cache break detection tracking prompt changes causing cache invalidation.

Other Flags

Domain 4: Memory and Knowledge Management (9 Flags)

Core Flags

TEAMMEM — 51 references Team memory system for cross-session memory sharing and team knowledge base.

• Automatic memory extraction
• Sensitive information protection (Secret Guard)
• Memory synchronization

Major Flags

EXTRACT_MEMORIES — 23 references Automatic memory extraction automatically analyzing and extracting information worth persisting at end of each turn.

EXPERIMENTAL_SKILL_SEARCH — 14 references Experimental skill search for enhanced skill discovery and retrieval capabilities.

SKILL_IMPROVEMENT — 11 references Automatic skill improvement optimizing skills based on usage feedback.

Other Flags

Domain 5: UI/UX and Platform Capabilities (31 Flags)

Core Flags

VOICE_MODE — 46 references Voice input mode supporting voice-to-text input.

• Streaming voice-to-text
• Push-to-talk shortcut (spacebar)

Major Flags

WEB_BROWSER_TOOL — 18 references Built-in browser based on Bun WebView API.

TERMINAL_PANEL — 16 references Terminal panel for enhanced terminal interface.

CHICAGO_MCP — 15 references Computer Use MCP integration supporting MCP protocol tool integration.

POWERSHELL_AUTO_MODE — 12 references PowerShell auto mode for Windows platform optimization.

UI/UX Other Flags

Platform Capability Flags

Experimental Flags (Selected)

The following Flags are in prototype exploration stage (1-2 references), possibly unstable:

User-Configurable Flags

The following Flags can be enabled via environment variables or settings files:

Environment Variable Configuration

# Feature switches
export CLAUDE_CODE_COORDINATOR_MODE=true
export CLAUDE_CODE_VOICE_ENABLED=true
export CLAUDE_CODE_ALWAYS_THINKING=true
export CLAUDE_CODE_EFFORT_LEVEL=high

settings.json Configuration

{
  "voiceEnabled": true,
  "alwaysThinkingEnabled": true,
  "experimentalSkills": true
}

Dependency Relationships Between Flags

Some Flags have dependency relationships:

DAEMON → requires BRIDGE_MODE
KAIROS_DREAM → can be independent of KAIROS
CCR_MIRROR → BRIDGE_MODE sub-mode
AGENT_TRIGGERS_REMOTE → AGENT_TRIGGERS extension
COORDINATOR_V2 → requires COORDINATOR_MODE
PROACTIVE_SUGGESTIONS → requires PROACTIVE
CONTEXT_COLLAPSE → requires CACHED_MICROCOMPACT

Dependency Design Principles:

• Sub-features can be enabled independently of parent features
• Hard dependencies expressed with &&
• Soft associations expressed with ||

Build Configuration and Flags

Public Build

Most Flags are false, only basic features enabled.

Internal Build (ant)

When USER_TYPE === 'ant':

• KAIROS
• Experimental skills
• A/B testing features
• Internal tools

Experimental Build

Specific Flag combinations for A/B testing:

• TREE_SITTER_BASH vs TREE_SITTER_BASH_SHADOW
• ABLATION_BASELINE: Ablation experiment baseline

How to Check Current Flag Status

# View in Claude Code
/flags

# Or check environment variables
echo $USER_TYPE

# Check settings file
cat ~/.claude/settings.json

Flag Evolution Trends

From the distribution of 89 Flags, Claude Code’s development direction is clear:

1. Background autonomy: KAIROS-related Flags are most numerous, indicating key direction
2. Intelligent permissions: TRANSCRIPT_CLASSIFIER deeply integrated, intelligent decision is the trend
3. Team collaboration: TEAMMEM and COORDINATOR_MODE, multi-Agent collaboration is key
4. Cross-platform: UI/UX domain Flags numerous, platform adaptation is ongoing work
5. Experimental exploration: 47% of Flags in prototype stage, many features being explored

Implications for Users

Anticipating Development Direction

• KAIROS-related Flags most numerous → Background autonomous Agent is the future focus
• VOICE_MODE has many references → Voice interaction will continue improving
• Few but concentrated memory system Flags → Memory is infrastructure, will iterate stably

Understanding Experimental Features

• 1-2 reference Flags: very experimental, may change anytime
• 10-29 reference Flags: relatively mature, can try
• 100+ reference Flags: core features, stable and reliable

Configuration Recommendations

• Production environment: only use mature Flags (30+ references)
• Personal exploration: can try medium-maturity Flags (10-29 references)
• Experimental testing: can accept prototype Flags (1-2 references), but be prepared

This checklist covers all 89 Feature Flags in Claude Code. Understanding these Flags enables you to:

• Anticipate Claude Code’s development direction
• Understand experimental feature implementation mechanisms
• Reasonably configure advanced features
• Apply similar techniques in your own projects