Feature gates reference
This page owns the feature-gate taxonomy visible in cli.js: GrowthBook evaluation, tengu_* gates/signals, environment kill switches, managed-policy gates, and CLI/settings gates that switch operational behavior.
Use Telemetry and tracing for telemetry/export behavior and Environment variables reference for canonical env-var names.
Source anchors
| Semantic alias | Source | Approximate location | String or symbol | Meaning |
|---|---|---|---|---|
| GrowthBookClient | cli.js | line ~129, byte 0xe138c | GrowthBook | Embedded feature/experiment client. |
| FeatureFlagLookupWrapper | cli.js | line ~253, byte 0x20c503 | Z$("..." | Feature-flag lookup wrapper pattern. |
| DisableGrowthBookEnv | cli.js | line ~432, byte 0x2fe03e | DISABLE_GROWTHBOOK | Environment kill switch for GrowthBook. |
| DisableCronEnv | cli.js | line ~1091, byte 0x4806fd | CLAUDE_CODE_DISABLE_CRON | Cron/scheduled-task kill switch. |
| KairosCronGate | cli.js | line ~1091, byte 0x480583 | isKairosCronEnabled | Scheduled-task feature gate wrapper. |
| CronSchedulerRuntime | cli.js | line ~9564, byte 0xcd3ba4 | createCronScheduler | Scheduled-task runtime. |
| DisableRemoteControlPolicy | cli.js | line ~185, byte 0x11a7a3 | disableRemoteControl | Managed policy gate for Remote Control. |
| DisableSkillShellPolicy | cli.js | line ~185, byte 0x11a8f2 | disableSkillShellExecution | Managed policy gate for skill/slash shell execution. |
Gate taxonomy
flowchart TD Decision[Runtime decision] --> GB[GrowthBook / tengu feature flags] Decision --> Env[Environment kill switches] Decision --> Policy[Managed settings / policy] Decision --> Cli[CLI flags] Decision --> Settings[User/project/local settings]
GB --> Behavior[Feature behavior] Env --> Behavior Policy --> Behavior Cli --> Behavior Settings --> Behavior Behavior --> Evidence[debug / telemetry evidence]| Gate type | Examples | What it controls |
|---|---|---|
| GrowthBook/feature flag | GrowthBook, Z$("..."), many tengu_* keys | Experiments, staged rollout, runtime feature defaults. |
| Environment kill switch | DISABLE_GROWTHBOOK, CLAUDE_CODE_DISABLE_CRON, telemetry/update/context UI env vars | Coarse process-level enable/disable decisions. |
| Managed policy/settings | disableRemoteControl, disableAllHooks, disableAgentView, disableSkillShellExecution, cleanupPeriodDays | Organization/admin and persisted configuration controls. |
| CLI flag | --debug, --debug-file, --add-trace-attribute, --no-session-persistence, --agents | Per-invocation behavior. |
| Provider/env selector | CLAUDE_CODE_USE_BEDROCK, CLAUDE_CODE_USE_VERTEX, CLAUDE_CODE_USE_FOUNDRY, CLAUDE_CODE_USE_ANTHROPIC_AWS, CLAUDE_CODE_USE_MANTLE | Provider/auth/reporting paths. |
High-signal feature-gate families
The table groups source-visible keys by nearby behavior. It does not claim every opaque tengu_* name is fully decoded.
| Family | Source-visible gates | Interpreted scope |
|---|---|---|
| Remote / bridge / CCR | tengu_ccr_bridge, tengu_remote_backend, tengu_bridge_repl_v2_cse_shim_enabled, tengu_bridge_attestation_enforce, tengu_bridge_attestation_enforce_config, tengu_ccr_v2_send_events_cli, tengu_bridge_requires_action_details, tengu_bridge_system_init, CLAUDE_CODE_REMOTE, disableRemoteControl | Remote Control, bridge transport, attestation, event forwarding. |
| Scheduled tasks / Kairos cron | tengu_kairos_cron, tengu_kairos_cron_durable, tengu_kairos_loop_dynamic, tengu_kairos_loop_persistent, tengu_kairos_loop_prompt, tengu_kairos_push_notifications, tengu_kairos_input_needed_push, CLAUDE_CODE_DISABLE_CRON | Cron/scheduled prompt feature family and autonomous-loop behavior. |
| Agents / background agents | tengu_auto_background_agents, tengu_agent_list_attach, tengu_mcp_subagent_prompt, tengu_slim_subagent_claudemd, CLAUDE_AGENT_SDK_DISABLE_BUILTIN_AGENTS, CLAUDE_CODE_DISABLE_AGENT_VIEW | Agent UI, background agents, subagent prompt/context behavior. |
| Auto-mode / permission automation | tengu_auto_mode_config, tengu_disable_bypass_permissions_mode, tengu_auto_notice_once, policy defaultMode=auto | Auto-mode defaults, consent, permission automation. |
| Tool/runtime security | tengu_streaming_tool_execution2, tengu_harbor_permissions, tengu_destructive_command_warning, tengu_bash_allowlist_strip_all, CLAUDE_CODE_DISABLE_ADVISOR_TOOL, CLAUDE_CODE_ENABLE_EXPERIMENTAL_ADVISOR_TOOL | Tool execution mode, warnings, advisory/permission behavior. |
| MCP/plugins/skills | tengu_mcp_directory_visibility, tengu_mcp_directory_bff, tengu_mcp_singleton_unwrap, tengu_skills_dashboard_enabled, tengu_plugin_official_mkt_git_fallback, CLAUDE_CODE_PLUGIN_PREFER_HTTPS, CLAUDE_CODE_PLUGIN_USE_ZIP_CACHE, CLAUDE_CODE_SYNC_PLUGIN_INSTALL | MCP discovery, plugin install/update, skills UI. |
| Context/model behavior | CLAUDE_CODE_DISABLE_1M_CONTEXT, DISABLE_COMPACT, DISABLE_INTERLEAVED_THINKING, USE_API_CONTEXT_MANAGEMENT, CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS, tengu_prompt_cache_1h_config, tengu_prompt_cache_diagnostics | Context window, compaction, thinking, prompt cache, beta flags. |
| Web/search/fetch adjacent | tengu_tool_search_unsupported_models, /api/web/domain_info?domain= | Web/domain metadata and model support guardrails. |
| UI/terminal behavior | CLAUDE_CODE_DISABLE_ALTERNATE_SCREEN, CLAUDE_CODE_NO_FLICKER, CLAUDE_CODE_DISABLE_MOUSE, CLAUDE_CODE_FORCE_SYNC_OUTPUT, CLAUDE_CODE_ACCESSIBILITY, CLAUDE_CODE_DISABLE_TERMINAL_TITLE | Terminal rendering and accessibility behavior. |
| Updates | DISABLE_UPDATES, DISABLE_AUTOUPDATER, FORCE_AUTOUPDATE_PLUGINS, auto-update settings | Native/plugin updater behavior. |
| Telemetry/diagnostics | CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC, DISABLE_TELEMETRY, DO_NOT_TRACK, DISABLE_ERROR_REPORTING, DISABLE_GROWTHBOOK, OTEL_LOG_TOOL_DETAILS, OTEL_LOG_TOOL_CONTENT, OTEL_LOG_USER_PROMPTS | Observability, analytics, error reporting, feature-evaluation fetching. |
Relationship to sessions and agents
flowchart TD Gates[Feature/env/policy gates] --> Session[Session envelope] Gates --> Agents[Agent/task scheduler] Gates --> Tools[Tool/security runtime] Gates --> Telemetry[Telemetry and tracing] Session --> Evidence[debug/telemetry evidence] Agents --> Evidence Tools --> Evidence| Runtime area | Gate interaction |
|---|---|
| Sessions/remote | Remote Control can be disabled by managed policy; bridge state and transcript mirrors emit observable frames. |
| Tools/permissions | Permission automation and streaming execution are feature-gated; denials can surface as stream frames and telemetry. |
| Agents/tasks | Background agents, subagent prompt shape, scheduled tasks, and auto-mode are behind feature/env/policy gates. |
| Scheduled tasks | CLAUDE_CODE_DISABLE_CRON disables cron; scheduled-task telemetry tracks missed/fire/expired behavior. |
| Hosted review/update | Hosted review preflight and native updater emit operational telemetry but do not own model turns. |
Safe interpretation rules
- Treat
tengu_*as evidence of a gate or telemetry event, not a fully decoded feature name unless adjacent strings confirm behavior. - Prefer env/policy strings for hard claims, because they are usually descriptive and user/admin-facing.
- Distinguish a gate from an exported event: feature gates decide behavior; telemetry/tracing reports behavior.
- Use Telemetry and tracing for sink/export details.
Related docs
Created and maintained by Yingting Huang.