Follow Discord
Sweep 22 Oct 2026 · 21:04Z Build v2.1.309 528 read Stable v2.1.308 Latest v2.1.308 Next v2.1.309 Feeds RSS JSON llms.txt llms-full.txt Unofficial
One family

Feature flags, all 2,095.

GrowthBook gates the client evaluates on startup. Most of what ships is behind one of these before it is announced.

Feature flags2,095seen across 150 builds
In v2.1.309900the newest inventory on disk
Gone316in an earlier build, not in the newest
Described0have a line in Anthropic's documentation
No docs line2,095no line in the documentation we hold

Every feature flag, and where it lived

937 of 2,095 feature flags One tick per build on this box, oldest on the left. A gap is a build the feature flag was not in.

A is the one account this site reads, N the anonymous baseline, D what the release compiled in. 1,158 of 2,095 gates have been read at all: a run only records the gates its own release wrote about. Claude Code evaluates its gates on Anthropic's servers, so a reading of off can be a rollout this account sits outside of, and none of it says anything about your install. How to read these.

○ not in v2.1.309, ◇ we found no documentation describing it.

Feature flagWhat it doesSpan v2.1.138 → v2.1.309 Status Reading
tengu_permission_explainer_shortcut_used ◇ An analytics event, not a feature flag. Claude Code logs it with no fields when the user presses the toggle-explanation shortcut (confirm:toggleExplanation, default ctrl+e) in a permission prompt while the explanation is hidden; if no explanation has been requested yet, this also starts generating one.never mined ·never mined never read tengu_permission_request_option_selected ◇ The Bash permission prompt only records your answer when it actually accepts the keypress.never mined ·never mined never read tengu_persistence_suppressed ◇ The "persistence suppressed" warning now shows once per host instead of once per process.never mined ·never mined never read tengu_pewter_canteen ◇ Default false. When true, publishing an artifact page that contains charts does not inject the chart runtime block into the page.v2.1.206 → v2.1.309 ◆in the build never read tengu_pewter_owl_model ◇ No description from anybody.v2.1.159 → v2.1.309 ◆in the build never read tengu_phantom_parent_write ◇ An analytics event, not a feature flag. Claude Code logs it, with no fields, while writing messages to the session transcript when a user message carries a sourceToolAssistantUUID that is not among the session's stored messages; that message is then parented to the previous message in the chain instead.never mined ·never mined never read tengu_pid_based_version_locking ◇ No description from anybody.never mined ·never mined never read tengu_plain_heron ◇ No description from anybody.v2.1.299 → v2.1.309 ◆in the build never read tengu_plan_exit ◇ An analytics event, not a feature flag. Claude Code logs it when the user picks an option in the exit plan mode dialog, with planLengthChars and outcome (the chosen option, such as no, ultraplan or yes-accept-edits), and on several approval paths also clearContext and hasFeedback.never mined ·never mined never read tengu_plan_exit_dialog_shown ◇ The plan-exit dialog was rebuilt on plain hooks and its feedback box now gets paste cleanup.never mined ·never mined never read tengu_plan_mode_interview_phase ◇ Default false. Turns on the iterative plan mode workflow: the full plan mode reminder uses an ## Iterative Planning Workflow section in place of the 5-phase workflow, the sparse reminder says Follow iterative workflow: explore codebase, interview user, write to plan incrementally., the EnterPlanMode tool description drops one section, and its result adds DO NOT write or edit any files except the plan file. CLAUDE_CODE_PLAN_MODE_INTERVIEW_PHASE set to a true value forces it on and a false value forces it off.v2.1.138 or earlier → v2.1.144 ○gone never read tengu_playful_lobster_fired ◇ Read-permission checks now detect hardlinked files and log it as shadow telemetry, without changing the decisionnever mined ·never mined never read tengu_plugin_ ◇ No description from anybody.v2.1.283 → v2.1.299 ○gone never read tengu_plugin_command_failed ◇ Failed plugin commands now flush their analytics before exiting instead of dropping the event.never mined ·never mined never read tengu_plugin_disabled_cli ◇ Plugin install/uninstall telemetry now confirms the plugin against the marketplace catalog before loggingnever mined ·never mined never read tengu_plugin_enabled_cli ◇ Plugin-enable telemetry now looks up plugin metadata asynchronously with added directory contextnever mined ·never mined never read tengu_plugin_enabled_for_session ◇ Skill usage stats include hashed skill names only for plugins from outside marketplaces.never mined ·never mined never read tengu_plugin_hint_seen ◇ Tool and command output can now embed a hidden tag that suggests installing a specific plugin.never mined ·never mined never read tengu_plugin_install_auto_activate ◇ Claude now records whether a freshly installed plugin actually activated, and why it didn't.never mined ·never mined never read tengu_plugin_installed_cli ◇ Plugin install/uninstall telemetry now confirms the plugin against the marketplace catalog before loggingnever mined ·never mined never read tengu_plugin_load_refresh_window_recovered ◇ Plugins from marketplaces skipped at startup now get installed anyway.never mined ·never mined never read tengu_plugin_menu_auto_reload ◇ Plugin menu now tells you when plugin changes are queued to apply after the current response finishesnever mined ·never mined never read tengu_plugin_menu_close_reload ◇ Plugin menu now tells you when plugin changes are queued to apply after the current response finishesv2.1.268 → v2.1.309 ◆in the build never read tengu_plugin_prune_cli ◇ An analytics event, not a feature flag. Claude Code logs it when claude plugin prune actually removes orphaned auto-installed plugins (not on a dry run or an aborted prompt), with the scope and removed_count.never mined ·never mined never read tengu_plugin_ratings_store_config ◇ Thumbs-up and thumbs-down ratings for plugins can be posted to your organization, off by default.never mined ·never mined never read tengu_plugin_remote_fetch ◇ Plugin fetch telemetry now sanitizes host and error values before sending them.never mined ·never mined never read tengu_plugin_renamed ◇ An analytics event, not a feature flag. Claude Code logs it when an enabled plugin id is not found in its marketplace but the marketplace has a renames entry for it, with outcome (renamed, removed or unresolved), chain_depth for a resolved chain, and for unresolved a reason (cycle, target-missing or chain-too-deep).never mined ·never mined never read tengu_plugin_uninstalled_cli ◇ Plugin install/uninstall telemetry now confirms the plugin against the marketplace catalog before loggingnever mined ·never mined never read tengu_plugin_updated_cli ◇ Plugin install/uninstall telemetry now confirms the plugin against the marketplace catalog before loggingnever mined ·never mined never read tengu_plugins_sync_list_failed ◇ Plugin sync failure logs and telemetry now break the failure down by kind and HTTP status.never mined ·never mined never read tengu_plugins_sync_manifest_failed ◇ An analytics event, not a feature flag. Claude Code logs it when fetching the session sync manifest fails while listing plugin entries, with unavailable set when the failure reason is unavailable; the same failure for skills logs tengu_skills_sync_manifest_failed.never mined ·never mined never read tengu_plugins_sync_mcp_relay_wait ◇ Internal plugin sync and MCP relay functions were renamed with no behavior changenever mined ·never mined never read tengu_plugins_sync_wait_timeout ◇ Internal plugin sync and MCP relay functions were renamed with no behavior changenever mined ·never mined never read tengu_plum_heron ◇ Default { off: true }. Config object for a notice about a newer model in the same family as the model saved in your user settings (state keys savedModelSuccessorShown and savedModelSuccessorAnswered). The notice is suppressed when off is true, when the model's family is turned off in families (falling back to a built-in per-family default), after it has been answered, after maxViews showings, or when it was shown more recently than minDaysApart allows. The object also accepts lead, text, action and done strings; a value that fails validation counts as { off: true }.v2.1.286 ○gone never read tengu_policy_limits_cache_state_at_first_prompt ◇ Three new telemetry events added for refusal-fallback prompts and policy-limits checksnever mined ·never mined never read tengu_policy_limits_fetch ◇ Policy limit fetch events now record how you authenticated and why the server said no.never mined ·never mined never read tengu_policy_limits_late_composition ◇ Billing, artifact, MCP and subagent call sites also take credentials explicitly now.never mined ·never mined never read tengu_policy_limits_principal_changed ◇ Local policy-limits cache now tracks which account is logged in and reacts to account switchesnever mined ·never mined never read tengu_policy_limits_verdict_recovered ◇ Claude Code now retries faster and recovers sooner when it can't fetch your policy limitsnever mined ·never mined never read tengu_polished_quill_killswitch ◇ Default false. In a non-interactive session, a typed slash command that matches no command is handed to the model as an unknown_command_fallback attachment carrying the name, the closest match within edit distance 2 and up to 40 model-invocable prompt command names. Setting the flag to true turns that fallback off, and the session prints Unknown command: /name instead.v2.1.273 → v2.1.309 ◆in the build never read tengu_post_compact_file_restore_success ◇ After compaction, Claude stops wasting its re-read budget on memory files it already injects.never mined ·never mined never read tengu_post_compact_survey_event ◇ An analytics event, not a feature flag. Claude Code logs it for the post-compaction feedback survey with event_type appeared, responded or abandoned, the appearance_id, and the response on responded.never mined ·never mined never read tengu_powershell_tool_command_executed ◇ Bash/PowerShell telemetry and sandbox checks now track whether the user typed the command directly.never mined ·never mined never read tengu_powershell_tool_command_failed ◇ The PowerShell tool's internal code was renamed and reorganized with no behavior change foundnever mined ·never mined never read tengu_powerup_lesson_completed ◇ An analytics event, not a feature flag. Claude Code logs it when a power-up lesson is marked done for the first time, after adding that lesson to powerupsUnlocked in its config, with lesson_id, unlocked_count and all_unlocked.never mined ·never mined never read tengu_powerup_lesson_opened ◇ An analytics event, not a feature flag. Claude Code logs it when a lesson is opened in the power-ups lesson screen, with the fields lesson_id, was_already_unlocked (whether that lesson was already in powerupsUnlocked) and unlocked_count.never mined ·never mined never read tengu_pr_prep_suggestion_rendered ◇ Before committing, Claude can be made to state which verify, simplify and code-review skills it ran.never mined ·never mined never read tengu_pr_status_cli ◇ No description from anybody.never mined ·never mined never read tengu_pr_writing_guidance_rendered ◇ Plain-language PR writing rules and repo PR template fetching exist but reach no prompt.never mined ·never mined never read tengu_precompute_compaction_setting_changed ◇ An analytics event, not a feature flag. Claude Code logs it when the Precompute compaction toggle in the settings list is changed, with the field enabled; the toggle (setting precomputeCompactionEnabled, shown as on when unset) appears only while the tengu_sepia_moth flag is on.never mined ·never mined never read tengu_precomputed_compact_arm_gated ◇ Claude Code can reuse a precomputed conversation summary instead of compacting fresh, gated offnever mined ·never mined never read tengu_precomputed_compact_discarded ◇ A new session.compact plugin hook can intercept and rewrite conversation compactionnever mined ·never mined never read tengu_precomputed_compact_rehydrated ◇ Claude Code can reuse a precomputed conversation summary instead of compacting fresh, gated offnever mined ·never mined never read tengu_prefill_boundary_mismatch ◇ No functional change: sandbox credential masking and reply-on-resume code was just renamednever mined ·never mined never read tengu_prefix_break_hashes ◇ New sampled telemetry records detailed hashes around cache-prefix breaks.never mined ·never mined never read tengu_prefix_ledger_notice ◇ Prompt assembly can tell the model which prefix sites changed since an earlier render, but the mode is hardcoded off.never mined ·never mined never read tengu_print_model_override_adopted ◇ SDK and print-mode turn telemetry now breaks down tool calls into MCP, tool-search, and built-in countsnever mined ·never mined never read tengu_projects_thread_state ◇ New internal tracking of time and reply counts since the last human message, used for Projects artifact comment repliesnever mined ·never mined never read tengu_projects_thread_state_stamp ◇ New internal tracking of time and reply counts since the last human message, used for Projects artifact comment repliesv2.1.265 → v2.1.309 ◆in the build never read tengu_prompt_cache_break ◇ Cache diagnostics know when the baseline came from disk and skip misleading difference reports.never mined ·never mined never read tengu_prompt_cache_diagnosis_received ◇ MCP server error lines are escaped before display, plus minor telemetry and version-string cleanups.never mined ·never mined never read tengu_prompt_invisible_strip ◇ Fleet's command bar and reply composer now strip invisible characters and block sending if any are foundnever mined ·never mined never read tengu_prompt_shell_permission ◇ Inline shell commands in a skill's instructions can now be handed to the model to run explicitly instead of executing silentlynever mined ·never mined never read tengu_prompt_suggestion_accepted ◇ No description from anybody.never mined ·never mined never read tengu_prompt_suggestion_shown ◇ No description from anybody.never mined ·never mined never read tengu_proud_pearl ◇ Two new feature switch names were added to an existing list of switch namesv2.1.307 → v2.1.309 ◆in the build never read tengu_push_notif_upsell_notification_shown ◇ The three remote-control suggestions were rewritten as plain tasks with identical wording.never mined ·never mined never read tengu_pyrite_wren ◇ No description from anybody.never mined ·never mined never read tengu_quartz_falcon ◇ No description from anybody.never mined ·never mined never read tengu_quartz_heron ◇ Default false. Picks the model for the built-in Explore agent: when false it uses haiku; when true it uses inherit (the parent's model), except on the first-party API when the parent model name contains none of haiku, sonnet or opus, where it uses opus.v2.1.152 → v2.1.197 ○gone never read tengu_quick_heron ◇ No description from anybody.v2.1.285 → v2.1.309 ◆in the build never read tengu_quiet_hollow ◇ No description from anybody.never mined ·never mined never read tengu_quota_auto_resume_armed ◇ Your session can now wait out a usage limit and pick the task back up by itself.never mined ·never mined never read tengu_quota_auto_resume_cancelled ◇ Rate-limit rejections now feed a controller that can arm or cancel an automatic resume after a quota block.never mined ·never mined never read tengu_quota_auto_resume_setting_changed ◇ Your session can now wait out a usage limit and pick the task back up by itself.never mined ·never mined never read tengu_ranch_rc_link_shown ◇ Two new telemetry events fire when Claude Code shows a verification-required message and its linknever mined ·never mined never read tengu_ranch_rc_shown ◇ Two new telemetry events fire when Claude Code shows a verification-required message and its linknever mined ·never mined never read tengu_rate_limit_options_menu_cancel_auto_resume ◇ Your session can now wait out a usage limit and pick the task back up by itself.never mined ·never mined never read tengu_rate_limit_options_menu_select_auto_resume ◇ Your session can now wait out a usage limit and pick the task back up by itself.never mined ·never mined never read tengu_rate_limit_options_menu_select_cedar_ember ◇ A second rate-limit option, 'cedar-ember', joins the existing 'juniper-tide' offer for using a reset nownever mined ·never mined never read tengu_rate_limit_options_menu_select_juniper_tide ◇ You could clear a five-hour session limit once a week, but the command stays hidden.never mined ·never mined never read tengu_rc_autostart_resolved ◇ One resolver now decides whether Remote Control starts and in which mode, weighing flags, settings and env vars.never mined ·never mined never read tengu_rc_long_turn_nudge_shown ◇ The three remote-control suggestions were rewritten as plain tasks with identical wording.never mined ·never mined never read tengu_rc_permission_nudge_shown ◇ The three remote-control suggestions were rewritten as plain tasks with identical wording.never mined ·never mined never read tengu_rc_pill_clicked ◇ The Remote Control status pill is now clickable and opens the session in your browser.never mined ·never mined never read tengu_rc_thread_reply_gate ◇ Default true. Gates the projects thread reply check for Remote Control: in an SDK main-thread session with the projects MCP server connected, a turn that ends without a successful terminal reply tool call (reply, no_reply_needed, post_message, react, update_message, update_status) gets a meta message telling the model that nothing reached the project thread and to call the reply tool. The flag also gates a wait for the projects mount, logged as tengu_bridge_projects_mount_wait.v2.1.273 → v2.1.309 ◆in the build never read tengu_rc_upsell_notification_shown ◇ The three remote-control suggestions were rewritten as plain tasks with identical wording.never mined ·never mined never read tengu_reactive_compact_attempt ◇ Automatic conversation compaction now flags when foreign or teammate content is being summarizednever mined ·never mined never read tengu_reactive_compact_failed ◇ Failed reactive compaction now reports a specific reason code, not just free textnever mined ·never mined never read tengu_reactive_compact_succeeded ◇ Compaction telemetry now records the split kind and head truncations, and flags when the whole conversation was summarized.never mined ·never mined never read tengu_reduce_motion_setting_changed ◇ An analytics event, not a feature flag. Claude Code logs it when the Reduce motion toggle in the settings list is changed, with the field enabled; the change also writes the prefersReducedMotion setting.never mined ·never mined never read tengu_refusal_fallback_preference_prompt ◇ Switching to a fallback model after a refusal can now ask you first, and remembers your answernever mined ·never mined never read tengu_refusal_fallback_resume_latch ◇ Safeguard and refusal notices now end with a Details line naming the refusal category.never mined ·never mined never read tengu_refusal_fallback_route_declined ◇ A new bypass lets Claude Code re-offer a fallback model for bio-related refusals even after the user already declined that routenever mined ·never mined never read tengu_refusal_fallback_setting_changed ◇ The setting that switches models on a flagged refusal now records when you change it.never mined ·never mined never read tengu_refusal_fallback_suppressed ◇ A new bypass lets Claude Code re-offer a fallback model for bio-related refusals even after the user already declined that routenever mined ·never mined never read tengu_refusal_fallback_triggered ◇ If compaction's model is refused, it retries on an allowed fallback instead of failing outright.never mined ·never mined never read tengu_refusal_same_model_retry ◇ Claude Code now recovers gracefully when a safety filter cuts off a response mid-tool-callnever mined ·never mined never read tengu_reinit_pending_redelivery ◇ SDK hook swaps no longer leave tool calls hanging; stuck ones now ask you to retry.never mined ·never mined never read tengu_relay_chain_v1 ◇ Default false. When true, the Bash tool prompt drops its "When issuing multiple commands:" section (run independent commands as parallel tool calls, chain dependent ones with &&, use ; only when failures don't matter, no newlines between commands).v2.1.138 or earlier → v2.1.197 ○gone never read
Feedback