Follow Discord
Sweep 22 Oct 2026 · 04:32Z Build v2.1.308 527 read Stable v2.1.308 Latest v2.1.308 Next v2.1.308 Feeds RSS JSON llms.txt llms-full.txt Unofficial
One family

Feature flags, all 2,081.

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

Feature flags2,081seen across 149 builds
In v2.1.308889the newest inventory on disk
Gone314in an earlier build, not in the newest
Described0have a line in Anthropic's documentation
No docs line2,081no line in the documentation we hold

Every feature flag, and where it lived

285 of 2,081 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. 1,143 of 2,081 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 v2.1.308, ◇ we found no documentation describing it.

Feature flagWhat it doesSpan v2.1.138 → v2.1.308 Status Reading
tengu_account_skills_sync_enabled ◇ Account skills-sync eligibility logic generalized into a reusable descriptorv2.1.234 ○gone AND v2.1.235 tengu_agent_list_attach ◇ Default false. When CLAUDE_CODE_AGENT_LIST_IN_MESSAGES is unset or not a recognised boolean, true sends the Agent tool's list of agent types as agent_listing_delta attachments in the conversation (after the first, only the added and removed types) and leaves it out of the tool description. A truthy CLAUDE_CODE_DISABLE_ATTACHMENTS or CLAUDE_CODE_SIMPLE keeps the list in the description.v2.1.138 or earlier → v2.1.177 ○gone AND v2.1.178 tengu_agile_glade ◇ Usage reports about individual features can now be thinned out or dropped per feature by a server setting, without a new releasev2.1.293 → v2.1.308 ◆in the build AND v2.1.293 tengu_amber_redwood2 ◇ You can now set when auto-compaction kicks in with a new --autocompact flag, and /autocompact works for everyone.v2.1.138 or earlier → v2.1.308 ◆in the build AND v2.1.221 tengu_amber_redwood3 ◇ You can now set when auto-compaction kicks in with a new --autocompact flag, and /autocompact works for everyone.v2.1.154 → v2.1.308 ◆in the build AND v2.1.221 tengu_artifact_hljs_highlight ◇ Artifact syntax highlighting and mermaid diagrams are now always on; three other gates were renamed and stay off.v2.1.212 → v2.1.238 ○gone AND v2.1.239 tengu_artifact_mermaid_diagrams ◇ Artifact syntax highlighting and mermaid diagrams are now always on; three other gates were renamed and stay off.v2.1.202 → v2.1.238 ○gone AND v2.1.239 tengu_async_snowglobe ◇ Claude Code recognises a new local memory import background task and reads a stale note setting for memory from remote settingsnever mined ·never mined AND v2.1.286 tengu_atomic_ocean ◇ The gzip decision now records exactly why it skipped, though compression stays off without a server value.v2.1.236 → v2.1.287 ○gone AND v2.1.300 tengu_auto_mode_config ◇ Auto-mode permission checks can score risk 0-100 instead of yes/no, but only for configured models.v2.1.138 or earlier → v2.1.308 ◆in the build AND v2.1.307 tengu_ax_screen_reader ◇ The information Claude Code gives the auto mode safety check can now carry an extra marker, and a new prompt-screening step was addedv2.1.300 → v2.1.308 ◆in the build AND v2.1.295 tengu_bg_low_mem_mb ◇ Low-memory threshold in MiB for background sessions, code default 1024; 0 or less turns the check off. Outside macOS the machine counts as low on memory when the system memory reading is below the threshold; on macOS the value only switches the check on and macOS's critical memory-pressure level decides. While memory is low, the background-session supervisor retires settled workers sooner and before spawning a new session, stops prewarming, keeps no pre-started spare, and a worker that crashes before starting reports possibly low memory.v2.1.138 or earlier → v2.1.308 ◆in the build AND v2.1.232 tengu_bg_prewarm_burst_concurrency ◇ Background session prewarm now checks the concurrency limit instead of the delay value.v2.1.206 → v2.1.308 ◆in the build AND v2.1.248 tengu_bg_prewarm_burst_delay_ms ◇ Default 15000 (milliseconds). How long the background-session supervisor waits, once it is ready, before its prewarm burst: restarting idle background sessions on a stale Claude Code version onto the running one, tengu_bg_prewarm_burst_concurrency (default 3) at a time. The burst is skipped if the supervisor is shutting down, the concurrency is 0 or less, or tengu_bg_attach_upgrade is off.v2.1.206 → v2.1.308 ◆in the build AND v2.1.248 tengu_bg_prewarm_per_sweep ◇ The kite network transport is on by default everywhere, and CLAUDE_CODE_HARBOR_KITE=0 now really turns it off.v2.1.163 → v2.1.308 ◆in the build AND v2.1.248 tengu_binary_music ◇ Users with a one-million-token context can be offered a choice of when auto-compact kicks in, from 300k to 900k tokensv2.1.300 → v2.1.308 ◆in the build AND v2.1.301 tengu_birch_lantern ◇ Default "off". Picks how /powerup is advertised in a session where the first-run onboarding was shown: "banner" shows an info notice ("New here? Type /powerup for a 5-minute tour..."), "step" adds a "Learn the moves" step after onboarding that can launch the tour, and any other value shows neither. CLAUDE_CODE_POWERUP_ONBOARDING set to banner or step overrides the flag.v2.1.181 → v2.1.308 ◆in the build AND v2.1.181 tengu_bison_cairn ◇ Default false. When on, the system prompt includes the # Delivering work section (guidance on doing the requested scope, handling uncertainty mid-task and proceeding after the user reaffirms a request). The section is also included when the cached client data has tengu_bison_cairn: true, for models with the fable_5_1_prompt_bundle capability (except under certain entrypoints), and for models with the opus_5_prompt_bundle capability while tengu_fennel_godwit is off; CLAUDE_CODE_BISON_CAIRN set to a true or false value overrides all of these.never mined ·never mined AND v2.1.217 tengu_bramble_lintel ◇ When a pass that saves memories from your conversation stops early, the next pass looks at the same turns againv2.1.138 or earlier → v2.1.308 ◆in the build AND v2.1.305 tengu_breezy_pelican ◇ A token limit set to a fraction of the context window, by default one sixth, is built in and stays off unless switched on remotelyv2.1.292 → v2.1.308 ◆in the build AND v2.1.292 tengu_bridge_attestation_enforce ◇ Tool requests from a bridged device can now be refused before the tool runs, but no tool uses this yetv2.1.141 → v2.1.287 ○gone AND v2.1.288 tengu_bridge_attestation_enforce_config ◇ The switch that turns on device attestation enforcement for the bridge is read through a safer path, and its settings now come from several sourcesv2.1.141 → v2.1.296 ○gone AND v2.1.288 tengu_bridge_min_version ◇ Remote Control now checks your Claude Code version meets a minimum before accepting a remotely started sessionnever mined ·never mined AND v2.1.273 tengu_bridge_recover_stale_epoch ◇ Remote sessions gain reconnect and reactivation heartbeats plus stale-connection recovery, all off here.v2.1.228 → v2.1.233 ○gone AND v2.1.234 tengu_bridge_system_init ◇ A dormant bridge startup switch was removed entirely; it now always reports off and can't be turned on.v2.1.138 or earlier → v2.1.246 ○gone AND v2.1.247 tengu_bridge_unarchive_on_resume ◇ Resuming an older reaped session now unarchives it first instead of failing.v2.1.242 → v2.1.303 ○gone AND v2.1.228 tengu_bright_ripple ◇ Memory context fetching gains a new 'laptop' path alongside the existing cowork path, both currently inactivev2.1.271 → v2.1.276 ○gone AND v2.1.277 tengu_brisk_meadow ◇ Whether 'inline tools' are enabled is now decided by a consolidated check with environment-variable overridesnever mined ·never mined AND v2.1.280 tengu_buzzing_lightning ◇ A new AppifactRepl tool runs a full JavaScript program against an Artifact's data instead of one call at a timenever mined ·never mined AND v2.1.272 tengu_buzzing_pike ◇ Claude Code can compress the requests it sends to the model in gzip blocks, controlled by CLAUDE_CODE_GZIP_REQUEST_BODY_BLOCKS and off by defaultv2.1.287 → v2.1.308 ◆in the build AND v2.1.287 tengu_byte_stream_idle_timeout_ms ◇ If you leave the timeout variables unset, Claude Code can now take stream-idle and request timeouts from a received configurationv2.1.157 → v2.1.308 ◆in the build AND v2.1.299 tengu_cached_pebble ◇ Claude Code has a new remote switch that falls back to off when the server sends nothing, and what it controls is not knownv2.1.302 → v2.1.303 ○gone AND v2.1.305 tengu_calm_bear ◇ A feature that the server could switch on is now hard-coded off, so a server-side setting can no longer enable itv2.1.288 → v2.1.289 ○gone AND v2.1.290 tengu_calm_eagle ◇ Claude Code adds two server-controlled switches, a path for remote-session tools over MCP, and a new label for Claude.ai proxy serversnever mined ·never mined AND v2.1.290 tengu_calm_seal ◇ Claude Code can now sort a model into a family such as Fable, Haiku, Mythos, Opus or Sonnet, and names it only when a server-side list allows itv2.1.292 → v2.1.308 ◆in the build AND v2.1.292 tengu_carved_slate ◇ Claude Code now only re-sends changed CLAUDE.md/instruction files instead of the whole block each turn, saving tokens.v2.1.251 → v2.1.267 ○gone AND v2.1.265 tengu_ccr_bundle_seed_enabled ◇ If a folder is too big to seed a cloud session, you now see the size, cap and biggest files.v2.1.227 → v2.1.241 ○gone AND v2.1.273 tengu_ccr_context_build_after_resume_enabled ◇ The wake-up and retry steps for background MCP tasks now depend on a new remote switch that is off by defaultv2.1.275 → v2.1.284 ○gone AND v2.1.285 tengu_ccr_keepalive_after_stop_enabled ◇ Remote (CCR) sessions gain keepalive messages after a turn stops, currently off by defaultv2.1.277 → v2.1.280 ○gone AND v2.1.281 tengu_ccr_no_subscriber_flush_ms ◇ Subscriber-aware flush buffering is disabled by its default of zero.v2.1.248 → v2.1.308 ◆in the build AND v2.1.248 tengu_ccr_replay_acks_at_request_dispatch ◇ A server setting used for one cloud session type was only renamed internally and still falls back to offv2.1.280 → v2.1.300 ○gone AND v2.1.301 tengu_ccr_sse_keepalive_max_interval_s ◇ Realtime SSE connection now negotiates its keepalive timing with the server instead of using a fixed valuev2.1.271 → v2.1.308 ◆in the build AND v2.1.296 tengu_ccr_stream_event_flush_ms ◇ A remotely tuned batching interval is now range-checked so a bad value cannot break event flushing.v2.1.223 → v2.1.308 ◆in the build AND v2.1.223 tengu_ccr_stream_first_flush_ms ◇ CCR client can now delay the very first streamed event separately from later ones, via a new first-flush timing settingv2.1.273 → v2.1.308 ◆in the build AND v2.1.273 tengu_ccr_stream_flush_on_first_delta ◇ Remote sessions gain server-controlled options to send streamed output on its first piece and to acknowledge messages only once readv2.1.285 → v2.1.300 ○gone AND v2.1.301 tengu_cfc_in_product_permissions ◇ The in-product permissions path for Claude for Chrome is hard-disabled here.v2.1.152 → v2.1.250 ○gone AND v2.1.251 tengu_cheeky_toucan ◇ A switched-off change lets Claude Code recover kept messages from an earlier compaction marker when the usual one is missingnever mined ·never mined AND v2.1.301 tengu_chert_bezel ◇ Default false. When on, Claude Code adds the thinking-token-count-2026-05-13 beta header to API requests for models that support interleaved thinking, provided the API provider is first-party, anthropicAws or Foundry and experimental betas are not disabled (by CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS or a HIPAA restriction).v2.1.147 → v2.1.183 ○gone AND v2.1.185 tengu_cicada_nap_ms ◇ Claude Code changed how it decides on and runs the background loading it does when it startsv2.1.138 or earlier → v2.1.308 ◆in the build AND v2.1.290 tengu_cinder_plover ◇ The AskUserQuestion prompt now receives a lean-prompt flag that can suppress a guidance snippet.v2.1.152 → v2.1.308 ◆in the build AND v2.1.301 tengu_cinder_swift ◇ Closing the control stream while a question waits can now cancel the turn, but that path stays off.v2.1.223 → v2.1.308 ◆in the build AND v2.1.223 tengu_cinder_wren ◇ The instructions Claude gets for AskUserQuestion are now chosen based on which other tools are available in the sessionv2.1.216 → v2.1.308 ◆in the build AND v2.1.301 tengu_classifier_disabled_surfaces ◇ The spinner can show a short headline of what Claude is actually working on right now.v2.1.138 or earlier → v2.1.308 ◆in the build AND v2.1.227 tengu_classifier_summary_heuristic_emit ◇ Lightweight conversation summaries are now always generated instead of needing a flag.v2.1.138 or earlier → v2.1.224 ○gone AND v2.1.225 tengu_claudeai_plugins_sync_enabled ◇ Flags controlling claude.ai plugin and skill sync were renamed to claudeai-scoped namesnever mined ·never mined AND v2.1.283 tengu_claudeai_skills_sync_enabled ◇ Flags controlling claude.ai plugin and skill sync were renamed to claudeai-scoped namesnever mined ·never mined AND v2.1.277 tengu_claudeai_skills_sync_shared_round ◇ How often Claude Code resyncs skills from claude.ai is now set by the server, and sessions can optionally share a recent syncv2.1.282 → v2.1.308 ◆in the build AND v2.1.282 tengu_cloth_snorkel ◇ Artifact MCP enablement no longer falls back to a remote flag, defaulting on unless an env var overrides itv2.1.209 → v2.1.234 ○gone AND v2.1.235 tengu_cobalt_awning ◇ Remote sessions gain bookkeeping for the time of the last full sign-in check before model requests, off unless a server value is setv2.1.285 → v2.1.291 ○gone AND v2.1.285 tengu_cobalt_lantern ◇ Quick web setup stays off, now behind a named helper needing a flag and two entitlements.v2.1.138 or earlier → v2.1.247 ○gone AND v2.1.248 tengu_cobalt_plinth_bryony ◇ Claude Code now always tells the artifact system it supports five-class permission asks, instead of depending on a setting or remote switchv2.1.273 → v2.1.294 ○gone AND v2.1.295 tengu_cobalt_plinth_burnet ◇ Artifact comment and subscription operations can each be switched on separately from the server.v2.1.268 → v2.1.308 ◆in the build AND v2.1.251 tengu_cobalt_plinth_direct ◇ Artifact availability now hinges on one gate, and direct upload kicks in for create paths.v2.1.172 → v2.1.238 ○gone AND v2.1.239 tengu_cobalt_plinth_hazel ◇ Creating an artifact from a type inside a cloud session sits behind a second switch, off.v2.1.248 → v2.1.266 ○gone AND v2.1.242 tengu_cobalt_plinth_holly ◇ Artifacts gain full pin and unpin actions, wired end to end but off by default behind a gate with an env override.v2.1.259 → v2.1.284 ○gone AND v2.1.285 tengu_cobalt_plinth_larch ◇ You could start a private artifact from a published template, but two server switches keep it shut here.v2.1.236 → v2.1.308 ◆in the build AND v2.1.267 tengu_cobalt_plinth_linden ◇ An alternate way of serving artifact content exists but is unreachable in normal use.v2.1.251 → v2.1.308 ◆in the build AND v2.1.251 tengu_cobalt_plinth_madder ◇ Several artifact feature switches were renamed and re-gated, discarding any prior rollout.v2.1.242 → v2.1.308 ◆in the build AND v2.1.287 tengu_cobalt_plinth_orpine ◇ Built but off unless enabled remotely: a shortcut that answers some artifact edit requests with direct file edits instead of a model replyv2.1.288 → v2.1.308 ◆in the build AND v2.1.288 tengu_cobalt_plinth_putguard ◇ Default true. When on, an artifact publish upload PUT (the page or its thumbnail image) that returns a 2xx status without an x-goog-generation response header is treated as failed: it is reported as status 403 with the error upload intercepted: 2xx without x-goog-generation, on the assumption that a proxy answered instead of Google Cloud Storage. When off, any 2xx response counts as a successful upload.v2.1.193 → v2.1.238 ○gone AND v2.1.239 tengu_cobalt_plinth_quince ◇ Published Artifacts get per-block anchor ids so links can point at a specific block.v2.1.236 → v2.1.238 ○gone AND v2.1.239 tengu_cobalt_plinth_reader_persist ◇ A prompt-and-context experiment switch was renamed and four other switches disappeared entirely.v2.1.191 → v2.1.245 ○gone AND v2.1.246 tengu_cobalt_plinth_rowan ◇ Claude could browse published artifact types to start from, but both actions are switched off.v2.1.248 → v2.1.266 ○gone AND v2.1.242 tengu_cobalt_plinth_sloe ◇ The Artifact tool's instructions now come in versions v0, v1 and v2, picked once per session by an env var, the session origin or a flagv2.1.281 → v2.1.295 ○gone AND v2.1.281 tengu_cobalt_thistle ◇ Artifact syntax highlighting and mermaid diagrams are now always on; three other gates were renamed and stay off.v2.1.211 → v2.1.238 ○gone AND v2.1.239 tengu_concurrent_candle ◇ A new gated 'start kit' feature can auto-bundle an artifact type, design system, and REPL toolv2.1.274 → v2.1.303 ○gone AND v2.1.274 tengu_copper_fox ◇ Forked subagents are on by default in interactive sessions; set CLAUDE_CODE_FORK_SUBAGENT to false to opt out.v2.1.138 or earlier → v2.1.231 ○gone AND v2.1.232 tengu_copper_kite ◇ With no API key, Claude Code can now fall back to a bearer token from the environment when fetching policy limits and remote settingsnever mined ·never mined AND v2.1.285 tengu_cozy_pixel ◇ A new fork_conversation control request lets a host app branch a session's history into a brand-new sessionv2.1.268 → v2.1.276 ○gone AND v2.1.277 tengu_cozy_teapot ◇ A remotely gated A/B test can loosen the bash-first system prompt wording.v2.1.271 → v2.1.306 ○gone AND v2.1.307 tengu_curious_river ◇ In Anthropic cloud sessions, a running MCP task stops keeping its cloud container up after a set number of hours, 168 by defaultv2.1.305 → v2.1.308 ◆in the build AND v2.1.305 tengu_curious_tower_stateless_models ◇ New gated 'tether' system can thread API requests together server-side instead of resending full context each turnv2.1.265 → v2.1.308 ◆in the build AND v2.1.265 tengu_dapper_lagoon ◇ An environment variable overrides whether something counts as first-party.v2.1.251 → v2.1.276 ○gone AND v2.1.251 tengu_dapper_truffle ◇ Auto mode's check of allowed MCP write tools now accepts rules from a command and adds a server-side decision path that is off by defaultv2.1.299 → v2.1.308 ◆in the build AND v2.1.299 tengu_dazzling_floyd ◇ You can type /review instead of /code-review, and pass a PR number as the target.v2.1.223 → v2.1.241 ○gone AND v2.1.246 tengu_deep_starlight ◇ Whole-PDF reads can skip a page-image render that only fed telemetry, and PDF renders can now be cancelledv2.1.281 → v2.1.306 ○gone AND v2.1.307 tengu_delegated_quail ◇ A background model-catalog comparison check now always has network access, though the feature stays off by defaultv2.1.251 → v2.1.308 ◆in the build AND v2.1.283 tengu_disable_bypass_permissions_mode ◇ Default false. When on, it disables bypassPermissions mode the same way the setting permissions.disableBypassPermissionsMode: "disable" does: startup mode selection skips bypassPermissions with the notice Bypass permissions mode was disabled by your organization policy, and the permission context marks bypass mode unavailable. A session already in bypass mode is also checked asynchronously against this gate; in the interactive UI it is switched to default mode, and on the other startup path Claude Code logs a warning and shuts down with exit code 1 and reason bypass_permissions_disabled.v2.1.138 or earlier → v2.1.233 ○gone AND v2.1.234 tengu_drifting_reef ◇ A new switch, off by default, makes computer-use actions in auto mode go through further permission checks even when an allow rule covers the whole toolv2.1.284 → v2.1.308 ◆in the build AND v2.1.284 tengu_enchanted_willow ◇ Mid-turn cost and token updates for print and SDK sessions now only run if a server-provided interval is set; with none, they are skippedv2.1.291 → v2.1.294 ○gone AND v2.1.291 tengu_fancy_crab ◇ Account memory tool inputs and results can be replaced with a placeholder, and remote sessions now fetch memory at the start and end of a turnv2.1.285 → v2.1.308 ◆in the build AND v2.1.285 tengu_feedback_survey_config ◇ The feedback survey checks for itself whether you're mid-prompt, so it interrupts you less.v2.1.162 → v2.1.308 ◆in the build AND v2.1.247 tengu_fennel_godwit ◇ On Opus 5, Claude is told not to spawn subagents, workflows or deep research unless you ask.v2.1.219 → v2.1.301 ○gone AND v2.1.219 tengu_fizzy_minsky ◇ In the desktop app and on claude.ai, a browser request with no extension connected can get a fixed setup reply instead of a model answerv2.1.300 ○gone AND v2.1.301 tengu_fleetview_simple ◇ The rule for turning on fleet view's simple mode is unchanged, but the way that code looks up keyboard shortcuts changedv2.1.216 → v2.1.308 ◆in the build AND v2.1.299 tengu_flint_harbor_prompt ◇ Default {}. Overrides for the /team-onboarding command: a string prompt replaces the built-in prompt template, a string guideTemplate replaces the text inserted at {{GUIDE_TEMPLATE}}, and a number windowDays sets how many days of usage data are scanned, floored and clamped to 1 through 365 (default 30). Fields of the wrong type use the built-in values.v2.1.138 or earlier → v2.1.308 ◆in the build AND v2.1.248 tengu_foamy_sedgewick ◇ A new instruction tells Claude to treat downloads and unpacked archives as untrusted and keep them apart from its own scriptsnever mined ·never mined AND v2.1.290 tengu_frame_publish_context ◇ Frame publish context is now permanently off and no server setting can enable it.v2.1.196 → v2.1.247 ○gone AND v2.1.248 tengu_gault_kestrel ◇ Default false. When on, the system prompt's guidance on hard-to-reverse actions ends its "Before deleting or overwriting, look at the target" sentence there, dropping the follow-on "If what you find contradicts how it was described, or you didn't create it, surface that instead of proceeding". The shorter wording is also used when CLAUDE_CODE_GAULT_KESTREL is set, when the Opus 5 prompt bundle applies (unless tengu_fennel_godwit is on), or when the client data key of this name is true.v2.1.257 → v2.1.259 ○gone AND v2.1.215 tengu_gb_eval_authed_enable ◇ Feature-flag requests can now attach auth asynchronously, but that path stays off in this build.v2.1.212 → v2.1.226 ○gone AND v2.1.227
Feedback