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_adopt_claim ◇ Stale adopted session claims now return their data marked stale instead of being discardednever mined ·never mined never read tengu_adopt_link ◇ Adopting an agent's transcript now hard-links the file, copying only when linking is impossible.never mined ·never mined never read tengu_adopt_relink ◇ On startup, subagent transcripts stored as symlinks are converted to hard links or copies.never mined ·never mined never read tengu_advisor_unknown_content ◇ Advisor output you couldn't see now shows an error or a clear "declined to advise" message.never mined ·never mined never read tengu_agent_frontmatter_mode_widening_carry_ignored ◇ An agent definition's permissionMode can no longer widen the mode inherited from its parent context.never mined ·never mined never read tengu_agent_hooks_origin_untrusted ◇ Agents in untrusted folders can no longer define inline MCP servers; trust the project to allow it.never mined ·never mined never read tengu_agent_proxy_download_queue ◇ Download backlog on the agent proxy is now reported as a telemetry event.never mined ·never mined never read tengu_agent_proxy_failure_note ◇ Proxy connection failures during a command are summarised afterwards, grouped by host, with a curl pointer.never mined ·never mined never read tengu_agent_summary_skipped ◇ Background-agent summaries now get rejected outright if unparseable, with a running count of failuresnever mined ·never mined never read tengu_agent_tool_selected ◇ SDK result messages now include a summary of the subagents your session started.never mined ·never mined never read tengu_agent_worktree_cwd_escape_blocked ◇ Agent commands now have a concrete worktree boundary, so escapes get blocked with a clearer log line.never mined ·never mined never read tengu_alder_wicket ◇ Default false. When on, the system prompt gets a # Delivering work section (section id scope_fidelity) telling the model to deliver what was asked at the intended scope, finish the whole task and say plainly what is missing. The section is also turned on by a truthy CLAUDE_CODE_ALDER_WICKET env var or a tengu_alder_wicket value of true in the cached client data.never mined ·never mined never read tengu_amber_astrolabe ◇ Default false. One of the switches that add the autonomy_append system prompt section ("You are operating autonomously...": proceed with reversible actions without asking, report findings without applying a fix when the user is only asking, finish the work before ending the turn). The section is added when tengu_amber_sextant is on (default true) and either the current model's settings turn it on or this check passes; the check passes when CLAUDE_CODE_AMBER_ASTROLABE is set, the client data key of this name is true, or this flag is on.v2.1.251 → v2.1.259 ○gone never read tengu_amber_quartz ◇ No description from anybody.never mined ·never mined never read tengu_amber_quill ◇ Default false. Turns on context-aware tips when the allow_context_tips policy permits them: every few turns a classifier side query picks one tip from a list of situation-matched entries (at most 3 per session) and shows it. While on, the spinner drops its built-in /clear and /btw hints and clears the regular spinner tip, unless CLAUDE_CODE_FORCE_TIP_ID or a spinnerTipsOverride is set.v2.1.191 → v2.1.211 ○gone never read tengu_ant_attribution_header_new ◇ No description from anybody.never mined ·never mined never read tengu_ant_yolo_equiv_strip_config ◇ No description from anybody.v2.1.169 → v2.1.309 ◆in the build never read tengu_api_cache_breakpoints ◇ An analytics event, not a feature flag. Claude Code logs it after placing prompt cache markers on the messages for an API request, with totalMessageCount, cachingEnabled, skipCacheWrite, forkPointPinned and markerCount; a caller can turn the event off for a given request.never mined ·never mined never read tengu_api_no_response_timeout ◇ Hung requests are aborted when headers never arrive, and you can tune the window.never mined ·never mined never read tengu_api_query ◇ Query and result telemetry now carry a proactivity level field from your session.never mined ·never mined never read tengu_api_subagent_model_not_found ◇ A subagent failing on a missing model now tells you exactly which model was sent.never mined ·never mined never read tengu_api_success ◇ API success and error telemetry now includes your session's proactivity level.never mined ·never mined never read tengu_api_unrecognized_model ◇ You get a one-time warning line if a query names a model Claude Code does not recognise.never mined ·never mined never read tengu_artifact_ask ◇ Artifact publishing now recognizes a "still settling" server state and retries instead of failingnever mined ·never mined never read tengu_artifact_ask_granted ◇ Claude Code now logs telemetry when it asks for, and receives, permission to publish an artifactnever mined ·never mined never read tengu_artifact_five_class_asks ◇ Pinning someone else's artifact can now be auto-allowed instead of prompting, under an experimentnever mined ·never mined never read tengu_artifact_foreign_read_unasked ◇ A new unified consent system governs sensitive artifact actions like delete and access changesnever mined ·never mined never read tengu_artifact_legacy_verb ◇ A telemetry event records which legacy artifact verb was steered onto the new schema.never mined ·never mined never read tengu_artifact_mcp_list ◇ MCP tool calls to artifact servers now pass through argument-rewriting and result-observing hooksnever mined ·never mined never read tengu_artifact_mcp_switch ◇ MCP config building gets a new caching layer, currently unread/offnever mined ·never mined never read tengu_artifact_preview ◇ An artifact preview action would screenshot a page locally at several widths, but it cannot run here.never mined ·never mined never read tengu_artifact_publish ◇ Artifact publishing can copy files from other artifacts server-side, but it's off by default.never mined ·never mined never read tengu_artifact_reject_breaker ◇ The Artifact tool stops looping after three identical failures and reports the problem to you.never mined ·never mined never read tengu_artifact_thumb_direct ◇ Default false. Turns on thumbnail capture for the inline upload lane: artifact publishing starts a thumbnail capture of the page there, skipped when the Chrome preflight fails or the page is too large. The signed upload lane captures a thumbnail whenever the server provides a thumbnail upload URL, regardless of the flag; with the flag on, a failed signed PUT that falls back to inline keeps that running capture for the fallback instead of cancelling it with the reason index_put_failed.v2.1.208 → v2.1.210 ○gone never read tengu_artifact_tool_recovered ◇ New telemetry tracks when the artifact tool disappears and reappears mid-session.never mined ·never mined never read tengu_artifact_tool_withheld ◇ New telemetry tracks when the artifact tool disappears and reappears mid-session.never mined ·never mined never read tengu_artifact_toolset ◇ Artifact tool messages now describe actions as unified list/read calls with a scope, not old per-type namesnever mined ·never mined never read tengu_artifact_verify ◇ Claude could read console errors and failed loads from real viewers of an artifact, but it's switched off.never mined ·never mined never read tengu_artifact_verify_republish ◇ Claude could read console errors and failed loads from real viewers of an artifact, but it's switched off.never mined ·never mined never read tengu_ask_user_question_completed ◇ AskUserQuestion now logs detailed completion stats and shows a transcript line for itself againnever mined ·never mined never read tengu_assistant_output_harness_content ◇ Claude's own replies are scanned for text imitating internal control markup, and matches counted only.never mined ·never mined never read tengu_at_mention_peer_ ◇ Typing @name could message another live Claude session; the feature is off by default.never mined ·never mined never read tengu_attach_serve_link ◇ New 'Remote Control' system lets a cloud session dispatch tool execution to your local computernever mined ·never mined never read tengu_attachment_file_too_large ◇ An analytics event, not a feature flag. Claude Code logs it when an @-mentioned file is larger than the attachment size limit and its extension is not in an exempt set, in which case the file is not attached; fields are size_bytes and mode.never mined ·never mined never read tengu_attention_budget_cycle ◇ A per-session attention budget with up and down shortcuts is built but completely unreachable.never mined ·never mined never read tengu_auq_park_interrupted_at_stream_close ◇ Closing the control stream while a question waits can now cancel the turn, but that path stays off.never mined ·never mined never read tengu_auq_park_preserve_reverted ◇ Questions left open when the stream closes could stay answerable, but that behaviour is switched off for you today.never mined ·never mined never read tengu_auq_park_preserved_at_shutdown ◇ Questions left open when the stream closes could stay answerable, but that behaviour is switched off for you today.never mined ·never mined never read tengu_auto_background_agents ◇ Default false. When true, or when CLAUDE_AUTO_BACKGROUND_TASKS is truthy, a subagent started in the foreground by the Agent tool is moved to the background automatically after 120000 ms (2 minutes) if it is still running. When both are off, no automatic backgrounding timer is set.v2.1.138 or earlier → v2.1.177 ○gone never read tengu_auto_compact_config ◇ No description from anybody.never mined ·never mined never read tengu_auto_default_nudge_resolved ◇ Saying yes to the auto-mode prompt now actually saves auto as your default permission mode.never mined ·never mined never read tengu_auto_default_nudge_shown ◇ Dialog view events are recorded once per showing rather than on every redraw.never mined ·never mined never read tengu_auto_dream_completed ◇ An analytics event, not a feature flag. Claude Code logs it when a background auto-dream memory consolidation run finishes, with token usage (cache_read, cache_created, output), sessions_reviewed, daily_logs_found, files_touched_count and team_memory_enabled.never mined ·never mined never read tengu_auto_dream_failed ◇ An analytics event, not a feature flag. Claude Code logs it when a background auto-dream memory consolidation run throws an error other than a user abort, with phase (fork while the consolidation agent runs, completion during the bookkeeping after it) and error_class.never mined ·never mined never read tengu_auto_dream_fired ◇ An analytics event, not a feature flag. Claude Code logs it when an auto-dream memory consolidation run starts after its time and session gates pass, with hours_since (hours since the last consolidation, rounded), sessions_since (sessions to review) and team_memory_enabled.never mined ·never mined never read tengu_auto_dream_toggled ◇ New internal function lets code programmatically edit the auto-memory and auto-dream settingsnever mined ·never mined never read tengu_auto_memory_toggled ◇ New internal function lets code programmatically edit the auto-memory and auto-dream settingsnever mined ·never mined never read tengu_auto_mode_availability_changed ◇ Claude Code now tracks whether auto mode's availability changes during a session, not just at startupnever mined ·never mined never read tengu_auto_mode_beta_latch ◇ The automatic retry that would drop a rejected classifier beta can never trigger in this build.never mined ·never mined never read tengu_auto_mode_context_over_ceiling ◇ New telemetry fires when auto-mode context is too large and gets withheld entirely.never mined ·never mined never read tengu_auto_mode_context_static_over_budget ◇ Auto-mode classifier context now trims progressively to stay within a byte budget.never mined ·never mined never read tengu_auto_mode_context_withheld ◇ New telemetry fires when auto-mode context is too large and gets withheld entirely.never mined ·never mined never read tengu_auto_mode_decision ◇ Subagent results can be tagged to show which parts Claude Code wrote versus the model.never mined ·never mined never read tengu_auto_mode_denial_dialog_auto_denied ◇ Auto mode can show a prompt that denies itself after two minutes when you are away; off by default.never mined ·never mined never read tengu_auto_mode_denial_limit_exceeded ◇ More auto-mode permission events record whether the tool was from an MCP server.never mined ·never mined never read tengu_auto_mode_entry_warning_shown ◇ New automatic prompts appear: a first-time auto mode warning, a spending threshold dialog and a worktree tip.never mined ·never mined never read tengu_auto_mode_env_onboarding_later ◇ Auto-mode environment setup onboarding now waits for repeated denials before offering againnever mined ·never mined never read tengu_auto_mode_fallback_to_ask ◇ In auto mode, a question can be classified and handled instead of halting the run.never mined ·never mined never read tengu_auto_mode_git_state_probe ◇ Auto mode sessions now probe local git state and report telemetry about it.never mined ·never mined never read tengu_auto_mode_hook_allow_funneled ◇ Hook-approved tool calls could still be checked by the auto-mode classifier, but the flag is off.never mined ·never mined never read tengu_auto_mode_malformed_tool_input ◇ Auto-mode approval events now note whether the tool came from an MCP server.never mined ·never mined never read tengu_auto_mode_outcome ◇ Auto-mode fallback telemetry now records why a fallback model was used.never mined ·never mined never read tengu_auto_mode_outer_retry ◇ Auto mode retries instead of failing when the model that routes your turn is briefly unavailable.never mined ·never mined never read tengu_auto_mode_repo_visibility_lookup_failed ◇ An analytics event, not a feature flag. Claude Code logs it when it cannot determine a repository's visibility (public or private) from the GitHub API, with a reason field: gitlab, bitbucket or other_host for a non-GitHub host, parse_error or missing_fields for an unusable API response, http_error when the request fails, and policy_denied when the visibility probe is not allowed by policy.never mined ·never mined never read tengu_auto_mode_server_context_stale ◇ Auto mode now re-checks artifact facts to catch stale 'allow' verdicts from the server classifiernever mined ·never mined never read tengu_auto_mode_server_fallback_notice ◇ The auto-mode fallback warning can now appear inside SDK host apps that support it, when tengu_expressive_galaxy allowsnever mined ·never mined never read tengu_auto_mode_server_fallback_notice_closed ◇ The auto-mode server-fallback notice now records "unanswered" when it closes without a choice, and the dialog can be cancelled from outsidenever mined ·never mined never read tengu_auto_mode_setup_wizard_answers ◇ Auto-mode setup is one screen now instead of three separate questions.never mined ·never mined never read tengu_auto_mode_setup_wizard_shown ◇ Reopening the auto-mode wizard can no longer let a stray keypress confirm instantly.never mined ·never mined never read tengu_autocompact_offer_resolved ◇ No description from anybody.never mined ·never mined never read tengu_autofix_pr ◇ No description from anybody.never mined ·never mined never read tengu_autofix_pr_result ◇ An analytics event, not a feature flag. Claude Code logs it when an /autofix-pr run ends, with a result field: success (a cloud autofix session was spawned or one is already monitoring the PR), success_current_session (monitoring set up in the current session), deferred_to_pr_steward (PR Steward is already handling the PR), cancelled, or failed together with an error_code.never mined ·never mined never read tengu_autofix_pr_started ◇ An analytics event, not a feature flag. Claude Code logs it when an /autofix-pr run starts, with action set to start and has_pr_number, has_repo_path and has_repo_ref recording which of those inputs were given.never mined ·never mined never read tengu_background ◇ No description from anybody.v2.1.287 ○gone never read tengu_basalt_3kr ◇ No description from anybody.never mined ·never mined never read tengu_bash_allowlist_strip_all ◇ Default false. Its value is recorded as stripAllBashFlag on the tengu_auto_mode_decision telemetry event sent when the auto mode classifier reviews a tool call; the fast-path and subagent-handoff decision events do not carry it, and no other code reads the flag in this version.v2.1.138 or earlier → v2.1.191 ○gone never read tengu_bash_ast_too_complex ◇ Bash's 'too complex to check' read-block guard now applies via a general table covering more refusal reasons, not just one hardcoded casenever mined ·never mined never read tengu_bash_command_clamp_denied ◇ If a clamped agent's permission check crashes, the shell command is denied rather than allowed.never mined ·never mined never read tengu_bash_dangerous_rm_too_complex ◇ The dangerous-rm safety check for bash commands can now be extended with an extra custom checknever mined ·never mined never read tengu_bash_security_check_triggered ◇ No description from anybody.never mined ·never mined never read tengu_bash_task_ack ◇ Backgrounded shell commands may return a structured task record instead of raw output.never mined ·never mined never read tengu_bash_task_delivered ◇ Finished background tasks now report whether their output was spliced back in or lost.never mined ·never mined never read tengu_bash_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_bash_tool_command_failed ◇ Bash tool telemetry now records a detailed breakdown of each command's shapenever mined ·never mined never read tengu_bedrock_upgrade_check ◇ Bedrock and Vertex now share one check for outdated pinned model variables.never mined ·never mined never read tengu_began_setup ◇ Onboarding now also treats a 'gateway' authentication setup as equivalent to OAuth for its flownever mined ·never mined never read tengu_bergotte_lantern ◇ No description from anybody.never mined ·never mined never read tengu_bg_adopt_upgrade_respawn ◇ Background sessions stop needlessly restarting when your running build is already newer.never mined ·never mined never read tengu_bg_agent_action ◇ Background-agent telemetry no longer records which agent template was usednever mined ·never mined never read tengu_bg_attach_outcome ◇ Background attach events now record cold starts, surface and first frame details.never mined ·never mined never read
Feedback