The Complete List: All Build Flags & Runtime Gates

Disclaimer

All analysis is based on a specific version of the Claude Code source at instructkr/claude-code. Claude Code is not open-source — all intellectual property belongs to Anthropic. This is independent source code analysis only.

This is the comprehensive reference for every gating mechanism we identified in the Claude Code source code. These fall into four categories: build-time feature flags, GrowthBook runtime flags, environment variables, and API beta headers.

Caution

This list was compiled through static analysis of the source code. Flag names, behaviors, and availability may change between versions. Some flags may be deprecated or not yet active.

Build-Time Feature Flags

These flags are evaluated at build time via feature() calls that reference definitions from bun:bundle. If a flag is not enabled in the build configuration, the associated code is tree-shaken out of the final bundle entirely — it does not exist at runtime.

Flag Name	Category	Description	Likely Ant-Only?
AGENT_MEMORY_SNAPSHOT	Memory	Agent memory persistence snapshots	Yes
AGENT_TRIGGERS	Scheduling	Scheduled agent execution	Yes
AGENT_TRIGGERS_REMOTE	Scheduling	Remote agent scheduling	Yes
ANTI_DISTILLATION_CC	Security	Anti-distillation content classifier	Yes
AUTO_THEME	UI	Automatic theme detection	No
AWAY_SUMMARY	UX	Away mode summarization	Yes
BG_SESSIONS	Sessions	Background session support	Yes
BRIDGE_MODE	Infrastructure	Remote bridge protocol	Yes
BUDDY / BUDDIES	UX	AI coaching companion	Yes
BUILDING_CLAUDE_APPS	Skills	Claude API skill	No
BUILTIN_EXPLORE_PLAN_AGENTS	Agents	Built-in agent templates	Yes
CCR_MIRROR	Infrastructure	Cloud compute resource mirroring	Yes
CHAT_MODES	UI	Chat mode selection UI	Yes
CONNECTOR_TEXT	Processing	Connector text summarization	Yes
CONTEXT_COLLAPSE	Performance	Context window optimization	Yes
COORDINATOR_MODE	Agents	Multi-worker orchestration	Yes
DAEMON	Infrastructure	Background daemon mode	Yes
DIRECT_CONNECT	Infrastructure	Direct WebSocket connections	Yes
ENHANCED_TELEMETRY_BETA	Analytics	Extended analytics	Yes
FORK_SUBAGENT	Agents	Cache-optimized subagent forking	Yes
HISTORY_PICKER / HISTORY_SNIP	Sessions	Session history features	Yes
KAIROS	Background Tasks	Async/background task system	Yes
KAIROS_BRIEF	Background Tasks	Brief formatting	Yes
KAIROS_CHANNELS	Background Tasks	MCP channel notifications	Yes
KAIROS_DREAM	Memory	Auto-memory consolidation	Yes
KAIROS_GITHUB_WEBHOOKS	Integrations	GitHub webhook integration	Yes
KAIROS_PUSH_NOTIFICATION	Notifications	Push notifications	Yes
MCP_RICH_OUTPUT	MCP	Rich MCP tool output	Yes
MCP_SKILLS	MCP	MCP tool skills integration	Yes
MESSAGE_ACTIONS	UI	Message action buttons	Yes
NATIVE_CLIENT_ATTESTATION	Security	Client attestation verification	Yes
NATIVE_CLIPBOARD_IMAGE	UI	Native clipboard image support	No
PROACTIVE	Agents	Autonomous execution	Yes
PROMPT_CACHE_BREAK_DETECTION	Performance	Detect prompt cache breaks	Yes
REACTIVE_COMPACT	Performance	Reactive context compaction	Yes
REVIEW_ARTIFACT	Review	Artifact review	No
RUN_SKILL_GENERATOR	Skills	Skill generation	No
SELF_HOSTED_RUNNER	Infrastructure	Self-hosted runner	Yes
SSH_REMOTE	Infrastructure	SSH remote execution	Yes
TRANSCRIPT_CLASSIFIER	Permissions	Auto-mode permission classifier	Yes
ULTRAPLAN	Planning	Ultra-detailed planning	Yes
ULTRATHINK	Thinking	Extended thinking	Yes
UPLOAD_USER_SETTINGS / DOWNLOAD_USER_SETTINGS	Settings	Cloud settings sync	Yes
VERIFICATION_AGENT	Agents	Code verification agent	Yes
VOICE_MODE	Input	Voice input/output	Yes
WEB_BROWSER_TOOL	Tools	Browser automation	Yes

Tip

The majority of flags are Ant-only. The few that ship in the public build (AUTO_THEME, BUILDING_CLAUDE_APPS, NATIVE_CLIPBOARD_IMAGE, REVIEW_ARTIFACT, RUN_SKILL_GENERATOR) tend to be UI polish or non-sensitive skill features.

Key GrowthBook Runtime Flags

GrowthBook flags (prefixed with tengu_) are evaluated at runtime by querying Anthropic’s feature-flagging service. Even if the code path exists in your build, these flags can enable or disable features server-side based on your user account, organization, or A/B test cohort.

Flag Name	Purpose
tengu_amber_flint	Agent swarms killswitch
tengu_amber_quartz_disabled	Voice mode killswitch
tengu_amber_stoat	Explore/Plan agents (default true)
tengu_ant_model_override	Model override for Anthropic employees
tengu_auto_mode_config	Auto-mode configuration (enabled / opt-in / disabled)
tengu_ccr_bridge_multi_session	Multi-session bridge
tengu_hive_evidence	Verification agent
tengu_harbor	MCP channels/notifications
tengu_kairos_cron	Scheduled agent execution
tengu_kairos_cron_durable	Durable scheduled task state
tengu_malort_pedway	Computer Use MCP configuration
tengu_max_version_config	Version killswitch
tengu_onyx_plover	Auto-dream configuration
tengu_otk_slot_v1	One-time key slots
tengu_remote_backend	Remote TUI backend
tengu_scratch	Scratchpad durable storage
tengu_sedge_lantern	Away summaries
tengu_session_memory	Session-to-session memory
tengu_sessions_elevated_auth_enforcement	Trusted device requirement
tengu_slim_subagent_claudemd	Skip CLAUDE.md for read-only agents
tengu_surreal_dali	Remote trigger tool
tengu_tool_pear	Strict tool use
tengu_transcript_classifier	Auto-mode classifier
tengu_willow_mode	Context optimization (off / light / full)

Note

The tengu_ prefix appears to be a codename convention. “Tengu” is a creature from Japanese folklore — a fitting mascot for a tool that watches and intervenes in your coding sessions.

Key Environment Variables

These environment variables can be set in your shell before launching Claude Code. Some are build-time gates, others are runtime toggles.

Variable	Purpose
USER_TYPE=ant	Build-time Anthropic employee gate
CLAUDE_CODE_COORDINATOR_MODE=1	Enable coordinator mode
CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1	Enable agent teams
CLAUDE_CODE_PROACTIVE=1	Enable proactive mode
CLAUDE_CODE_UNDERCOVER=1	Force undercover mode
CLAUDE_CODE_TEAMMATE_COMMAND=/path	Custom teammate binary
CLAUDE_CODE_SIMPLE=1	Simple mode for coordinator
CLAUDE_CODE_PLAN_MODE_REQUIRED=1	Require plan mode
CLAUDE_INTERNAL_FC_OVERRIDES='{}'	Feature flag overrides (JSON)
CLAUDE_CODE_GB_BASE_URL=...	GrowthBook override
CLAUDE_CODE_TERMINAL_RECORDING=1	Terminal recording
MONOREPO_ROOT_DIR	Anthropic monorepo detection
ALLOW_ANT_COMPUTER_USE_MCP	Override computer use gate

Caution

Setting USER_TYPE=ant as an environment variable at runtime will not change the build-time feature flags. The USER_TYPE check happens during the build/bundle step, not at runtime. You would need to rebuild from source with this value set.

Ant-Only API Beta Headers

These HTTP headers are sent to the Anthropic API to unlock server-side beta features. They are only included when running the internal (Ant) build.

Header Constant	Purpose
CLI_INTERNAL_BETA_HEADER	Internal CLI beta access — unlocks Ant-only API capabilities
SUMMARIZE_CONNECTOR_TEXT_BETA_HEADER	Anti-distillation connector text processing
AFK_MODE_BETA_HEADER	Away-from-keyboard mode — signals the user is not actively watching

These headers are particularly important because even if you enable the client-side code paths, the API server will reject requests that include these headers unless your account is authorized. This is a server-side gate that cannot be bypassed by modifying the client.