Feature flags, all 2,095.

tengu_sparrow_ledger ◇ Default false. When on, adds a Work step by step: section to the system prompt telling the model to reproduce the symptom before editing, edit the source, then re-observe the symptom until it is gone. The CLAUDE_CODE_VERIFY_PROMPT environment variable turns on the same section.v2.1.138 or earlier → v2.1.205 ○gone AND v2.1.308 tengu_spicy_pearl ◇ No description from anybody.v2.1.299 → v2.1.309 ◆in the build AND v2.1.308 tengu_sprightly_lagoon ◇ Default false. When on, the checks for per-turn effort and per-turn timing both report unsupported, so requests carry no per-turn effort or now values and those checks add no per-turn-control-2026-07-01 beta.v2.1.260 → v2.1.301 ○gone AND v2.1.308 tengu_starry_locket ◇ Two server switches, off unless turned on remotely, can hide or skip MCP servers from claude.ai-synced plugins that a terminal cannot sign in tov2.1.283 → v2.1.309 ◆in the build AND v2.1.283 tengu_steady_marshmallow ◇ Identity changes mid-session can now defer a policy reload instead of immediately reloading as an account changenever mined ·never mined AND v2.1.274 tengu_steady_sunbeam ◇ 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_stone_shell ◇ Rewritten memory rules would make Claude save corrections and environment quirks immediately, before replying.v2.1.212 → v2.1.285 ○gone AND v2.1.286 tengu_structured_output_strict ◇ Tool descriptions can differ by session proactive level; the streaming behaviour behind it is off.v2.1.203 → v2.1.309 ◆in the build AND v2.1.301 tengu_subagent_cache_evict ◇ Default false. When on, model requests marked evictCacheOnComplete that carry sticky betas add the prompt-caching-evict-2026-05-12 beta. Provider checks come first: the request must go to the first-party API at a first-party base URL (or Claude Platform on AWS with no ANTHROPIC_AWS_BASE_URL), with neither CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS set nor HIPAA mode active, and under those conditions a truthy CLAUDE_CODE_SUBAGENT_CACHE_EVICT turns it on without consulting the flag.v2.1.198 → v2.1.309 ◆in the build AND v2.1.308 tengu_synchronous_bunny ◇ Claude Code no longer checks for the GitHub app when it is connected through a provider other than Anthropic's own servicev2.1.286 → v2.1.306 ○gone AND v2.1.285 tengu_tab_read_sep ◇ The tab-aware read separator flag is read fresh each time but still off.v2.1.138 or earlier → v2.1.309 ◆in the build AND v2.1.248 tengu_tallow_quoit_gate ◇ No description from anybody.v2.1.307 → v2.1.309 ◆in the build AND v2.1.308 tengu_teal_corbel_newel ◇ A new unclear flag likely affects auto-allow rules for artifact reads.v2.1.260 → v2.1.309 ◆in the build AND v2.1.273 tengu_team_discovery ◇ Default false. When on, the allow_team_discovery policy permits it, and the login is a claude.ai subscription (OAuth with the user:inference scope), the context-tips classifier fetches /api/claude_code/discovery/team_usage (the skills and MCP servers used across the team, with user counts), caches the result for 24 hours in cache/team-discovery.json under the config directory, and uses those lists. A 403 or an empty result yields no team data, and that empty result is cached as well.v2.1.191 → v2.1.211 ○gone AND v2.1.191 tengu_teleport_send_to_cloud ◇ A menu to send your session to the cloud exists but stays hidden unless the server enables it.v2.1.229 → v2.1.309 ◆in the build AND v2.1.247 tengu_thinking_block_resumption ◇ Resuming an interrupted thinking block now accepts model aliases, but the feature stays off.v2.1.211 → v2.1.309 ◆in the build AND v2.1.271 tengu_thinking_display_updates ◇ Claude Code can present thinking as connector text only, but nothing enables it here.v2.1.240 → v2.1.259 ○gone AND v2.1.240 tengu_ticklish_whisper ◇ Auto mode can show a prompt that denies itself after two minutes when you are away; off by default.v2.1.246 → v2.1.309 ◆in the build AND v2.1.246 tengu_tidy_axolotl ◇ Claude Code reads a server-controlled switch whose built-in value is on; what it controls is not knownv2.1.292 → v2.1.299 ○gone AND v2.1.300 tengu_tool_memory_cgroup ◇ On Linux you can cap memory for shell commands so a runaway process cannot eat your machine.v2.1.233 → v2.1.245 ○gone AND v2.1.295 tengu_tool_pear ◇ A new capability gate tied to the opus_5_5_prompt_bundle flag can unlock certain model features, and a tool_pear beta header now depends on a flag not yet read for this accountv2.1.138 or earlier → v2.1.309 ◆in the build AND v2.1.280 tengu_trace_lantern ◇ Detailed beta tracing now also requires that the session is not diskless, so it stops in sessions that avoid writing to diskv2.1.138 or earlier → v2.1.309 ◆in the build AND v2.1.295 tengu_tranquil_crescent ◇ Claude Code adds three new internal feature gates: typed_koala, zinc_harbor, tranquil_crescentnever mined ·never mined AND v2.1.275 tengu_tussock_oriole ◇ Default false. When on, the team-artifacts tip can show New from your team: ..., listing project skills and commands by teammates that the user has not seen yet, and skill names from project settings can show a · by <author> attribution. Off, neither appears.v2.1.138 or earlier → v2.1.309 ◆in the build AND v2.1.117 tengu_twinkling_globe ◇ New remote gate tengu_twinkling_globe controls whether Claude Code treats account policy limits as absent instead of fetching themnever mined ·never mined AND v2.1.308 tengu_twinkly_prism ◇ Two server switches, off unless turned on remotely, can hide or skip MCP servers from claude.ai-synced plugins that a terminal cannot sign in tov2.1.283 → v2.1.309 ◆in the build AND v2.1.283 tengu_umber_petrel ◇ Default false. When on, auto memory is turned off for sessions whose model matches an entry in the tengu_sepia_cormorant model list, unless auto memory has been explicitly forced on or off.v2.1.138 or earlier → v2.1.309 ◆in the build AND v2.1.308 tengu_umber_sluice ◇ The rule stopping Claude publishing artifacts you didn't ask for can now be removed remotely.v2.1.247 → v2.1.309 ◆in the build AND v2.1.247 tengu_valiant_crescent ◇ With CLAUDE_AUTO_BACKGROUND_TASKS set, running foreground subagents move to the background when you have a queued message waitingv2.1.290 → v2.1.309 ◆in the build AND v2.1.290 tengu_validated_clover ◇ A new internal check called tengu_validated_clover appeared in Claude Code's code, but what it does isn't known yetnever mined ·never mined AND v2.1.308 tengu_vast_tulip ◇ New tool-host attestation floor check replaces old attestation execution logic, gated behind two flagsv2.1.277 → v2.1.309 ◆in the build AND v2.1.277 tengu_vellum_anchor ◇ The out-of-credits message can add a "progress saved" note, off unless a server enables it.v2.1.229 → v2.1.309 ◆in the build AND v2.1.236 tengu_vellum_lantern ◇ Default false. Turns on the alternate prompt mode that CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT controls, for eligible models (claude-opus-4-7, model IDs ending in -eap, or models listed in the simple_system_prompt client data or the tengu_velvet_cascade flag). -eap and listed models get the mode whether or not this flag is on, and a truthy or explicitly falsy CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT overrides both.v2.1.138 or earlier → v2.1.153 ○gone AND v2.1.120 tengu_vellum_siding ◇ When the product-feedback policy changes, Claude Code now re-sends every feature switch to the VS Code extension, not just two of themv2.1.143 → v2.1.309 ◆in the build AND v2.1.284 tengu_velvet_gannet ◇ Claude Code has a new check for resumed interrupted turns where it cannot ask you for permission, behind a switch that is off by defaultv2.1.295 → v2.1.309 ◆in the build AND v2.1.305 tengu_velvet_hammer ◇ Default false. When true, or when the per-model variant tengu_velvet_hammer_<bucket> is true for the current model (the bucket is a model family such as opus, sonnet or haiku; the base bucket uses the plain name), the Edit tool skips its "File has not been read yet. Read it first before writing to it." rejection, so a file that was never read, or was only partially read, can be edited.v2.1.166 → v2.1.195 ○gone AND v2.1.308 tengu_velvet_mallet ◇ An alternate prompt wording for two tools exists but stays off unless you set an environment variable.v2.1.166 → v2.1.195 ○gone AND v2.1.228 tengu_velvet_tide ◇ Which models get the simple system prompt is now decided locally, not by remote config.v2.1.210 → v2.1.309 ◆in the build AND v2.1.227 tengu_velvety_ripple ◇ Messages from other sessions now come with notes telling the model they are not from you, or that some could not be shownv2.1.287 → v2.1.309 ◆in the build AND v2.1.287 tengu_violin_bridgepin ◇ New display styles for background tool results and for repository-trust folder-wait refusalsv2.1.277 → v2.1.309 ◆in the build AND v2.1.277 tengu_violin_duet ◇ Claude Code gains switches for a connected mode for cloud sessions and a file sync that depends on it, both off unless turned on remotelynever mined ·never mined AND v2.1.284 tengu_violin_guarneri ◇ Text telling Claude that Claude in Chrome needs no enabling has moved, and a new remote switch now sits where it wasv2.1.300 → v2.1.303 ○gone AND v2.1.300 tengu_violin_luthier ◇ A new --allow-unattended-tool-calls option lets cloud sessions use the current folder for one run, with auto-mode commands running without askingv2.1.290 → v2.1.309 ◆in the build AND v2.1.290 tengu_violin_maple ◇ Claude Code gains switches for a connected mode for cloud sessions and a file sync that depends on it, both off unless turned on remotelyv2.1.284 → v2.1.308 ○gone AND v2.1.284 tengu_violin_rosin ◇ A GetTask tool that reports the state of a background shell command is built in but off unless switched on remotelyv2.1.283 → v2.1.309 ◆in the build AND v2.1.283 tengu_violin_scroll ◇ Claude Code adds an unreleased one-shot 'reannounce' request for remote tools, off by defaultv2.1.280 → v2.1.309 ◆in the build AND v2.1.280 tengu_violin_wood ◇ If you have too many untracked files, cloud seeding refuses and tells you which budget you blew.v2.1.227 → v2.1.309 ◆in the build AND v2.1.295 tengu_vscode_feedback_survey ◇ Default false. When on, the session's SDK initialize response carries a feedback_survey_config (the tengu_feedback_survey_config value, with its probability overridden by the feedbackSurveyRate setting when set, plus the last time a survey was shown) for VS Code extension sessions and certain other entrypoints. No config is sent when CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY is set or the allow_product_feedback policy is off.v2.1.162 → v2.1.309 ◆in the build AND v2.1.308 tengu_vscode_review_upsell ◇ A resume-precheck gate is now passed to the VS Code extension, on by default.v2.1.138 or earlier → v2.1.309 ◆in the build AND v2.1.284 tengu_walnut_prism ◇ Default false. When on (or when CLAUDE_CODE_OWNERSHIP_FRAME is true), system prompt wording shifts to ownership framing in the prompt variants that carry these sections: the opening line becomes You work alongside the user on software engineering tasks and own the outcome of what you take on. (or an output-style variant with the same framing), the confirm-before-risky-actions sentence drops approval in one context doesn't extend to the next, and the When a task has been agreed, the approval covers it end to end paragraph is left out.v2.1.161 → v2.1.215 ○gone AND v2.1.308 tengu_walnut_spire ◇ Plugin eval checks your credentials first and aborts instead of failing every case.v2.1.198 → v2.1.268 ○gone AND v2.1.235 tengu_watchdog_skip_nonstreaming_fallback ◇ Default false. When on, a streaming request that the stream idle watchdog aborts is not retried as a non-streaming request. CLAUDE_CODE_REMOTE has the same effect, and CLAUDE_CODE_DISABLE_NONSTREAMING_FALLBACK or tengu_disable_streaming_to_non_streaming_fallback disable the fallback for every stream failure.v2.1.211 → v2.1.309 ◆in the build AND v2.1.232 tengu_webagent_mcp_bridge ◇ The hearthbot chat bridge can now auto-allow many more read-only actions without a permission promptv2.1.273 → v2.1.274 ○gone AND v2.1.273 tengu_wild_gosling ◇ A new remotely controlled switch is off by default and never applies to organisations under the HIPAA policyv2.1.290 → v2.1.309 ◆in the build AND v2.1.290 tengu_willow_crate ◇ Opening the diff panel records one event with your terminal width bucket.v2.1.228 → v2.1.259 ○gone AND v2.1.228 tengu_willow_tern ◇ A writing-style prompt block can be forced on with an environment variable; otherwise it stays off.v2.1.248 → v2.1.309 ◆in the build AND v2.1.247 tengu_windows_credman ◇ Several internal services now receive login credentials directly instead of grabbing whatever token is lying around.never mined ·never mined AND v2.1.269 tengu_wise_comet ◇ Manual compaction now decides whether to strip thinking blocks, gated for adaptive thinking only.v2.1.260 → v2.1.309 ◆in the build AND v2.1.260 tengu_wobbly_dahl ◇ Claude Code can tell the model the user pays per token and what a Bash call costs, but both notes fall back to offv2.1.290 → v2.1.309 ◆in the build AND v2.1.290 tengu_woven_harbor ◇ Cloud sessions from review or plan sources can use a newer creation endpoint allowing --project grouping.v2.1.227 → v2.1.241 ○gone AND v2.1.224 tengu_zinc_harbor ◇ Background tasks left running after their owning agent disappears are now automatically stoppedv2.1.275 → v2.1.309 ◆in the build AND v2.1.275