Appendix C: Feature Flag Reference

Feature flags are the core mechanism for implementing compile-time dead code elimination in Claude Code. All flags are injected as boolean values by the compiler during the bundling phase: when a flag is false, the code branch it guards is removed entirely, thereby reducing artifact size and protecting unreleased features.

Type Definitions:

• Compile-time: The flag is evaluated during the Bun bundling phase; code branches that resolve to false are completely absent from the output artifact.
• Compile-time + Runtime Gating: The flag is true at compile time, but additional runtime conditions (such as environment variables or server-side configuration) must also be satisfied to truly activate the feature.

How to Use This Reference:

• Navigate to the target flag using the C.1 - C.12 category sections
• The “Impact Scope” column describes the specific capabilities enabled when the flag is activated
• The “Dependencies” column describes prerequisite relationships between flags
• The “Tool Enablement Conditions Reference” table in Appendix B lists the specific tools enabled by each flag and can be used for cross-referencing

---

C.1 Core Interaction Mode Flags

Core interaction mode flags control the primary operating modes of Claude Code. These flags define how the system interacts with users, IDEs, or other systems.

Feature Flag	Type	Description	Impact Scope	Dependencies	Related Chapter
PROACTIVE	Compile-time	Proactive mode. When enabled, the agent can proactively offer suggestions and automatically execute background tasks during user idle periods	Enables SleepTool, proactive suggestion logic, background task triggers	No prerequisites	Chapter 5
KAIROS	Compile-time + Runtime Gating	Assistant mode (formerly Harbor). IDE-oriented channel-based collaboration, including channel messaging, session recovery, and the complete feature set	Enables channel communication, session recovery, SleepTool, SendUserFileTool, PushNotificationTool, and the full feature set	No prerequisites, though some sub-features require their own sub-flags	Chapter 6
KAIROS_BRIEF	Compile-time	KAIROS lightweight subset. Only enables basic capabilities such as session recovery, without the full channel functionality	Session recovery and basic interaction only	Depends on KAIROS context	Chapter 6
KAIROS_CHANNELS	Compile-time + Runtime Gating	Channel functionality toggle independent of the full KAIROS feature set, allowing channel messaging alone to be enabled	Sending, receiving, and routing of channel messages	Can be enabled independently of the full KAIROS feature set	Chapter 6
KAIROS_DREAM	Compile-time	KAIROS Dream mode, used for loading additional skill sets	Loading and registration of Dream-specific skill sets	Depends on KAIROS	Chapter 6
KAIROS_GITHUB_WEBHOOKS	Compile-time	GitHub Webhook integration. Receives GitHub event pushes in KAIROS mode	SubscribePRTool, GitHub event listening and routing	Depends on KAIROS	Chapter 6
KAIROS_PUSH_NOTIFICATION	Compile-time	KAIROS push notification capability	Notification logic for PushNotificationTool	Depends on KAIROS	Chapter 6
BRIDGE_MODE	Compile-time + Runtime Gating	IDE bridge mode. When enabled, the REPL communicates with the IDE (e.g., VS Code) via a bridging protocol, supporting permission callback pipelines	IDE bidirectional communication, JWT authentication, permission callbacks, state synchronization	Requires an IDE plugin	Chapter 7
COORDINATOR_MODE	Compile-time + Runtime Gating	Coordinator mode. Acts as a coordinator in multi-agent scenarios, distributing tasks and aggregating results	AgentTool, TaskStopTool, SendMessageTool coordinator extensions	No prerequisites	Chapter 8
VOICE_MODE	Compile-time	Voice mode. Enables voice input/output, push-to-talk key bindings, and voice integration modules	Voice input capture, speech synthesis output, PTT key bindings	Requires microphone and speaker hardware support	Chapter 12
DAEMON	Compile-time	Background daemon mode. Allows Claude Code to run long-term as a daemon process	Daemon startup logic, persistent services, scheduled task framework	No prerequisites	Chapter 11
BUDDY	Compile-time	Companion Sprite. Displays an interactive animated character in the REPL interface, providing emotional feedback	Buddy component rendering, animation state machine, interaction event handling	No prerequisites	Chapter 12

KAIROS Flag Family Relationship Diagram:

flowchart TD
    KAIROS["KAIROS<br/>Master Flag"] --> KB["KAIROS_BRIEF<br/>Lightweight Subset"]
    KAIROS --> KC["KAIROS_CHANNELS<br/>Channel Communication<br/>Can Be Enabled Independently"]
    KAIROS --> KD["KAIROS_DREAM<br/>Dream Skill Set"]
    KAIROS --> KG["KAIROS_GITHUB_WEBHOOKS<br/>GitHub Integration"]
    KAIROS --> KP["KAIROS_PUSH_NOTIFICATION<br/>Push Notifications"]

    classDef main fill:#4a90d9,stroke:#2c5f8a,color:#fff
    classDef sub fill:#7eb8da,stroke:#4a90d9,color:#fff

    class KAIROS main
    class KB,KC,KD,KG,KP sub

C.2 Agent and Subtask Flags

This group of flags controls the agent’s task decomposition, scheduling, and automated execution capabilities.

Feature Flag	Type	Description	Impact Scope	Dependencies	Related Chapter
FORK_SUBAGENT	Compile-time	Fork sub-agent. Allows creating independent sub-agents via a fork mechanism within a conversation to handle subtasks	AgentTool fork capability, sub-agent context isolation	No prerequisites	Chapter 8
AGENT_TRIGGERS	Compile-time	Agent triggers. Enables scheduled task dispatch (cron) and periodic background agent execution	CronCreateTool, CronDeleteTool, CronListTool	Depends on DAEMON or a long-running environment	Chapter 9
AGENT_TRIGGERS_REMOTE	Compile-time	Remote agent triggers. Supports execution of remotely hosted scheduled triggers	RemoteTriggerTool, remote trigger communication protocol	Depends on AGENT_TRIGGERS	Chapter 9
ULTRAPLAN	Compile-time	Ultra-planning mode. Provides an interactive planning dialog that allows users to review and modify execution plans for complex tasks	Enhanced planning interface for EnterPlanModeTool/ExitPlanModeV2Tool	No prerequisites	Chapter 5
VERIFICATION_AGENT	Compile-time	Verification agent. Automatically initiates a verification process after task completion	Post-task automatic verification logic, verification result reports	No prerequisites	Chapter 8
BUILTIN_EXPLORE_PLAN_AGENTS	Compile-time	Built-in explore-plan agents. Provides pre-installed exploration and planning sub-agents	ExploreAgent, PlanAgent, and other built-in agent definitions	Depends on FORK_SUBAGENT	Chapter 8
AGENT_MEMORY_SNAPSHOT	Compile-time	Agent memory snapshot. Supports snapshotting and restoring agent state	Snapshot creation, serialized storage, restore loading logic	No prerequisites	Chapter 10
WORKFLOW_SCRIPTS	Compile-time	Workflow scripts. Enables local workflow task processors, supporting automated script orchestration	WorkflowTool, workflow script parsing and execution engine	No prerequisites	Chapter 9
TEMPLATES	Compile-time	Template system. Enables the Job Classifier for identifying and routing different types of user requests	Job Classifier, request type identification, routing dispatch	No prerequisites	Chapter 9

Agent Flag Dependency Relationships:

flowchart TD
    FS["FORK_SUBAGENT"] --> BE["BUILTIN_EXPLORE_PLAN_AGENTS<br/>Requires fork capability"]
    AT["AGENT_TRIGGERS"] --> AR["AGENT_TRIGGERS_REMOTE<br/>Requires local trigger foundation"]

    classDef root fill:#4a90d9,stroke:#2c5f8a,color:#fff
    classDef child fill:#7eb8da,stroke:#4a90d9,color:#fff

    class FS,AT root
    class BE,AR child

C.3 Context Management and Compression Flags

Context management flags control how Claude Code handles the token limits of the context window. Together, these flags form a multi-layered compression strategy system, ranging from preventive lightweight compression to emergency aggressive pruning.

Feature Flag	Type	Description	Impact Scope	Dependencies	Related Chapter
REACTIVE_COMPACT	Compile-time + Runtime Gating	Reactive compression. Automatically triggers context compression when tokens approach the threshold, rather than waiting for user confirmation	Automatic compression trigger logic, threshold detector	No prerequisites	Chapter 4
CONTEXT_COLLAPSE	Compile-time	Context collapse. Provides a more aggressive context reduction strategy than traditional compression, including a dedicated collapse UI and recovery mechanism	CtxInspectTool, collapse UI component, content recovery logic	No prerequisites	Chapter 4
CACHED_MICROCOMPACT	Compile-time	Cached micro-compaction. Maintains prompt cache boundaries during micro-compaction, avoiding the overhead of cache invalidation	Micro-compaction cache boundary markers, cache-aware compression strategy	Depends on Prompt Cache API support	Chapter 4
HISTORY_SNIP	Compile-time	History pruning (Snip Compact). Intelligently prunes processed conversation history, retaining key information while significantly reducing token usage	SnipTool, history message pruning algorithm	No prerequisites	Chapter 4
COMPACTION_REMINDERS	Compile-time	Compression reminders. Displays reminder messages to the user during the compression process	Pre- and post-compression user notification UI	Depends on compression operations	Chapter 4
PROMPT_CACHE_BREAK_DETECTION	Compile-time	Prompt cache break detection. Detects and reports prompt cache boundary breaks during compression operations	Cache breakpoint detection logic, break report generation	Depends on compression operations	Chapter 4
TOKEN_BUDGET	Compile-time	Token budget manager. Tracks and visualizes token usage budgets, providing budget overrun warnings	Token usage tracking, budget warning UI, usage statistics	No prerequisites	Chapter 4
COMMIT_ATTRIBUTION	Compile-time	Commit attribution. Adds pre- and post-compression context attribution markers to code commits after compression	Compression context to commit message mapping logic	Depends on compression operations	Chapter 4

Compression Flag Strategy Matrix:

flowchart LR
    subgraph CompressionIntensity["Compression Intensity (Light to Heavy)"]
        direction LR
        MC["micro-compact<br/>CACHED_MICROCOMPACT<br/>Light - Cache-Friendly"] --> RC["reactive compact<br/>REACTIVE_COMPACT<br/>Medium - Auto-Triggered"]
        RC --> HS["history snip<br/>HISTORY_SNIP<br/>Medium-Heavy - Continuous Pruning"]
        HS --> CC["context collapse<br/>CONTEXT_COLLAPSE<br/>Extreme - Emergency Reduction"]
    end

    classDef light fill:#c8e6c9,stroke:#4caf50,color:#333
    classDef medium fill:#fff9c4,stroke:#fbc02d,color:#333
    classDef heavy fill:#ffccbc,stroke:#e64a19,color:#333
    classDef extreme fill:#ef5350,stroke:#b71c1c,color:#fff

    class MC light
    class RC medium
    class HS heavy
    class CC extreme

Compression Strategy	Corresponding Flag	Compression Intensity	Cache-Friendly	Applicable Scenario
micro-compact	CACHED_MICROCOMPACT	Light	High	Preventive compression when approaching the threshold
reactive compact	REACTIVE_COMPACT	Medium	Medium	Automatically triggered standard compression
history snip	HISTORY_SNIP	Medium-Heavy	Low	Intelligent pruning of history messages
context collapse	CONTEXT_COLLAPSE	Extreme	Low	Aggressive reduction in emergency situations

Recommended Configuration: Enabling REACTIVE_COMPACT + CACHED_MICROCOMPACT + TOKEN_BUDGET together provides the best context management experience, offering both preventive protection and budget visualization.

C.4 Permission and Security Flags

Permission and security flags control the agent’s autonomous execution boundaries and security auditing capabilities. Together, these flags define the Claude Code trust model.

Feature Flag	Type	Description	Impact Scope	Dependencies	Related Chapter
TRANSCRIPT_CLASSIFIER	Compile-time	Transcript classifier. Automatically determines the permission mode (including auto mode) based on conversation content, replacing purely manual permission switching	Permission mode automatic inference, auto mode decision logic	No prerequisites	Chapter 3
BASH_CLASSIFIER	Compile-time	Bash classifier. Performs security classification on Bash commands; high-confidence safe commands can be automatically approved	Bash command security assessment, read-only command auto-approval	No prerequisites	Chapter 3
HARD_FAIL	Compile-time	Hard fail mode. Terminates directly on critical errors rather than degrading gracefully	Error handling strategy, critical fault termination logic	No prerequisites	Chapter 3
NATIVE_CLIENT_ATTESTATION	Compile-time	Native client attestation. Enables platform-native client identity verification mechanisms	Client identity verification, platform security integration	Requires platform security framework support	Chapter 3
ANTI_DISTILLATION_CC	Compile-time	Anti-distillation protection. Prevents model outputs from being used for model distillation attacks	Output watermarking, distillation detection logic	No prerequisites	Chapter 3

Security Flag Combination Recommendations:

Security Level	Recommended Combination	Description
Maximum Security (Enterprise Environment)	TRANSCRIPT_CLASSIFIER + BASH_CLASSIFIER + HARD_FAIL + NATIVE_CLIENT_ATTESTATION + ANTI_DISTILLATION_CC	All security features enabled
Standard Security	TRANSCRIPT_CLASSIFIER + BASH_CLASSIFIER	Automated classification assistance, reduced manual intervention
Development Debugging	HARD_FAIL only	Quickly surface errors for easier debugging

C.5 Tool and Skill Flags

Tool and skill flags control the extensibility of available tools in Claude Code and tool behaviors.

Feature Flag	Type	Description	Impact Scope	Dependencies	Related Chapter
MONITOR_TOOL	Compile-time	Monitor tool. Provides monitoring capability when the Bash tool executes background tasks	MonitorTool, background task output monitoring	No prerequisites	Chapter 7
WEB_BROWSER_TOOL	Compile-time	Web browser tool. Enables the built-in browser panel, supporting web browsing and content extraction	WebBrowserTool, browser panel UI	No prerequisites	Chapter 7
MCP_SKILLS	Compile-time	MCP skill discovery. Allows dynamic discovery and loading of skills from MCP servers	MCP skill discovery protocol, dynamic skill registration	Depends on MCP connection	Chapter 7
EXPERIMENTAL_SKILL_SEARCH	Compile-time	Experimental skill search. Enables semantic-based skill indexing and search capabilities	Semantic skill indexing, skill search engine	No prerequisites	Chapter 7
SKILL_IMPROVEMENT	Compile-time	Skill improvement. Supports automatic optimization and iteration of installed skills	Skill auto-optimization logic, iterative improvement engine	Depends on skill system	Chapter 7
RUN_SKILL_GENERATOR	Compile-time	Skill generator runner. Supports dynamic generation of new skills	Skill generation tool, dynamic skill creation workflow	Depends on skill system	Chapter 7
BUILDING_CLAUDE_APPS	Compile-time	Claude app building mode. Loads a dedicated skill set for building Claude applications	Claude Apps dedicated skill set, application templates	Depends on skill system	Chapter 7
REVIEW_ARTIFACT	Compile-time	Review artifact. Loads code review-related skills	Code review skill set, review templates	Depends on skill system	Chapter 7
HOOK_PROMPTS	Compile-time	Hook prompts. Allows hooks to inject custom prompts into the conversation flow	Hook prompt injection mechanism, dynamic prompt extension	Depends on hook system	Chapter 9
CONNECTOR_TEXT	Compile-time	Connector text blocks. Supports rendering special connector text block types within the message stream	Connector text renderer, special text block types	No prerequisites	Chapter 7
UDS_INBOX	Compile-time	Unix Domain Socket inbox. Receives messages from other processes via UDS	ListPeersTool, UDS message listener	Requires UDS system support	Chapter 7
MCP_RICH_OUTPUT	Compile-time	MCP rich output. Allows MCP tools to return structured rich media content	MCP output format extension, rich media rendering	Depends on MCP connection	Chapter 7
TREE_SITTER_BASH	Compile-time	Tree-sitter Bash parsing. Uses Tree-sitter for precise AST parsing of Bash commands	Bash command AST parser, precise security analysis	No prerequisites	Chapter 3
TREE_SITTER_BASH_SHADOW	Compile-time + Runtime Gating	Tree-sitter Bash shadow mode. Runs Tree-sitter in parallel alongside the existing parser for result comparison and verification	Parallel parsing comparison, result consistency verification	Depends on TREE_SITTER_BASH	Chapter 3

Skill Flag Relationship Diagram:

flowchart TD
    Skill["Skill System<br/>Foundation"] --> MCPS["MCP_SKILLS<br/>Discover Skills from MCP"]
    Skill --> ESS["EXPERIMENTAL_SKILL_SEARCH<br/>Semantic Skill Search"]
    Skill --> SI["SKILL_IMPROVEMENT<br/>Automatic Skill Optimization"]
    Skill --> RSG["RUN_SKILL_GENERATOR<br/>Dynamic Skill Generation"]
    Skill --> BCA["BUILDING_CLAUDE_APPS<br/>Claude Apps Dedicated Skills"]
    Skill --> RA["REVIEW_ARTIFACT<br/>Review Dedicated Skills"]

    classDef root fill:#4a90d9,stroke:#2c5f8a,color:#fff
    classDef sub fill:#7eb8da,stroke:#4a90d9,color:#fff

    class Skill root
    class MCPS,ESS,SI,RSG,BCA,RA sub

C.6 Session and Persistence Flags

Session and persistence flags control the lifecycle management and state persistence capabilities of Claude Code sessions.

Feature Flag	Type	Description	Impact Scope	Dependencies	Related Chapter
BG_SESSIONS	Compile-time	Background sessions. Supports maintaining independent session instances in the background, allowing long-running tasks to execute detached from the foreground	Background session manager, session serialization and recovery	No prerequisites	Chapter 11
AWAY_SUMMARY	Compile-time	Away summary. Automatically generates a conversation summary covering the period when the user was away, displayed upon their return	Away detection, summary generation, return display	Depends on session persistence	Chapter 11
FILE_PERSISTENCE	Compile-time	File persistence. Enables session-level file persistence tracking	File change tracking, state consistency during session recovery	No prerequisites	Chapter 11
NEW_INIT	Compile-time	New initialization flow. Uses an improved session initialization logic	Session initialization flow, startup optimization	No prerequisites	Chapter 11

Recommended Configuration: For scenarios requiring long-running tasks, it is recommended to enable BG_SESSIONS + AWAY_SUMMARY + FILE_PERSISTENCE together to ensure tasks can reliably execute in the background and seamlessly resume when the user returns.

C.7 Memory and Knowledge Management Flags

Memory and knowledge management flags control Claude Code’s cross-session knowledge storage, retrieval, and sharing capabilities.

Feature Flag	Type	Description	Impact Scope	Dependencies	Related Chapter
TEAMMEM	Compile-time + Runtime Gating	Team memory. Enables a team-level shared memory file system, supporting read/write and synchronization of a team knowledge base	Team memory file read/write, knowledge base synchronization mechanism	No prerequisites	Chapter 10
EXTRACT_MEMORIES	Compile-time	Memory extraction. Automatically extracts reusable knowledge fragments from conversations at session end and writes them to memory files	Automatic knowledge extraction, memory file writing	Depends on memory system	Chapter 10
LODESTONE	Compile-time	Lodestone. Enables enhanced memory retrieval and matching mechanisms	Memory relevance scoring, enhanced retrieval algorithm	Depends on memory system	Chapter 10
MEMORY_SHAPE_TELEMETRY	Compile-time	Memory shape telemetry. Collects anonymous telemetry data on memory file shape and usage patterns	Memory file statistical analysis, anonymous telemetry reporting	Depends on memory system	Chapter 10

Memory Flag Enhancement Path: Enabling LODESTONE improves the relevance accuracy of memory retrieval; combined with EXTRACT_MEMORIES, knowledge can be automatically distilled from conversations. In a team environment, additionally enabling TEAMMEM enables team-level knowledge sharing.

C.8 Remote and Connectivity Flags

Remote and connectivity flags control Claude Code’s connectivity capabilities across different network environments and deployment scenarios.

Feature Flag	Type	Description	Impact Scope	Dependencies	Related Chapter
SSH_REMOTE	Compile-time	SSH remote mode. Supports connecting to remote machines via SSH to run Claude Code	SSH connection management, remote environment adaptation	Requires SSH infrastructure	Chapter 11
DIRECT_CONNECT	Compile-time	Direct connect mode. Supports bypassing proxies to connect directly to the Anthropic API	Proxy bypass logic, direct connection network configuration	No prerequisites	Chapter 11
CHICAGO_MCP	Compile-time	Chicago MCP. Enables a specific MCP server configuration and Computer Use integration	Computer Use toolset, dedicated MCP configuration	Depends on MCP integration	Chapter 7
CCR_AUTO_CONNECT	Compile-time	CCR auto-connect. Automatically establishes a connection to the CCR (Claude Code Remote) service	CCR service auto-discovery, connection establishment	Depends on CCR service	Chapter 11
CCR_MIRROR	Compile-time	CCR mirror. Supports bidirectional mirroring of session state between local and remote	Session state bidirectional synchronization, conflict resolution	Depends on CCR service	Chapter 11
CCR_REMOTE_SETUP	Compile-time	CCR remote setup. Enables a one-click remote environment configuration flow	Remote environment auto-configuration, dependency installation	Depends on CCR service	Chapter 11
SELF_HOSTED_RUNNER	Compile-time	Self-hosted runner. Supports running agents on self-hosted infrastructure	Self-hosted execution environment, infrastructure adaptation	No prerequisites	Chapter 11
BYOC_ENVIRONMENT_RUNNER	Compile-time	BYOC environment runner. Supports agent execution in “Bring Your Own Cloud” environments	BYOC environment integration, multi-cloud adaptation	No prerequisites	Chapter 11

Remote Deployment Scenarios and Flag Combinations:

Deployment Scenario	Recommended Flag Combination
Local development	No remote flags needed
SSH remote development	SSH_REMOTE
CCR cloud development	CCR_AUTO_CONNECT + CCR_MIRROR + CCR_REMOTE_SETUP
Self-hosted server	SELF_HOSTED_RUNNER + DIRECT_CONNECT
BYOC enterprise environment	BYOC_ENVIRONMENT_RUNNER + DIRECT_CONNECT

C.9 UI and Interface Flags

UI and interface flags control the visual presentation and interaction capabilities of the Claude Code terminal interface.

Feature Flag	Type	Description	Impact Scope	Dependencies	Related Chapter
MESSAGE_ACTIONS	Compile-time + Runtime Gating	Message actions. Enables contextual action buttons on messages (e.g., copy, regenerate, etc.)	Message action button UI, context menus	No prerequisites	Chapter 12
TERMINAL_PANEL	Compile-time + Runtime Gating	Terminal panel. Enables an independent terminal panel in full-screen layout (shortcut Meta+J)	TerminalCaptureTool, terminal panel component, full-screen layout	Depends on full-screen terminal environment	Chapter 12
QUICK_SEARCH	Compile-time	Quick search. Enables in-conversation quick search functionality	Search UI, conversation content indexing	No prerequisites	Chapter 12
HISTORY_PICKER	Compile-time	History picker. Provides a visual conversation history browsing and switching interface	History browsing UI, session switcher	No prerequisites	Chapter 12
AUTO_THEME	Compile-time + Runtime Gating	Auto theme. Automatically switches between light and dark themes based on system preferences	Theme detection, auto-switching logic	Requires system theme API support	Chapter 12
STREAMLINED_OUTPUT	Compile-time	Streamlined output. Reduces redundant visual elements in the interface, providing a more compact output style	Compact output rendering, visual element reduction	No prerequisites	Chapter 12
NATIVE_CLIPBOARD_IMAGE	Compile-time	Native clipboard image. Supports pasting images directly from the system clipboard into conversations	Clipboard image reading, image format conversion	Requires platform clipboard API support	Chapter 12

C.10 Settings Synchronization Flags

Settings synchronization flags control the bidirectional synchronization of user configurations between local and cloud storage.

Feature Flag	Type	Description	Impact Scope	Dependencies	Related Chapter
UPLOAD_USER_SETTINGS	Compile-time	Upload user settings. Syncs local user settings to the cloud	Local-to-cloud configuration upload, conflict detection	Requires cloud service	Chapter 10
DOWNLOAD_USER_SETTINGS	Compile-time + Runtime Gating	Download user settings. Pulls and applies user settings from the cloud to the local environment	Cloud-to-local configuration download, configuration merging	Depends on UPLOAD_USER_SETTINGS	Chapter 10

Note: Settings synchronization typically requires both flags to be enabled together for complete bidirectional sync. UPLOAD_USER_SETTINGS is responsible for pushing local changes to the cloud, while DOWNLOAD_USER_SETTINGS is responsible for pulling existing configurations in a new environment.

C.11 Telemetry and Diagnostic Flags

Telemetry and diagnostic flags control Claude Code’s runtime data collection, performance profiling, and debugging capabilities. These flags are primarily used for internal quality assurance and performance optimization.

Feature Flag	Type	Description	Impact Scope	Dependencies	Related Chapter
COWORKER_TYPE_TELEMETRY	Compile-time	Coworker type telemetry. Collects and reports type information about coworkers (e.g., IDE, terminal)	Collaboration environment detection, anonymous type reporting	No prerequisites	Chapter 13
ENHANCED_TELEMETRY_BETA	Compile-time	Enhanced telemetry Beta. Enables extended anonymous usage telemetry data collection	Extended telemetry data collection, anonymous statistics	No prerequisites	Chapter 13
PERFETTO_TRACING	Compile-time	Perfetto tracing. Integrates the Chrome Perfetto tracing framework for performance profiling and timeline visualization	Perfetto tracing integration, performance data export	No prerequisites	Chapter 13
SHOT_STATS	Compile-time	Shot statistics. Collects and displays detailed statistics for each API call	API call statistics, latency analysis	No prerequisites	Chapter 13
SLOW_OPERATION_LOGGING	Compile-time	Slow operation logging. Records operations whose execution time exceeds a threshold to assist with performance diagnostics	Slow operation detection, threshold alerts, performance logging	No prerequisites	Chapter 13
ABLATION_BASELINE	Compile-time	Ablation baseline. Serves as the baseline control group in A/B experiments for evaluating the impact of new features	A/B experiment framework, baseline data collection	No prerequisites	Chapter 13

Diagnostic Flag Combination Recommendations:

Diagnostic Purpose	Recommended Combination
Performance profiling	PERFETTO_TRACING + SHOT_STATS + SLOW_OPERATION_LOGGING
Usage statistics	COWORKER_TYPE_TELEMETRY + ENHANCED_TELEMETRY_BETA
A/B experiments	ABLATION_BASELINE + target flags to be tested

C.12 Infrastructure and Build Flags

Infrastructure and build flags control low-level technical behavior and build configuration. These flags typically do not require direct user attention, but are important in specific deployment and debugging scenarios.

Feature Flag	Type	Description	Impact Scope	Dependencies	Related Chapter
UNATTENDED_RETRY	Compile-time	Unattended retry. Automatically retries on API call failures without user intervention	API retry logic, backoff strategy, maximum retry count	No prerequisites	Chapter 13
IS_LIBC_GLIBC	Compile-time	Detects whether the target platform’s libc is the glibc implementation, used for binary compatibility determination	Compatibility selection during binary distribution	No prerequisites	Appendix
IS_LIBC_MUSL	Compile-time	Detects whether the target platform’s libc is the musl implementation (e.g., Alpine Linux), used for binary compatibility determination	Compatibility selection during binary distribution	No prerequisites	Appendix
POWERSHELL_AUTO_MODE	Compile-time	PowerShell auto mode. Provides dedicated automated permission configuration for Windows PowerShell environments	PowerShell environment permission auto-configuration	Windows environments only	Appendix
ALLOW_TEST_VERSIONS	Compile-time	Allow test versions. Accepts pre-release/test version numbers during version checks	Version check logic, pre-release version acceptance	No prerequisites	Appendix
SKIP_DETECTION_WHEN_AUTOUPDATES_DISABLED	Compile-time	Skip detection when auto-updates are disabled. When auto-updates have been explicitly disabled, skips version detection logic to reduce startup latency	Version detection skip, startup optimization	No prerequisites	Appendix
DUMP_SYSTEM_PROMPT	Compile-time	Dump system prompt. When enabled, outputs the complete system prompt to a log or file for debugging purposes	System prompt export, debug logging	No prerequisites	Appendix
OVERFLOW_TEST_TOOL	Compile-time	Overflow test tool. Provides a dedicated test tool for verifying context overflow handling logic	OverflowTestTool, overflow scenario simulation	Testing environments only	Appendix
ULTRATHINK	Compile-time	Deep thinking mode. Enables Extended Thinking capability	Extended Thinking API calls, thinking token processing	No prerequisites	Chapter 5
TORCH	Compile-time	Torch mode. Experimental enhanced reasoning capability	Enhanced reasoning engine, experimental reasoning strategies	No prerequisites	Chapter 5

Debugging Tip: When troubleshooting abnormal system behavior, DUMP_SYSTEM_PROMPT is one of the most effective diagnostic tools available. It can export the complete system prompt sent to the model, making it easy to inspect whether instructions are assembled correctly.

---

C.13 Flag Usage Statistics

Category	Count
Core Interaction Modes	12
Agent and Subtasks	9
Context Management and Compression	8
Permission and Security	5
Tools and Skills	14
Session and Persistence	4
Memory and Knowledge Management	4
Remote and Connectivity	8
UI and Interface	7
Settings Synchronization	2
Telemetry and Diagnostics	6
Infrastructure and Build	10
Total	89

Note: The flag list above is compiled from system architecture analysis and may change with version iterations. The specific activation method for each feature flag is determined by the build configuration, and some flags also require runtime gating conditions (such as specific activation detection functions) to truly activate the corresponding features.

---

C.14 Common Configuration Scenario Recommendations

The following lists several typical usage scenarios and their recommended feature flag combinations to help readers configure based on their actual needs:

flowchart TD
    Start{"Choose Your Scenario"}
    Start -->|Daily Development| S1["Scenario 1: Standard Mode<br/>BASH_CLASSIFIER<br/>TRANSCRIPT_CLASSIFIER<br/>REACTIVE_COMPACT<br/>TOKEN_BUDGET"]
    Start -->|IDE Collaboration| S2["Scenario 2: IDE Integration<br/>BRIDGE_MODE<br/>KAIROS + KAIROS_CHANNELS<br/>BASH_CLASSIFIER<br/>REACTIVE_COMPACT"]
    Start -->|Multi-Agent| S3["Scenario 3: Multi-Agent Collaboration<br/>COORDINATOR_MODE<br/>FORK_SUBAGENT<br/>BUILTIN_EXPLORE_PLAN_AGENTS<br/>VERIFICATION_AGENT"]
    Start -->|CI/CD| S4["Scenario 4: Automated CI/CD<br/>DAEMON<br/>AGENT_TRIGGERS<br/>UNATTENDED_RETRY<br/>HARD_FAIL"]
    Start -->|Enterprise Security| S5["Scenario 5: Security Audit<br/>HARD_FAIL<br/>TRANSCRIPT_CLASSIFIER<br/>BASH_CLASSIFIER<br/>NATIVE_CLIENT_ATTESTATION"]
    Start -->|Performance Tuning| S6["Scenario 6: Performance Debugging<br/>PERFETTO_TRACING<br/>SHOT_STATS<br/>SLOW_OPERATION_LOGGING<br/>DUMP_SYSTEM_PROMPT"]

    classDef decision fill:#ffeb3b,stroke:#f9a825,color:#333
    classDef scene fill:#4a90d9,stroke:#2c5f8a,color:#fff

    class Start decision
    class S1,S2,S3,S4,S5,S6 scene

Scenario 1: Daily Development (Standard Mode)

Suitable for everyday use by most developers, providing a balanced set of features and security.

Core Flags: BASH_CLASSIFIER + TRANSCRIPT_CLASSIFIER + REACTIVE_COMPACT + TOKEN_BUDGET

Scenario 2: IDE Integration Development

Suitable for scenarios where Claude Code is used via plugins in VS Code or JetBrains.

Core Flags: BRIDGE_MODE + KAIROS + KAIROS_CHANNELS + BASH_CLASSIFIER + REACTIVE_COMPACT

Scenario 3: Multi-Agent Collaboration

Suitable for complex scenarios requiring task distribution and coordination among multiple agents.

Core Flags: COORDINATOR_MODE + FORK_SUBAGENT + BUILTIN_EXPLORE_PLAN_AGENTS + VERIFICATION_AGENT + AGENT_MEMORY_SNAPSHOT

Scenario 4: Automated CI/CD Integration

Suitable for unattended execution in CI/CD pipelines.

Core Flags: DAEMON + AGENT_TRIGGERS + UNATTENDED_RETRY + HARD_FAIL + WORKFLOW_SCRIPTS

Scenario 5: Security Audit Mode

Suitable for enterprise environments with stringent security requirements.

Core Flags: HARD_FAIL + TRANSCRIPT_CLASSIFIER + BASH_CLASSIFIER + NATIVE_CLIENT_ATTESTATION + ANTI_DISTILLATION_CC + ULTRAPLAN

Scenario 6: Performance Debugging and Optimization

Suitable for troubleshooting performance issues or optimizing system behavior.

Core Flags: PERFETTO_TRACING + SHOT_STATS + SLOW_OPERATION_LOGGING + DUMP_SYSTEM_PROMPT + PROMPT_CACHE_BREAK_DETECTION

---

C.15 Global Flag Dependency Relationship Diagram

The following diagram illustrates the primary dependency relationships and grouping structure among feature flags:

flowchart TD
    subgraph InteractionModeLayer["Interaction Mode Layer"]
        PROACTIVE
        KAIROS --> KAIROS_BRIEF
        KAIROS --> KAIROS_CHANNELS
        KAIROS --> KAIROS_DREAM
        KAIROS --> KAIROS_GITHUB_WEBHOOKS
        KAIROS --> KAIROS_PUSH_NOTIFICATION
        BRIDGE_MODE
        COORDINATOR_MODE
        VOICE_MODE
        DAEMON
        BUDDY
    end

    subgraph AgentLayer["Agent Layer"]
        FORK_SUBAGENT --> BUILTIN_EXPLORE_PLAN_AGENTS
        AGENT_TRIGGERS --> AGENT_TRIGGERS_REMOTE
        ULTRAPLAN
        VERIFICATION_AGENT
        AGENT_MEMORY_SNAPSHOT
        WORKFLOW_SCRIPTS
        TEMPLATES
    end

    subgraph ContextManagementLayer["Context Management Layer"]
        REACTIVE_COMPACT
        CONTEXT_COLLAPSE
        CACHED_MICROCOMPACT
        HISTORY_SNIP
        COMPACTION_REMINDERS
        PROMPT_CACHE_BREAK_DETECTION
        TOKEN_BUDGET
        COMMIT_ATTRIBUTION
    end

    subgraph SecurityLayer["Security Layer"]
        TRANSCRIPT_CLASSIFIER
        BASH_CLASSIFIER
        HARD_FAIL
        NATIVE_CLIENT_ATTESTATION
        ANTI_DISTILLATION_CC
    end

    subgraph ToolAndSkillLayer["Tool and Skill Layer"]
        TREE_SITTER_BASH --> TREE_SITTER_BASH_SHADOW
        MONITOR_TOOL
        WEB_BROWSER_TOOL
        MCP_SKILLS
        MCP_RICH_OUTPUT
        SkillSystem["Skill System"] --> EXPERIMENTAL_SKILL_SEARCH
        SkillSystem --> SKILL_IMPROVEMENT
        SkillSystem --> RUN_SKILL_GENERATOR
        SkillSystem --> BUILDING_CLAUDE_APPS
        SkillSystem --> REVIEW_ARTIFACT
        HOOK_PROMPTS
        CONNECTOR_TEXT
        UDS_INBOX
    end

    subgraph RemoteAndConnectivityLayer["Remote and Connectivity Layer"]
        SSH_REMOTE
        DIRECT_CONNECT
        CCR_AUTO_CONNECT --> CCR_MIRROR
        CCR_REMOTE_SETUP
        SELF_HOSTED_RUNNER
        BYOC_ENVIRONMENT_RUNNER
    end

    subgraph InfrastructureLayer["Infrastructure Layer"]
        UNATTENDED_RETRY
        ULTRATHINK
        TORCH
        IS_LIBC_GLIBC
        IS_LIBC_MUSL
        POWERSHELL_AUTO_MODE
        DUMP_SYSTEM_PROMPT
    end

    classDef layer1 fill:#4a90d9,stroke:#2c5f8a,color:#fff
    classDef layer2 fill:#8fbc8f,stroke:#5a8a5a,color:#fff
    classDef layer3 fill:#ff9800,stroke:#e65100,color:#fff
    classDef layer4 fill:#ef5350,stroke:#c62828,color:#fff
    classDef layer5 fill:#ce93d8,stroke:#7b1fa2,color:#fff
    classDef layer6 fill:#7eb8da,stroke:#4a90d9,color:#fff
    classDef layer7 fill:#a0a0a0,stroke:#666,color:#fff

    class PROACTIVE,KAIROS,KAIROS_BRIEF,KAIROS_CHANNELS,KAIROS_DREAM,KAIROS_GITHUB_WEBHOOKS,KAIROS_PUSH_NOTIFICATION,BRIDGE_MODE,COORDINATOR_MODE,VOICE_MODE,DAEMON,BUDDY layer1
    class FORK_SUBAGENT,BUILTIN_EXPLORE_PLAN_AGENTS,AGENT_TRIGGERS,AGENT_TRIGGERS_REMOTE,ULTRAPLAN,VERIFICATION_AGENT,AGENT_MEMORY_SNAPSHOT,WORKFLOW_SCRIPTS,TEMPLATES layer2
    class REACTIVE_COMPACT,CONTEXT_COLLAPSE,CACHED_MICROCOMPACT,HISTORY_SNIP,COMPACTION_REMINDERS,PROMPT_CACHE_BREAK_DETECTION,TOKEN_BUDGET,COMMIT_ATTRIBUTION layer3
    class TRANSCRIPT_CLASSIFIER,BASH_CLASSIFIER,HARD_FAIL,NATIVE_CLIENT_ATTESTATION,ANTI_DISTILLATION_CC layer4
    class TREE_SITTER_BASH,TREE_SITTER_BASH_SHADOW,MONITOR_TOOL,WEB_BROWSER_TOOL,MCP_SKILLS,MCP_RICH_OUTPUT,EXPERIMENTAL_SKILL_SEARCH,SKILL_IMPROVEMENT,RUN_SKILL_GENERATOR,BUILDING_CLAUDE_APPS,REVIEW_ARTIFACT,HOOK_PROMPTS,CONNECTOR_TEXT,UDS_INBOX layer5
    class SSH_REMOTE,DIRECT_CONNECT,CCR_AUTO_CONNECT,CCR_MIRROR,CCR_REMOTE_SETUP,SELF_HOSTED_RUNNER,BYOC_ENVIRONMENT_RUNNER layer6
    class UNATTENDED_RETRY,ULTRATHINK,TORCH,IS_LIBC_GLIBC,IS_LIBC_MUSL,POWERSHELL_AUTO_MODE,DUMP_SYSTEM_PROMPT layer7