Feature flags, all 2,081.

tengu_serene_teapot ◇ Claude Code no longer opens the server notification stream for claude.ai connectors, and answers that request itself with a 405v2.1.281 → v2.1.308 ◆in the build AND v2.1.281 tengu_serialized_petal ◇ A new setting controlled from Anthropic's servers is a number that is 0 unless the server sets it and two conditions are metv2.1.302 → v2.1.308 ◆in the build AND v2.1.302 tengu_sharded_moonbeam ◇ Claude Code can now show a rate-limit status from recently-remembered windows when a fresh server reading isn't availablev2.1.267 → v2.1.308 ◆in the build AND v2.1.267 tengu_shimmering_firefly ◇ New instructions teach a status classifier that a message opening with a request means work is blocked, behind a switch that defaults offv2.1.292 → v2.1.308 ◆in the build AND v2.1.292 tengu_shimmying_mochi ◇ Claude Code's bash-safety checker now rejects more risky variable-expansion patternsv2.1.274 → v2.1.308 ◆in the build AND v2.1.274 tengu_shiny_stardust ◇ A gated code path can now make certain search-matched MCP tools load eagerly instead of staying deferredv2.1.280 → v2.1.308 ◆in the build AND v2.1.280 tengu_shoji_engine ◇ No description from anybody.v2.1.193 → v2.1.202 ○gone AND v2.1.308 tengu_silent_heron ◇ When a background shell command produces no new output for a while, Claude is sent a check-in note, if a server sets the waiting timev2.1.290 → v2.1.308 ◆in the build AND v2.1.290 tengu_slate_bittern ◇ A new prompt section adds reduced-delegation guidance for Opus-class models, skipped if already cached in memory.v2.1.257 → v2.1.308 ◆in the build AND v2.1.301 tengu_slate_fern ◇ Default false. Turns on the unused plugin check: user-installed marketplace plugins not used for at least days days and sessions sessions feed the plugin-disuse-review startup tip (You haven't used the ... plugin in a while) and the days-since-last-use shown in the plugin list. Plugins that provide LSP servers, themes, output styles, monitors or workflows are never flagged, and the check is off when the strictKnownMarketplaces policy is set. true uses 14 days and 10 sessions; an object can set numeric days and sessions, each falling back to those values.v2.1.172 → v2.1.186 ○gone AND v2.1.308 tengu_slate_harbor ◇ REPL contexts moved to a registry that releases timers even on error paths.v2.1.138 or earlier → v2.1.274 ○gone AND v2.1.248 tengu_slate_harrier ◇ Claude no longer spends up to a minute investigating before asking you a clarifying question.v2.1.138 or earlier → v2.1.222 ○gone AND v2.1.223 tengu_slate_kestrel ◇ Default false. When on, a claude.ai OAuth login with no known subscription type (null) counts as eligible for organization policy limits; otherwise an OAuth login is eligible only on an enterprise or team subscription. A login with an API key is eligible either way, and the other checks (first-party API, the token's scope) still apply.v2.1.138 or earlier → v2.1.148 ○gone AND v2.1.308 tengu_slate_marten ◇ Claude Code now reports whether early attachment preparation was off, on, or on including PDFs, controlled by two switches off by defaultv2.1.292 → v2.1.308 ◆in the build AND v2.1.292 tengu_slate_meadow ◇ Default false. When true, enables the background-agents fleet gate (isAgentsFleetEnabled) and adds the built-in Claude agent to the built-in agent list, unless agent view is disabled: by CLAUDE_CODE_DISABLE_AGENT_VIEW, the disableAgentView setting, or a third-party API provider. Both are also on without the flag when Claude Code runs in essential-traffic-only mode (for example with CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC set).v2.1.138 or earlier → v2.1.140 ○gone AND v2.1.308 tengu_slate_quill ◇ Default false. One of the gates for the DesignSync tool and the /design, /design-consent, /design-revoke and /design-login commands. They are off when the allow_design_sync policy check fails or another internal check disables them, on when a further internal check passes, and otherwise follow this flag.v2.1.160 → v2.1.245 ○gone AND v2.1.308 tengu_slate_quoin ◇ Artifact syntax highlighting and mermaid diagrams are now always on; three other gates were renamed and stay off.v2.1.239 → v2.1.267 ○gone AND v2.1.239 tengu_slate_siskin ◇ Config for the memory-write survey, default {enabled: false, timeoutMs: 8000, throttleMs: 30000, summaryLineThreshold: 5}; valid fields override the defaults, and a value that fails the schema is logged and ignored entirely. With enabled on (and product feedback allowed, among other checks), a main-agent write to a file directly inside the memory directory, other than MEMORY.md, is shown in an Approve/Reject dialog, and rejecting undoes the write. throttleMs spaces dialogs apart, the dialog closes with a timeout outcome after timeoutMs, and a write over summaryLineThreshold lines gets a one-sentence model summary.v2.1.138 or earlier → v2.1.161 ○gone AND v2.1.120 tengu_slate_wicket ◇ Claude Code can now refuse names with hidden or direction-changing characters, < or >, or more than 256 charactersv2.1.299 → v2.1.308 ◆in the build AND v2.1.299 tengu_sleepy_rabbit ◇ The plan and auto mode check now also looks at the mode that was active before a rewrite, and a line-reading helper sits behind a server switchv2.1.285 → v2.1.308 ◆in the build AND v2.1.285 tengu_sleepy_ripple ◇ A server-controlled switch, on unless turned off remotely, now decides whether a step runs just before Claude Code compacts a conversationv2.1.307 → v2.1.308 ◆in the build AND v2.1.307 tengu_sleepy_shore ◇ The web search tool has a built-in choice between a quick, cheap search and a slower, thorough one, but it stays off unless configuredv2.1.282 → v2.1.308 ◆in the build AND v2.1.283 tengu_sleepy_snowflake ◇ New 'sleepy_snowflake' setting controls how widely kept reminders are delivered, per model and conversationv2.1.271 → v2.1.308 ◆in the build AND v2.1.282 tengu_slim_subagent_claudemd ◇ Default true. When on, subagents whose definition sets omitClaudeMd (the built-in Explore and Plan agents in this version) start without the CLAUDE.md content in their user context, unless the caller passes its own user context. When off, they receive the CLAUDE.md content like other subagents.v2.1.138 or earlier → v2.1.195 ○gone AND v2.1.308 tengu_sm_config ◇ No description from anybody.never mined ·never mined AND v2.0.68 tengu_smooth_harbor ◇ A server flag can widen when subagent requests send the extended usage-limit header.v2.1.246 → v2.1.308 ◆in the build AND v2.1.292 tengu_snappy_plum ◇ A new way to pass MCP server metadata to plugins was added; it stays off unless switched on remotelyv2.1.295 → v2.1.308 ◆in the build AND v2.1.295 tengu_snoopy_creek ◇ Claude Code now sends usage data about each permission decision, recording what kind of rule or check made itv2.1.307 → v2.1.308 ◆in the build AND v2.1.307 tengu_snug_reef ◇ MCP servers that failed with a retryable error can be retried automatically at the start of each turn, for a limited number of turnsv2.1.295 → v2.1.308 ◆in the build AND v2.1.295 tengu_snug_storm ◇ A notice for Max and Team plans about an API credit link is built in, with its wording supplied remotely and a limit on how often it showsv2.1.302 → v2.1.308 ◆in the build AND v2.1.302 tengu_soft_slate_nudge ◇ Default "baseline". Setting it to "off" stops the periodic todo and task reminders that are otherwise attached after 10 assistant turns without a todo or task update and 10 turns since the last reminder; any other value keeps them. CLAUDE_CODE_TODO_REMINDER_MODE overrides it.v2.1.178 → v2.1.308 ◆in the build AND v2.1.178 tengu_sorrel_trellis_gusset ◇ Claude Code has a new remotely controlled switch, off unless turned on, with a check for an "arbiter" permission mode beside itv2.1.288 → v2.1.308 ◆in the build AND v2.1.288 tengu_sorrel_trellis_lintel ◇ A comment sent moments before a watch starts could get a reply, but the flag is off.v2.1.236 → v2.1.267 ○gone AND v2.1.308 tengu_sorted_otter ◇ The dialog for MCP servers that ask you to sign in again with extra permissions now falls back to on when the server sends no settingv2.1.288 → v2.1.308 ◆in the build AND v2.1.288 tengu_sorted_zebra ◇ Behind a switch that is off by default, auto mode registers tools reached through toolAliases or warns when an allow rule would skip its checkv2.1.295 → v2.1.308 ◆in the build AND v2.1.295 tengu_spicy_blossom ◇ Running plugin install --replace without a name@marketplace id now says that id is needed and that nothing was installedv2.1.287 → v2.1.308 ◆in the build AND v2.1.287 tengu_splendid_sutton ◇ A new internal check decides when Claude Code sends extra gateway hint headers on first-party requestsv2.1.273 → v2.1.308 ◆in the build AND v2.1.283 tengu_squishy_newt ◇ Messages relayed from a project's coordinator session are handled by a switch that defaults to on, and the model is told when one is waitingv2.1.282 → v2.1.300 ○gone AND v2.1.301 tengu_squishy_spring ◇ Worktree file copying can now try a fast atomic rename before falling back to the old method, behind a gatev2.1.275 → v2.1.308 ◆in the build AND v2.1.275 tengu_staged_ladybug ◇ Claude Code now counts turns since a person last wrote and can act once a limit is reached, but this is off unless switched on remotelyv2.1.296 → v2.1.308 ◆in the build AND v2.1.300 tengu_starry_squid ◇ Programs built on the SDK can now receive tool calls and results from subagents nested inside other subagentsv2.1.290 → v2.1.308 ◆in the build AND v2.1.290 tengu_starry_waterfall ◇ No description from anybody.v2.1.304 → v2.1.308 ◆in the build AND v2.1.308 tengu_startup_announcements ◇ Default []. A list of startup announcements, each with id, text, optional title, footer and requiresModel, priority (default 0), maxImpressions (default 3) and accentBar (default true). Claude Code picks the highest-priority entry that has been shown fewer than maxImpressions times and whose requiresModel, if set, passes a model check; a value that fails the schema falls back to [].never mined ·never mined AND v2.1.187 tengu_startup_notice ◇ Default "". A non-empty string is shown as a high-priority, warning-colored notification for 30 seconds. The value is re-read when flag values update: a new string replaces the notification and an empty one removes it.v2.1.140 → v2.1.308 ◆in the build AND v2.1.289 tengu_static_title_under_mux ◇ Inside tmux, screen or zellij your terminal title stays static instead of animating.v2.1.236 → v2.1.308 ◆in the build AND v2.1.236 tengu_steadfast_wren ◇ Claude Code adds several new remotely controlled switches whose purpose is not yet visiblenever mined ·never mined AND v2.1.299 tengu_steady_finch ◇ A recently added server-controlled switch was swapped for a new one that falls back to off; what it controls is still unknownnever mined ·never mined AND v2.1.305 tengu_steady_mitten ◇ Enterprise organizations can be shown an in-app survey, off unless switched on remotely and skipped with CLAUDE_CODE_DISABLE_FEEDBACK_SURVEYv2.1.304 → v2.1.308 ◆in the build AND v2.1.304 tengu_steady_plover ◇ When you resume a session whose last reply was recently cut off, Claude Code now treats it as an interrupted turn and can nudge to continuev2.1.299 → v2.1.308 ◆in the build AND v2.1.301 tengu_steady_plum ◇ The effort-downgrade nudge now works per model instead of just one, but each model needs its own flag turned on.never mined ·never mined AND v2.1.308 tengu_steady_puppy ◇ Claude Code can now prepare background agents orphaned by a restart so they can be resumed, and the restart notice says which onesv2.1.304 → v2.1.308 ◆in the build AND v2.1.304 tengu_still_heron ◇ If the same usage-limit refusal repeats with the same limits, Claude Code marks it as a repeat, skips the model request and holds the result for later deliveryv2.1.292 → v2.1.308 ◆in the build AND v2.1.292 tengu_stream_watchdog_default_on ◇ Stalled streams are now detected across direct, gateway and Bedrock connections, and you can force it with an env var.v2.1.138 or earlier → v2.1.308 ◆in the build AND v2.1.248 tengu_streamed_thimble ◇ A session waiting at the login notice now clears itself when you log in from another Claude Code windowv2.1.281 → v2.1.308 ◆in the build AND v2.1.281 tengu_streaming_tool_execution2 ◇ Default false. Turns on streaming tool execution in the query loop: tool calls are handed to an executor as their tool_use blocks arrive while the model's response is still streaming, completed results are emitted during the stream, and the remaining results are collected after it ends. The executor is discarded and recreated when a response is retried or restarted.v2.1.138 or earlier → v2.1.153 ○gone AND v2.1.308 tengu_structured_creek ◇ In auto mode, tool calls that Claude Code generates itself now skip the classifier and are allowedv2.1.281 → v2.1.284 ○gone AND v2.1.285 tengu_sunny_locket ◇ A remotely controlled option can compact a long, idle conversation shortly before its one-hour prompt cache runs outv2.1.283 → v2.1.306 ○gone AND v2.1.307 tengu_sunset_penguin_opus46 ◇ Default "2026-06-29". Date (parsed with Date.parse) on which Opus 4.6 fast mode is removed: when fast mode is turned on while the current model is claude-opus-4-6 and the date is still in the future, Claude Code shows the warning Opus 4.6 fast mode is deprecated and will be removed on <date>, with the date formatted like Jun 29 in UTC. An unparseable or past date shows no warning.v2.1.174 → v2.1.191 ○gone AND v2.1.174 tengu_sunset_penguin_opus47 ◇ Default "2026-07-25". Removal date for Opus 4.7 fast mode: when fast mode is turned on (via /fast or the picker) while the current model is claude-opus-4-7 and the date is still in the future, a warning notice Opus 4.7 fast mode is deprecated and will be removed on <date> appears, with the date shown like Jul 25 in UTC. An unparseable or past date shows no notice.v2.1.196 → v2.1.220 ○gone AND v2.1.196 tengu_surface_failed_mcp_servers ◇ MCP servers that fail to connect are now reported to you by default.v2.1.205 → v2.1.308 ◆in the build AND v2.1.247 tengu_swann_brevity ◇ No description from anybody.never mined ·never mined AND v2.1.66 tengu_swift_lynx ◇ A hidden new tool lets Claude open the page for a startup promotional offerv2.1.275 → v2.1.308 ◆in the build AND v2.1.296 tengu_swift_whistle ◇ Subagent report truncation accounting for a 'harness tail' section stays inert behind a flag.v2.1.261 → v2.1.308 ◆in the build AND v2.1.261 tengu_swirling_meteor ◇ New gate 'tengu_swirling_meteor' controls grouping of background task notificationsv2.1.274 → v2.1.308 ◆in the build AND v2.1.299 tengu_swirling_squid ◇ Artifact settings reports now include inherited_type_grant when an environment variable or the tengu_swirling_squid flag turns it onv2.1.281 → v2.1.308 ◆in the build AND v2.1.281 tengu_synthetic_fog ◇ When a remote tool answer is too big, Claude Code now shrinks embedded images to fit before refusing, and the size limit can be set remotelyv2.1.284 → v2.1.308 ◆in the build AND v2.1.284 tengu_tallow_quoit ◇ Team and Enterprise members who hit their own spend limit in VS Code can get a 'Request more' button that notifies their organization's adminsv2.1.307 → v2.1.308 ◆in the build AND v2.1.307 tengu_teleport_github_access_precheck ◇ Teleporting a session to the cloud now checks GitHub access ahead of time and can fail earlierv2.1.268 → v2.1.308 ◆in the build AND v2.1.300 tengu_temporal_hare ◇ Claude Code has a new check for a slimmed-down prompt variant, off unless it is switched on remotelyv2.1.301 → v2.1.308 ◆in the build AND v2.1.301 tengu_tender_truffle ◇ A check that matched MCP servers against a remote list can now read a single on/off switch instead, depending on how it is callednever mined ·never mined AND v2.1.300 tengu_terminal_sidebar ◇ Session status in your terminal tab needs a server gate as well as the setting.v2.1.138 or earlier → v2.1.308 ◆in the build AND v2.1.301 tengu_tern_alloy ◇ Default "off". Any other value makes the subagent-fanout-nudge tip eligible, with a 3-session cooldown, on the first-party API for users not signed in with a claude.ai subscription. The value copy_b shows For big tasks, tell Claude to use subagents. They work in parallel and keep your main thread clean.; any other non-off value shows Say "fan out subagents" and Claude sends a team. Each one digs deep so nothing gets missed.v2.1.138 or earlier → v2.1.140 ○gone AND v2.1.308 tengu_thistle_grebe ◇ Subagent nudging behaviour can be picked per model, but the whole feature stays behind a gate.never mined ·never mined AND v2.1.227 tengu_tide_elm ◇ Default "off". Any other value makes the effort-high-nudge tip, which suggests /effort high and has a 3-session cooldown, eligible on the first-party API for users not signed in with a claude.ai subscription, when the current model supports effort, no policy effortLevel is set, CLAUDE_CODE_EFFORT_LEVEL is unset or invalid (its values auto and unset also hide the tip), and the effortLevel setting is not high, xhigh or max. The value copy_b shows Use /effort high for better one-shot answers. Claude thinks it through first.; any other non-off value shows Working on something tricky? /effort high gives better first answers.v2.1.138 or earlier → v2.1.140 ○gone AND v2.1.308 tengu_tidy_abacus ◇ /usage can now point out what is costing you most and suggest fixes such as /compact or lower effort, which you can queue as a follow-upv2.1.305 → v2.1.308 ◆in the build AND v2.1.305 tengu_tidy_cloud ◇ When a response stops arriving because it was suspended, Claude Code now retries it a limited number of times before its usual retry handlingv2.1.290 → v2.1.308 ◆in the build AND v2.1.290 tengu_tidy_eclipse ◇ When a tool request contains details the tool does not recognise, Claude Code can now drop them and try againv2.1.290 → v2.1.308 ◆in the build AND v2.1.290 tengu_tidy_fern ◇ New gate tengu_tidy_fern can add an extra value into how user messages are builtv2.1.274 → v2.1.308 ◆in the build AND v2.1.274 tengu_tidy_frost ◇ Subscribers who get the 1-hour prompt cache can be moved to a 5-minute cache for certain models when a server-side switch is onv2.1.293 → v2.1.308 ◆in the build AND v2.1.293 tengu_tidy_lemon ◇ A remotely controlled option lets team members send a more-usage request to their admin directly from /rate-limit-optionsv2.1.283 → v2.1.308 ◆in the build AND v2.1.283 tengu_tidy_origami ◇ Claude Code reads a new remote switch whose purpose is not known, and treats it as off when the server sends nothingv2.1.286 → v2.1.308 ◆in the build AND v2.1.286 tengu_tidy_sloth ◇ Claude is now told that a background task's result reaches it only if the task finishes before Claude gives its final responsev2.1.292 → v2.1.308 ◆in the build AND v2.1.292 tengu_tidy_widget ◇ A new tengu_tidy_widget gate helps decide whether a listing's preconditions are adopted when publishing an artifactv2.1.281 → v2.1.308 ◆in the build AND v2.1.281 tengu_timber_lark ◇ Default "off". Controls the loop-command-nudge tip for /loop: "off" hides it, "copy_b" shows Use /loop 5m check the deploy to run any prompt on a schedule. Set it and forget it., and any other value shows /loop runs any prompt on a recurring schedule. Great for monitoring deploys, babysitting PRs, or polling status. The tip also needs cron scheduling to be enabled (CLAUDE_CODE_DISABLE_CRON not set true and tengu_kairos_cron on) plus a further eligibility check, and has a cooldown of 3 sessions.v2.1.138 or earlier → v2.1.140 ○gone AND v2.1.308 tengu_tingly_hopper ◇ Always-loaded claude.ai MCP tools are held out of the declared tool set and loaded lazily until the tool-use decision settles.v2.1.257 → v2.1.267 ○gone AND v2.1.308 tengu_tingly_pearl ◇ axScreenReader is now read only from trusted settings, and Claude Code explains when a project's request to enable it was ignoredv2.1.305 → v2.1.308 ◆in the build AND v2.1.305 tengu_tingly_pike ◇ How Claude Code prompts the model to continue after hitting its output limit, and when it gives up, can now be set by a server switchv2.1.305 → v2.1.308 ◆in the build AND v2.1.305 tengu_tips_mod ◇ The spinner tip line shown while Claude works was rebuilt as an internal 'tips' pluginnever mined ·never mined AND v2.1.273 tengu_toasty_breeze ◇ Hitting the 5-hour limit could offer to keep working at reduced priority, but it is off for everyone.v2.1.239 → v2.1.308 ◆in the build AND v2.1.308 tengu_tool_cgroup_exclude_classes ◇ Memory limits now reach npm, git, tmux, LSP, MCP and hooks, but everything except Bash is excluded.never mined ·never mined AND v2.1.246 tengu_tool_detach ◇ A long-running tool call can now be moved to the background instead of blocking when a new message comes inv2.1.277 → v2.1.308 ◆in the build AND v2.1.287 tengu_tool_search_unsupported_models ◇ Default null. A list of model name substrings for which tool search is treated as unsupported: a model whose name contains one (case-insensitive) does not get tool search and reports model_unsupported where that reason is used. A non-array value falls back to the built-in list claude-3-5-haiku, claude-3-haiku.v2.1.138 or earlier → v2.1.308 ◆in the build AND v2.1.308 tengu_tools_exclusion_killswitch ◇ Tool presets can now exclude specific tools with a '!ToolName' syntaxv2.1.273 → v2.1.308 ◆in the build AND v2.1.273 tengu_tranquil_cloud ◇ Claude Code now strips invisible characters like zero-width marks from pasted or typed text before sending itv2.1.277 → v2.1.308 ◆in the build AND v2.1.300 tengu_tranquil_fern ◇ Resuming a session that was in plan mode now puts you back in plan mode.v2.1.239 → v2.1.299 ○gone AND v2.1.275 tengu_tranquil_starlight ◇ The long Claude in Chrome guidance on GIFs, console logs, alerts and tabs is gone from its old place, and tool search now mentions a batch browser toolv2.1.305 → v2.1.308 ◆in the build AND v2.1.305 tengu_tranquil_summit ◇ Auto mode now stops itself instead of looping forever when safety checks keep failingv2.1.280 → v2.1.304 ○gone AND v2.1.280 tengu_transcript_local_gc ◇ CLAUDE_CODE_TRANSCRIPT_LOCAL_GC, which turns on local transcript clean-up, is now accepted from managed policy settingsv2.1.217 → v2.1.308 ◆in the build AND v2.1.287 tengu_truncated_response_recovery ◇ A cut-off streaming reply in claude -p is now retried automatically, asking the model to resume.v2.1.246 → v2.1.308 ◆in the build AND v2.1.246 tengu_twinkling_boole ◇ Claude now warns when dispatching multiple write-capable agents into the same working directoryv2.1.280 → v2.1.308 ◆in the build AND v2.1.280