Feature flags, all 2,081.

tengu_left_arrow_editing_guard ◇ In vim normal mode, pressing left arrow on an empty prompt now works as expected.v2.1.218 → v2.1.308 ◆in the build AND v2.1.257 tengu_level_wren ◇ Attachment uploads now have separate remote settings for preview and scaled copies, and Claude is told when an image went out scaled downv2.1.287 → v2.1.308 ◆in the build AND v2.1.292 tengu_lexical_crystal ◇ No description from anybody.v2.1.287 → v2.1.308 ◆in the build AND v2.1.308 tengu_lilac_loom ◇ Default null. A promotional "feature of the week" campaign payload: an object with feature, optional command, startsAt and endsAt (ISO 8601 with a timezone), and optional display fields such as titleLabel, tips, announcementLines and creditless. A payload that fails validation is logged and ignored, and the campaign is active only between startsAt and endsAt.v2.1.242 → v2.1.308 ◆in the build AND v2.1.308 tengu_linear_brook ◇ A cloud session that failed to start now shows the actual reason instead of a generic error.v2.1.251 → v2.1.308 ◆in the build AND v2.1.251 tengu_linear_fog ◇ The focus view tip now stays hidden when spinnerTipsEnabled is false and is shown according to a remote experimentv2.1.305 → v2.1.308 ◆in the build AND v2.1.305 tengu_linear_owl ◇ Apps embedding Claude Code can supply the wording of a permission prompt from a PreToolUse hook, off unless switched on remotelyv2.1.299 → v2.1.308 ◆in the build AND v2.1.299 tengu_linen_orbit ◇ The memory-tools switch was renamed, resetting any rollout, so everyone is back on file-based memory.v2.1.238 → v2.1.308 ◆in the build AND v2.1.281 tengu_linked_cake ◇ When finding autoMemoryDirectory, Claude Code now skips settings sources that fail a new check, so they cannot move the memory folderv2.1.295 → v2.1.308 ◆in the build AND v2.1.295 tengu_linked_clover ◇ Workflow agents can now pause before spawning until an auto-continue-at-usage-limit condition clearsv2.1.271 → v2.1.308 ◆in the build AND v2.1.271 tengu_loggia_passage ◇ When Claude Code moves to another model after a refusal, it can now carry the original effort level over to that modelv2.1.287 → v2.1.308 ◆in the build AND v2.1.287 tengu_loggia_roster ◇ The server can list models that should be routed differently when a refusal fallback happens.v2.1.227 → v2.1.306 ○gone AND v2.1.227 tengu_long_context_survey_probability ◇ Default "", which means 0.2. Chance that the long-context feedback survey opens once its token threshold is crossed: the value goes through Number, an empty or non-finite value uses 0.2, and the result is clamped to 0 through 1. The roll is made once per survey component, so a failed roll is not retried.v2.1.196 → v2.1.308 ◆in the build AND v2.1.308 tengu_long_context_survey_question_variant ◇ Default "". Selects the long-context survey question: instruction_following asks How well is Claude following the instructions you gave earlier in this conversation? (optional), and any other value uses the generic survey question.v2.1.191 → v2.1.308 ◆in the build AND v2.1.308 tengu_long_context_survey_threshold ◇ Default "". Token count at which the long-context feedback survey may appear; the value must be all digits and greater than 0, and anything else leaves the threshold unset so the survey never triggers. The count it is compared with is chosen by tengu_long_context_survey_trigger_mode.v2.1.191 → v2.1.308 ◆in the build AND v2.1.308 tengu_long_context_survey_trigger_mode ◇ Default "". Which token count is compared with tengu_long_context_survey_threshold: virtual_context uses a context-token count computed from the current transcript, and any other value (cumulative) uses the session's cumulative token count.v2.1.196 → v2.1.308 ◆in the build AND v2.1.308 tengu_lucid_anchor ◇ Adding tools mid-conversation can still force a session reset, gated by an unread flagv2.1.268 → v2.1.308 ◆in the build AND v2.1.269 tengu_lucky_cerf ◇ New guidance nudges Claude to weigh the real cost before spawning a subagentv2.1.271 → v2.1.306 ○gone AND v2.1.307 tengu_lucky_cerf_text ◇ New guidance nudges Claude to weigh the real cost before spawning a subagentv2.1.271 → v2.1.304 ○gone AND v2.1.271 tengu_lucky_quill ◇ Two new internal feature gates were added: tengu_groovy_eclipse and tengu_lucky_quillv2.1.267 → v2.1.308 ◆in the build AND v2.1.267 tengu_lucky_whisper ◇ Two remote switches, off by default, control keeping a file-sending tool's results in subagents and how unanswered tool requests are foundv2.1.288 → v2.1.308 ◆in the build AND v2.1.300 tengu_luminous_whistle ◇ Reasoning text from a non-Claude model can survive a model switch, off unless enabled.v2.1.251 → v2.1.278 ○gone AND v2.1.308 tengu_madrone_spindle_newel ◇ Claude now previews whether it would be allowed to auto-reply to a new artifact comment while it's still busyv2.1.265 → v2.1.308 ◆in the build AND v2.1.265 tengu_malort_pedway ◇ Screenshots could be scaled and zoomed to a region, but both config tables switch it off.never mined ·never mined AND v2.1.242 tengu_maple_rung ◇ A nudge's eligibility check was rewritten around a different allowed-state helper, still gated by the same flag.v2.1.211 → v2.1.308 ◆in the build AND v2.1.271 tengu_marble_finch ◇ Auto mode's action checker can use a stricter wording covering connected apps and accounts; CLAUDE_CODE_AUTO_MODE_CANDIDATE_WORDING selects itv2.1.284 → v2.1.285 ○gone AND v2.1.284 tengu_marble_heron ◇ Your session can now wait out a usage limit and pick the task back up by itself.v2.1.234 → v2.1.308 ◆in the build AND v2.1.307 tengu_marlin_porch ◇ The check that enables scroll-region terminal rendering (DECSTBM) was rewired to new guard functionsv2.1.138 or earlier → v2.1.308 ◆in the build AND v2.1.269 tengu_max_version_config ◇ Claude Code now reads a pinned version from remote configuration and uses it only when it is newer than the version you are runningnever mined ·never mined AND v2.1.287 tengu_mcp_claudeai_eligibility_gate ◇ Default false. When on, a claude.ai connector (claudeai-proxy MCP server) whose config has eligible: false is not connected and is reported as needs-auth.v2.1.212 → v2.1.308 ◆in the build AND v2.1.308 tengu_mcp_connect_suspend_redial ◇ If an MCP server connection fails because the computer slept or its clock jumped, Claude Code says so and tries connecting again a limited number of timesv2.1.307 → v2.1.308 ◆in the build AND v2.1.307 tengu_mcp_connect_timeout_retry ◇ MCP connect timeouts now respect the global timeout and get retried instead of failing.v2.1.227 → v2.1.308 ◆in the build AND v2.1.307 tengu_mcp_discover_projection_prior ◇ A new gate resolves MCP discovery projection to legacy-only, all or off, and defaults to off.v2.1.259 → v2.1.308 ◆in the build AND v2.1.259 tengu_mcp_discovery_cache ◇ The /mcp view now counts cached servers separately so you can tell them from live connections.v2.1.221 → v2.1.237 ○gone AND v2.1.238 tengu_mcp_discovery_cache_enable ◇ Calling a tool on a disabled MCP server would tell you to re-enable it, but that path is off.v2.1.238 → v2.1.308 ◆in the build AND v2.1.238 tengu_mcp_drop_invalid_tool_schemas ◇ Default [], a list of hostnames. For an MCP server whose url hostname equals or is a subdomain of a listed entry (or for every server when the list contains ""), a tool whose input schema fails Claude Code's API-compatibility check is skipped and logged as Skipping tool "...": its input schema would be rejected by the Anthropic API. For other servers the tool is kept and only a warning is logged that requests including it may fail; servers without a string url only match "".v2.1.259 ○gone AND v2.1.216 tengu_mcp_issuer_strict_echo ◇ MCP OAuth now checks a server's echoed issuer, but only records mismatches rather than blocking.v2.1.224 → v2.1.308 ◆in the build AND v2.1.224 tengu_mcp_legacy_sse_fallback ◇ MCP servers using newer HTTP now automatically retry with an older HTTP+SSE connection method if the newer one failsv2.1.265 → v2.1.308 ◆in the build AND v2.1.307 tengu_mcp_listen_reopen_park ◇ A flaky MCP server that keeps dropping event streams now gets a pause instead of endless instant reconnects.v2.1.233 → v2.1.308 ◆in the build AND v2.1.233 tengu_mcp_listen_reopen_park_tuning ◇ A flaky MCP server that keeps dropping event streams now gets a pause instead of endless instant reconnects.v2.1.233 → v2.1.308 ◆in the build AND v2.1.233 tengu_mcp_local_oauth_blocked_hosts ◇ Default {"hosts": ["microsoft365.mcp.claude.com", "gmail.mcp.claude.com", "gcal.mcp.claude.com"]}. An http or sse MCP server whose URL host (lowercased, trailing dot removed) is in hosts is classified as Anthropic-hosted for authentication, with a message that it doesn't support local OAuth and should be connected under Settings, Connectors on claude.ai. A missing or non-array hosts, or one with no string entries, uses the default list.v2.1.162 → v2.1.308 ◆in the build AND v2.1.308 tengu_mcp_negotiation_server_denylist ◇ Remote config can force legacy MCP handshakes or suppress listening for named servers, matched by hostname.v2.1.259 ○gone AND v2.1.308 tengu_mcp_normalize_root_combinators ◇ Default [], a list of hostnames. For an MCP server whose url hostname equals or is a subdomain of a listed entry (or for every server when the list contains ""), a tool whose input schema uses top-level combinators is kept with a flattened schema and a note prepended to its description. For other servers such a tool is skipped with Skipping tool "...": its input schema uses top-level ..., which the Anthropic API does not accept; servers without a string url only match "".v2.1.195 → v2.1.259 ○gone AND v2.1.216 tengu_mcp_proxy_needs_approval_retry ◇ Retrying MCP calls that need approval now uses one shared rule, so behaviour is consistent.v2.1.210 → v2.1.308 ◆in the build AND v2.1.295 tengu_mcp_server_policy_bypass_exempt ◇ The rule that exempts MCP server policy denials in remote bypass-permissions sessions behaves the same as beforev2.1.196 → v2.1.308 ◆in the build AND v2.1.288 tengu_mcp_skills ◇ Prompt autocomplete no longer suggests MCP server URL templates for input like /server:...://v2.1.161 → v2.1.303 ○gone AND v2.1.282 tengu_mcp_startup_policy_seed ◇ Remote sessions can now seed MCP startup policy state behind a new gate.v2.1.191 → v2.1.308 ◆in the build AND v2.1.290 tengu_mcp_stateless_skip_init ◇ Default true. When on, a claude.ai connector (claudeai-proxy) with stateless: true connects statelessly: MCP initialize is answered from its cached init response when that parses and carries a supported protocol version (otherwise a real initialize runs), and when tengu_mcp_discover_projection_prior is not off, server/discover can be resolved from a cached projection. It also turns on the anthropic-mcp-client-capabilities and MCP-Protocol-Version request headers even when tengu_mcp_discover_projection_prior is off.v2.1.169 → v2.1.308 ◆in the build AND v2.1.308 tengu_mcp_strip_trailing_xml_tags ◇ MCP result post-processing still uses the same flag to decide whether to strip trailing XML-like tags.v2.1.186 → v2.1.308 ◆in the build AND v2.1.299 tengu_mcp_task_worker_wake ◇ The wake-up and retry steps for background MCP tasks now depend on a new remote switch that is off by defaultv2.1.285 → v2.1.308 ◆in the build AND v2.1.299 tengu_mcp_tasks ◇ Claude Code now checks a remote switch for an MCP tasks feature that stays off unless the server turns it onv2.1.285 → v2.1.296 ○gone AND v2.1.299 tengu_mcp_url_elicitation ◇ MCP server configs accept a new bareElicitationCapability option that gives the server the plain elicitation capability instead of the legacy URL onev2.1.281 → v2.1.308 ◆in the build AND v2.1.300 tengu_media_byte_cap ◇ The size limit on images and media sent to the model now comes from new remote settings, with checks and a separate second limitv2.1.191 → v2.1.296 ○gone AND v2.1.299 tengu_medlar_quoin ◇ Artifact reply posts can now declare which earlier reply they continuev2.1.235 → v2.1.272 ○gone AND v2.1.235 tengu_mellow_beacon ◇ A server-controlled setting can make some background requests to the model point back to the reply they came fromv2.1.288 → v2.1.308 ◆in the build AND v2.1.288 tengu_mellow_heron ◇ Claude Code can now track prompts that were merged or queued together, for rewinding and spotting unanswered turns, when a switch is onv2.1.299 → v2.1.308 ◆in the build AND v2.1.299 tengu_mellow_hollerith ◇ New built-in 'Claude Test' plugin runs plain-language specs against your local dev server in a real browser and reports PASS/FAILv2.1.274 → v2.1.295 ○gone AND v2.1.300 tengu_melodic_octopus ◇ When --worktree uses a pull or merge request number, Claude Code fetches the full PR reference and refuses if it can't tell which commit is meantv2.1.300 → v2.1.308 ◆in the build AND v2.1.302 tengu_melodic_wolf ◇ Subagent results can be tagged to show which parts Claude Code wrote versus the model.v2.1.242 → v2.1.308 ◆in the build AND v2.1.277 tengu_mem_push_delete_mode ◇ Default corroborate. How a memory store sync push treats a previously synced file that is now missing locally: immediate deletes it remotely on that push, never never deletes it (and skips the mass-delete hold check), and corroborate deletes it only after it has been missing on at least 2 walks at least 30000 ms apart. Other flag values mean corroborate; a set CLAUDE_CODE_MEMORY_PUSH_DELETE_MODE overrides the flag.v2.1.212 → v2.1.308 ◆in the build AND v2.1.308 tengu_memdir_write_lint ◇ A new check can refuse or clean memory files containing invisible characters or tag-like text; it falls back to off unless enabled remotelyv2.1.284 → v2.1.308 ◆in the build AND v2.1.284 tengu_memoized_turtle ◇ No description from anybody.v2.1.295 → v2.1.308 ◆in the build AND v2.1.308 tengu_memory_bulk_inflate ◇ Team memory bulk-inflate telemetry now only fires for non-account memory, plus a new bail-outv2.1.161 → v2.1.308 ◆in the build AND v2.1.275 tengu_memory_stream_list ◇ Default true. When on, memory store sync first tries to list a store's entries through the backend's metadata stream (exportMetadata), falling back to the paged list call when that fails. Off, or with CLAUDE_CODE_DISABLE_MEMORY_STREAM_LIST set, every listing uses the paged list.v2.1.210 → v2.1.308 ◆in the build AND v2.1.210 tengu_mermaid_mod ◇ Mermaid flowcharts and sequence diagrams in replies now render as text art in the terminalnever mined ·never mined AND v2.1.277 tengu_merry_popcorn ◇ Claude Code can now label each approval of a tool call as coming from a person, a classifier, a mode or a hookv2.1.288 → v2.1.308 ◆in the build AND v2.1.288 tengu_misty_aurora ◇ In remote sessions, a permission request left unanswered long enough can be declined automatically, with Claude told not to retryv2.1.292 → v2.1.308 ◆in the build AND v2.1.292 tengu_misty_sunbeam ◇ When a PDF has too many pages to read at once, Claude Code can read the first pages and report the total instead of failingv2.1.300 → v2.1.308 ◆in the build AND v2.1.300 tengu_modular_seal ◇ Claude Code can now count Claude's refusals since the conversation was last summarised, behind a setting that is off by defaultv2.1.295 → v2.1.308 ◆in the build AND v2.1.295 tengu_moonlit_island ◇ Claude Code can mark a round of polled events as declared and hand them to Claude in one message, but only behind a remote switchv2.1.293 → v2.1.308 ◆in the build AND v2.1.293 tengu_moonlit_lagoon ◇ A timing check used while Claude Code is idle or waiting on you now first consults a server-controlled switchv2.1.287 → v2.1.308 ◆in the build AND v2.1.287 tengu_moonlit_lobster ◇ Changing the model now refreshes the data Claude Code loaded at startup, as part of reworked model-switch handlingv2.1.285 → v2.1.287 ○gone AND v2.1.285 tengu_moonlit_toucan ◇ 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 v2.1.296 tengu_morning_brief_config ◇ Default null. An object whose skillPrompt string replaces the built-in prompt of the /morning skill when, after trimming, it is 500 to 50000 characters long. A valid CLAUDE_CODE_MORNING_BRIEF_PROMPT takes precedence, and the flag is not read when the entrypoint is remote_cowork or remote_trigger.v2.1.207 ○gone AND v2.1.207 tengu_morning_brief_skill ◇ Default false. Enables the /morning skill (render a morning brief as an HTML artifact, or set it up as a recurring weekday task) when CLAUDE_CODE_IS_COWORK is true and the entrypoint is not remote_cowork or remote_trigger. A true CLAUDE_CODE_ENABLE_MORNING_BRIEF enables the skill in those sessions and entrypoints regardless of this flag.v2.1.207 ○gone AND v2.1.207 tengu_mossy_crown ◇ A new accessFallbackModels setting lists models Claude Code may fall back to when the main model can't be usedv2.1.292 → v2.1.308 ◆in the build AND v2.1.299 tengu_mossy_gray ◇ A limit set at 2000 for every provider can now be removed by a server-side switch when Claude Code runs interactivelyv2.1.299 → v2.1.308 ◆in the build AND v2.1.299 tengu_mossy_lantern ◇ Tools added or removed mid-conversation can be surfaced to the model, but the beta capability is gated off by default.v2.1.257 → v2.1.268 ○gone AND v2.1.308 tengu_mossy_thrush ◇ When a further server-side switch is on, the changed-file check can judge files through an outputs link and count each outcomev2.1.300 → v2.1.308 ◆in the build AND v2.1.300 tengu_moth_lantern ◇ Default false. When on and CLAUDE_MEMORY_STORES is unset or blank, a saved org memory store selection for the current account (orgMemorySelection) is sent when Claude Code requests the org memory credential, and cached store-picker data is returned alongside the org memory mounts. When off, a saved selection is sent only if it is public_projects.v2.1.213 → v2.1.217 ○gone AND v2.1.213 tengu_mutable_teapot ◇ Claude Code can now remove ask rules that come from MCP server policy before sending your permission rules to the auto mode classifiernever mined ·never mined AND v2.1.282 tengu_nankeen_kestrel ◇ Windows sandbox TLS interception could use a persistent managed certificate authority, but that path is disabled here.v2.1.212 → v2.1.308 ◆in the build AND v2.1.221 tengu_native_cursor ◇ Default false. When true, Claude Code uses the terminal's native cursor, the same as setting CLAUDE_CODE_NATIVE_CURSOR, provided the DECSTBM renderer is not in use; the result is cached for the process. CLAUDE_CODE_ACCESSIBILITY and one other internal condition turn the native cursor on without this flag. On Windows the same switches also feed a second terminal setup check that does not look at the DECSTBM renderer.v2.1.163 → v2.1.308 ◆in the build AND v2.1.308 tengu_neapolitan ◇ The checks deciding whether subagents are offered remote isolation gained two new conditions at the frontv2.1.178 → v2.1.308 ◆in the build AND v2.1.288 tengu_nifty_finch ◇ Claude Code can now work out an alternate model to suggest from the current one, behind a server switchv2.1.286 → v2.1.308 ◆in the build AND v2.1.286 tengu_nifty_lemur ◇ Five analytics events cover the weekly limit-reset flow.v2.1.248 → v2.1.308 ◆in the build AND v2.1.308 tengu_nimble_wren ◇ The built-in web-fetch helper now uses Haiku when certain model and provider checks pass, and is told not to retry sites that refuse itv2.1.300 → v2.1.308 ◆in the build AND v2.1.302 tengu_noble_mountain ◇ Write tool now accepts file_text, file_content, and path as aliases for content and file_path, rewriting input before plugin hooks see itv2.1.280 → v2.1.308 ◆in the build AND v2.1.280 tengu_noble_snowflake ◇ A new telemetry event, tengu_bridge_transcript_deferral, reports each stage of Remote Control's deferred transcript uploadv2.1.281 → v2.1.308 ◆in the build AND v2.1.281 tengu_noble_wozniak ◇ When the agents view opens, background housekeeping can now wait until you have trusted the folder instead of starting straight awayv2.1.300 → v2.1.308 ◆in the build AND v2.1.300 tengu_non_deferrable_builtins ◇ Default null. A list of tool names that tool search does not defer, so they are loaded with the request. The value may be a plain array or an object keyed by model name substring with a * fallback; it is merged with a non_deferrable_builtins list from another config source, and its entries also feed a set of MCP server names alongside the always-allow rules.v2.1.176 → v2.1.308 ◆in the build AND v2.1.308 tengu_observer_agents_enabled ◇ Default true. Kill switch for observer agents, the observer field in an agent definition: they also require CLAUDE_CODE_EXPERIMENTAL_OBSERVER_AGENTS, and a false value here leaves every agent unobserved. The same gate also switches a second code path elsewhere in the bundle.v2.1.200 → v2.1.308 ◆in the build AND v2.1.200 tengu_observer_subagent_fanout ◇ Default true. When false, an observer pairing is not extended to the subagents an observed agent spawns. When on, fan-out still requires observer agents to be enabled, stops at depth 2, and is skipped for agents with observeSubagents: false.v2.1.218 → v2.1.308 ◆in the build AND v2.1.218 tengu_onyx_plover ◇ Default null. Configures auto dream memory consolidation: enabled: true turns it on when the autoDreamEnabled setting is unset, enabled or available being true makes the feature available, and minHours (default 24) and minSessions (default 5) set how long and how many sessions must pass since the last consolidation before another runs. Non-positive or non-numeric values use the defaults.v2.1.138 or earlier → v2.1.308 ◆in the build AND v2.1.94 tengu_orchid_mantis ◇ Default false. When on and tengu_orchid_mantis_v2 is off, the system prompt's session guidance tells the model it may end a reply with a one-line offer to /schedule a background agent for a natural future follow-up, with signals for one-time and recurring offers; this needs the Skill tool and a schedule skill to be available. Separately, it switches the /schedule skill's when-to-use text to one that also covers one-time runs such as "open a cleanup PR for X in 2 weeks".v2.1.138 or earlier → v2.1.205 ○gone AND v2.1.272 tengu_orchid_mantis_v2 ◇ Default false. When on, and the Skill tool and a schedule skill are available, the system prompt's session guidance carries a stricter /schedule rule: no offer by default, and an offer only when the turn left a named artifact with a quotable future obligation (a flag with a ramp date, a .skip with a removal condition, a job ID with an ETA, a dated TODO), at most once per session. In that guidance it takes precedence over tengu_orchid_mantis.v2.1.138 or earlier → v2.1.205 ○gone AND v2.1.272 tengu_orderly_lantern ◇ WebFetch URL approval prompts now remember approved addresses and can retry without asking, and a switch can make them appear one at a timev2.1.292 → v2.1.308 ◆in the build AND v2.1.292 tengu_orford_ness ◇ Default false. Turns on sending error reports (logged errors, unhandled rejections and uncaught exceptions) to Datadog error tracking, with a per-process report cap. It is checked last: reporting also needs DISABLE_ERROR_REPORTING unset, the first-party API provider, a Claude Code version of at least 2.1.193, the allow_error_reporting policy check passing, and other internal checks.v2.1.191 → v2.1.274 ○gone AND v2.1.308 tengu_org_config_required ◇ No description from anybody.v2.1.293 → v2.1.308 ◆in the build AND v2.1.308 tengu_org_policy_denied ◇ Blocked remote-session actions from org policy now emit a tengu_org_policy_denied telemetry eventnever mined ·never mined AND v2.1.296