Environment variables, all 1,448.

CLAUDE_CODE_SYNTAX_HIGHLIGHT ◇ Set to 0, false, no or off to turn off syntax highlighting of code; the theme picker then shows Syntax highlighting disabled (via CLAUDE_CODE_SYNTAX_HIGHLIGHT=<value>). Any other value, or unset, leaves highlighting on.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_SYSTEM_PROMPT_GB_FEATURE ◇ Read only when CLAUDE_CODE_REMOTE is true. Names a feature flag whose value, if it is a non-empty string, is used as the session's custom system prompt in place of the one the session was given; a missing, empty or non-string flag value keeps the original.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_TAG_ISMETA_MESSAGES ◇ No description from anybody.v2.1.160 → v2.1.308 ◆in the build CLAUDE_CODE_TAGS ◇ Free-form string copied, when set, into the tags field of the environment context Claude Code attaches to its analytics event metadata. A function-hooks telemetry module also reads it into a tags field.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_TASK_LIST_ID ◇ Fixes the ID of the task list the task tools read and write; tasks are stored as JSON files under a tasks/<id>/ directory, with characters other than letters, digits, _ and - replaced by -. Without it the ID comes from the team or session context. When it is set, a forked session does not copy the parent's tasks.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_TEAM_ONBOARDING ◇ Accepts banner or step; other values are ignored. Either value runs the first-run onboarding flow at startup even when onboarding is already complete, and takes precedence over the tengu_cedar_inlet flag for how the "On a team?" prompt (ask a teammate to run /team-onboarding and paste the guide) appears afterwards: step as a screen at the end of the onboarding flow, banner as an info notice. The prompt stays off when the allow_team_onboarding policy disallows it or the login is a claude.ai Pro or Max subscription.v2.1.138 or earlier → v2.1.179 ○gone CLAUDE_CODE_TEAM_TEARDOWN_PARK_TIMEOUT_MS ◇ Five more environment variables, including stream and byte watchdogs, can be set in settings.json.v2.1.206 → v2.1.308 ◆in the build CLAUDE_CODE_TEE_SDK_STDOUT ◇ Boolean (1, true, yes or on). In a remote session that is not a bridge, every SDK message sent to the remote backend is also written as a JSON line to the pipe named by CLAUDE_RUNNER_ACTIVITY_FD, or to stdout when that is unset or unusable, along with session_state_changed and turn_starting system events. The self-hosted runner sets it to 1 for the sessions it launches.v2.1.139 → v2.1.308 ◆in the build CLAUDE_CODE_TERMINAL_MCP_TOOLS ◇ Comma-separated list of MCP tool names (entries trimmed, empty ones dropped). When the model ends a turn with no text after a successful call to one of these tools earlier in the same turn, the thinking-only-response retry nudge is skipped. When Claude Code condenses recent messages for its session status classifier, the string text input of these tools' calls counts as message text.v2.1.181 → v2.1.308 ◆in the build CLAUDE_CODE_TERMINAL_RECORDING ◇ No description from anybody.v2.1.160 → v2.1.308 ◆in the build CLAUDE_CODE_TEST_ALLOW_REAL_NETWORK ◇ A module that blocks network access during test runs ships complete but does nothing here.v2.1.239 → v2.1.308 ◆in the build CLAUDE_CODE_TEST_FIXTURES_ROOT ◇ No description from anybody.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_TEST_FORCE_DENY ◇ No description from anybody.v2.1.160 → v2.1.308 ◆in the build CLAUDE_CODE_TEST_NO_GIT_BASH ◇ No description from anybody.v2.1.160 → v2.1.308 ◆in the build CLAUDE_CODE_TEST_NO_PWSH ◇ No description from anybody.v2.1.160 → v2.1.308 ◆in the build CLAUDE_CODE_THINKING_DISPLAY_UPDATES ◇ Set CLAUDE_CODE_THINKING_DISPLAY_UPDATES to force the new thinking display on or off yourself.v2.1.240 → v2.1.308 ◆in the build CLAUDE_CODE_THISTLE_GREBE ◇ Subagent delegation nudges are now switched off for Opus 5; you can also control them by env var.v2.1.215 → v2.1.308 ◆in the build CLAUDE_CODE_THRIFTY_SONIC ◇ Auto mode can be told to do file reads, searches and edits through Bash, but it's off by default.v2.1.221 → v2.1.308 ◆in the build CLAUDE_CODE_TICKLISH_WHISPER ◇ Auto mode can show a prompt that denies itself after two minutes when you are away; off by default.never mined ·never mined CLAUDE_CODE_TICKLISH_WHISPER_TIMEOUT_MS ◇ Auto mode can show a prompt that denies itself after two minutes when you are away; off by default.never mined ·never mined CLAUDE_CODE_TMPDIR ◇ If the socket directory for session-to-session messaging looks unsafe, messaging is turned off with a fix hint.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_TMUX_PREFIX ◇ Set by Claude Code when it launches itself in a new tmux session: the global tmux prefix read from tmux show-options -g prefix, default C-b when that fails. Read back inside that session by the tmux session: notice, which tells you to detach with <prefix> d.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_TMUX_PREFIX_CONFLICTS ◇ Set by Claude Code when it launches itself in a new tmux session: 1 when the tmux prefix is one of C-b, C-c, C-d, C-t, C-o, C-r, C-s, C-g or C-e, otherwise empty. When true, the tmux session: notice tells you to press the prefix twice and then d to detach, because Claude uses that key.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_TMUX_SESSION ◇ When you exit a worktree session, removing the worktree no longer holds up the exit dialog and runs after the session endsv2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_TMUX_TRUECOLOR ◇ Any non-empty value keeps 24-bit color inside tmux. By default, when TMUX is set and the detected color level is truecolor, Claude Code lowers it to 256 colors.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_TOASTY_THIMBLE ◇ A short reminder can be slipped in after tool results, set by an environment variable or server config.v2.1.236 → v2.1.308 ◆in the build CLAUDE_CODE_TODO_REMINDER_MODE ◇ off stops the periodic todo_reminder and task_reminder attachments, which are otherwise added after at least 10 assistant turns without a todo or task tool call and 10 turns since the last reminder. baseline keeps them. When unset, the tengu_soft_slate_nudge flag decides, default baseline.v2.1.178 → v2.1.308 ◆in the build CLAUDE_CODE_TOOL_MEMORY_CGROUP_EXCLUDE ◇ Memory limits now reach npm, git, tmux, LSP, MCP and hooks, but everything except Bash is excluded.v2.1.246 → v2.1.308 ◆in the build CLAUDE_CODE_TOOL_MEMORY_LIMIT ◇ On Linux you can cap memory for shell commands so a runaway process cannot eat your machine.v2.1.233 → v2.1.308 ◆in the build CLAUDE_CODE_TOTAL_TOKENS_REMINDER ◇ Claude now gets a running count of tokens left by default, re-anchored on each of your prompts.v2.1.176 → v2.1.308 ◆in the build CLAUDE_CODE_TOTAL_TOKENS_REMINDER_AFTER_USER_TURN ◇ Claude now gets a running count of tokens left by default, re-anchored on each of your prompts.v2.1.202 → v2.1.308 ◆in the build CLAUDE_CODE_TOTAL_TOKENS_REMINDER_BUDGET ◇ The token countdown reminder budget is read once at start, so mid-session config changes are ignored.v2.1.196 → v2.1.308 ◆in the build CLAUDE_CODE_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 CLAUDE_CODE_TRANSCRIPT_SLIM_TOOL_RESULTS ◇ Five new Claude Code environment variables are now on the list passed to remote sessions, with no other use in this buildv2.1.299 → v2.1.308 ◆in the build CLAUDE_CODE_TRIGGER_ID ◇ Reported, after passing through a helper, as the trigger_id field of the tengu_started analytics event at startup. Claude Code also clears it from the environment of a child session it spawns.v2.1.207 → v2.1.308 ◆in the build CLAUDE_CODE_TUI_JUST_SWITCHED ◇ After the terminal interface relaunches, fullscreen is only restored where it is actually supported.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_TUI_TRIAL ◇ You can start a throwaway fullscreen session with an env var, and the upsell stops after three showings.v2.1.236 → v2.1.308 ◆in the build CLAUDE_CODE_TURN_UPDATES ◇ Claude can narrate what it's about to do and post progress updates; set CLAUDE_CODE_TURN_UPDATES.v2.1.229 → v2.1.308 ◆in the build CLAUDE_CODE_TWO_STAGE_CLASSIFIER ◇ No description from anybody.v2.1.160 → v2.1.308 ◆in the build CLAUDE_CODE_ULTRAREVIEW_PREFLIGHT_FIXTURE ◇ A non-empty value is parsed and validated as an ultrareview preflight response and used in place of the request to /v1/ultrareview/preflight. If it fails validation the preflight result is null.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_ULTRAREVIEW_QUOTA_FIXTURE ◇ Your remaining free cloud reviews can appear in the slash menu and in tips after commits.v2.1.224 → v2.1.308 ◆in the build CLAUDE_CODE_USE_ANTHROPIC_AWS ◇ A new provider-managed env var list decides gateway sign-in error wording.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_USE_ANTHROPIC_GOOGLE_CLOUD ◇ A new provider-managed env var list decides gateway sign-in error wording.v2.1.213 → v2.1.308 ◆in the build CLAUDE_CODE_USE_BEDROCK ◇ CLAUDE_CODE_USE_GATEWAY now passes down to spawned agent processes, so gateway routing is no longer dropped.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_USE_CCR_V2 ◇ Set by Claude Code to 1, together with CLAUDE_CODE_WORKER_EPOCH, in the environment of a session child process that bridge mode spawns after registering it as a CCR v2 worker. It is declared as a boolean in the env schema, but no code in this version reads it.v2.1.138 or earlier → v2.1.202 ○gone CLAUDE_CODE_USE_COWORK_PLUGINS ◇ Claude Code shows notices saying why hooks are off, covering unreadable managed settings, bare mode, safe mode, managed policy and --settingsv2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_USE_FOUNDRY ◇ Base-URL safety checks now group Foundry and Bedrock Mantle with the other providers.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_USE_GATEWAY ◇ CLAUDE_CODE_USE_GATEWAY now passes down to spawned agent processes, so gateway routing is no longer dropped.v2.1.195 → v2.1.308 ◆in the build CLAUDE_CODE_USE_MANTLE ◇ CLAUDE_CODE_USE_GATEWAY now passes down to spawned agent processes, so gateway routing is no longer dropped.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_USE_NATIVE_FILE_SEARCH ◇ No description from anybody.v2.1.160 → v2.1.308 ◆in the build CLAUDE_CODE_USE_POWERSHELL_TOOL ◇ New structured reason codes let a host app tell why Claude Code refused to startv2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_USE_VERTEX ◇ A new provider-managed env var list decides gateway sign-in error wording.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_USER_DIALOG_TIMEOUT_MS ◇ Set dialogExpiry to control how long a forwarded permission prompt waits before resolving to its safe default.v2.1.169 → v2.1.308 ◆in the build CLAUDE_CODE_USER_EMAIL ◇ If there is no logged-in email, Claude Code can now use the CLAUDE_CODE_USER_EMAIL environment variable as the email it gives the modelv2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_VELVET_FALCON ◇ Tri-state override for which models count as the falcon class, which claude-fable-5 and claude-mythos-5 always belong to. A true value puts every model in the class and a false value (such as 0 or false) leaves only those two; when unset, a model is also in the class if its name contains the non-empty velvet_falcon_model from cached client data or tengu_velvet_falcon_model. Falcon-class models report the family falcon and get extra system prompt guidance such as the # Communicating with the user section.v2.1.163 → v2.1.173 ○gone CLAUDE_CODE_VERIFY_PROMPT ◇ When true (1, true, yes or 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 tengu_sparrow_ledger flag turns on the same section when this variable is not true.v2.1.138 or earlier → v2.1.205 ○gone CLAUDE_CODE_VOICE_FORWARD_INTERIMS_TYPED ◇ Environment variables are now read through one typed table, with the same names and behaviour.v2.1.141 → v2.1.308 ◆in the build CLAUDE_CODE_WALNUT_SPIRE ◇ Plugin eval checks your credentials first and aborts instead of failing every case.v2.1.207 → v2.1.268 ○gone CLAUDE_CODE_WARN_DISK_FREE_GIB ◇ Setting CLAUDE_CODE_SKIP_PLUGIN_LOCAL_MCP_SERVERS keeps only remote MCP servers from plugins, and several other environment variables are newv2.1.304 → v2.1.308 ◆in the build CLAUDE_CODE_WARN_MEMORY_FREE_GIB ◇ Setting CLAUDE_CODE_SKIP_PLUGIN_LOCAL_MCP_SERVERS keeps only remote MCP servers from plugins, and several other environment variables are newv2.1.304 → v2.1.308 ◆in the build CLAUDE_CODE_WEB_FETCH_AGENT ◇ The built-in web-fetch agent refuses observer and worktree roles and just runs locally.v2.1.232 → v2.1.308 ◆in the build CLAUDE_CODE_WEB_SEARCH_FAST_ARG ◇ 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 CLAUDE_CODE_WEB_SEARCH_REFILLS_PER_HOUR ◇ WebSearch is limited by an hourly refilling budget instead of a simple count, adjustable with CLAUDE_CODE_WEB_SEARCH_REFILLS_PER_HOURv2.1.290 → v2.1.308 ◆in the build CLAUDE_CODE_WEBFETCH_CACHE_TTL_MS ◇ You can set how long WebFetch keeps pages cached, in milliseconds.v2.1.233 → v2.1.308 ◆in the build CLAUDE_CODE_WEBFETCH_DEADLINE_MS ◇ WebFetch requests now enforce a timeout and fail with a clear deadline error if exceededv2.1.268 → v2.1.308 ◆in the build CLAUDE_CODE_WEBFETCH_PROXY_PATH ◇ A path for routing WebFetch through a proxy endpoint. When the API provider is first-party and the value starts with / and contains no ://, WebFetch skips its direct fetch and domain preflight and POSTs {url} (with http:// rewritten to https://) to ANTHROPIC_BASE_URL (default https://api.anthropic.com) plus this path, with a 40 second timeout; a failed proxy call fails the fetch with the web-fetch-ccr-proxy error. Any other value is ignored.v2.1.166 → v2.1.168 ○gone CLAUDE_CODE_WEBFETCH_USE_CCR_PROXY ◇ Routing WebFetch through the CCR proxy now depends on CLAUDE_CODE_WEBFETCH_USE_CCR_PROXY instead of being worked out from session statev2.1.169 → v2.1.308 ◆in the build CLAUDE_CODE_WEBSEARCH_CCR_PROXY_FAST ◇ 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 ○gone CLAUDE_CODE_WEBSEARCH_CITATIONS ◇ With CLAUDE_CODE_WEBSEARCH_CITATIONS set, Claude is no longer told to end web search answers with a written "Sources:" listv2.1.295 → v2.1.308 ◆in the build CLAUDE_CODE_WEBSEARCH_CITATIONS_DECLARED ◇ A new environment variable, CLAUDE_CODE_WEBSEARCH_CITATIONS_DECLARED, joins the existing web search citations switchv2.1.296 ○gone CLAUDE_CODE_WEBSEARCH_USE_CCR_PROXY ◇ The web search tool has a built-in choice between a quick, cheap search and a slower, thorough one, but it stays off unless configuredv2.1.182 → v2.1.308 ◆in the build CLAUDE_CODE_WEBSOCKET_AUTH_FILE_DESCRIPTOR ◇ Child worker processes no longer inherit your session id, resume prompt or auth handles.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_WHIMSICAL_ELEPHANT ◇ A new environment variable, CLAUDE_CODE_WHIMSICAL_ELEPHANT, overrides the remote switch for a reminder tag on turns passed between agentsv2.1.284 → v2.1.308 ◆in the build CLAUDE_CODE_WILLOW_TERN ◇ A writing-style prompt block can be forced on with an environment variable; otherwise it stays off.v2.1.247 → v2.1.308 ◆in the build CLAUDE_CODE_WISE_COMET ◇ A gated experiment controls whether adaptive thinking content gets stripped from output.v2.1.260 → v2.1.308 ◆in the build CLAUDE_CODE_WITTY_DEWDROP ◇ Claude Code no longer reads CLAUDE_CODE_ADOPT_UNDERIVABLE_PARKED_PERMISSION, and a denied waiting permission request is no longer retiredv2.1.307 → v2.1.308 ◆in the build CLAUDE_CODE_WORKER_CHECKIN_SCHEDULE ◇ One schedule variable replaces the two single-interval check-in variables, and setting the old ones no longer changes anythingv2.1.283 → v2.1.308 ◆in the build CLAUDE_CODE_WORKER_EPOCH ◇ The background worker now drops duplicate telemetry events within an upload and logs how many.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS ◇ Workflow tool's concurrent-agent limit can now be overridden with an environment variablev2.1.269 → v2.1.308 ◆in the build CLAUDE_CODE_WORKFLOW_PREFIX_STAGGER_MS ◇ Parallel workflow agents stagger their first call to share cache; tune with an env var.v2.1.229 → v2.1.308 ◆in the build CLAUDE_CODE_WORKFLOW_PROMPT_PROVENANCE ◇ Workflow runs now label whether a person or a schedule triggered them, and quote the request.never mined ·never mined CLAUDE_CODE_WORKFLOW_SIZE_WARNING_AGENTS ◇ Agent-count threshold for the Large workflow warning: it shows when a workflow schedules more agents than this. It takes precedence over a cap from the workflowSizeGuideline setting, the tengu_ochre_gantry flag's agents value and the default 25; a value that is not a positive number is ignored. The warning is not computed while ultracode is active or when that flag has enabled: false, and it also fires on the separate token threshold.v2.1.203 → v2.1.308 ◆in the build CLAUDE_CODE_WORKFLOW_SIZE_WARNING_TOKENS ◇ Token threshold for the Large workflow warning on a running workflow. The warning fires when the tokens used so far, or a projection of average tokens per started agent (70000 before any agent has started) times the number of scheduled agents, exceeds it. Default is the tokens field of the tengu_ochre_gantry flag, else 1500000; an unset, non-numeric or 0 value uses the default, and no warning is computed while ultracode is active or when that flag has enabled: false.v2.1.203 → v2.1.308 ◆in the build CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL ◇ Setting CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL chooses which model subagents use during a workflow run, separately from other subagentsv2.1.296 → v2.1.308 ◆in the build CLAUDE_CODE_WORKFLOWS ◇ Environment flags are now parsed centrally, so values are compared to explicit true or false.v2.1.146 → v2.1.308 ◆in the build CLAUDE_CODE_WORKSPACE_HOST_PATHS ◇ When set to a non-empty string, every OpenTelemetry event Claude Code logs gets a workspace.host_paths attribute holding the value split on |.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CONFIG_DIR ◇ Resumed sessions no longer inherit a stale config directory or abort on a failed file copy.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_CONTEXT_COLLAPSE ◇ No description from anybody.v2.1.160 → v2.1.274 ○gone CLAUDE_CONTEXT_COLLAPSE_MODEL ◇ The CLAUDE_CONTEXT_COLLAPSE_MODEL environment variable has been removedv2.1.160 → v2.1.274 ○gone CLAUDE_COWORK_MEMORY_EXTRA_GUIDELINES ◇ Extra text added to the memory instructions in the system prompt: it is appended to the auto memory section (after the Memory directory: line in one variant, ahead of the memory index lines in the other), except where CLAUDE_COWORK_MEMORY_GUIDELINES has replaced that section. When non-empty after trimming, it is also added as an extra guideline in the Persistent Agent Memory prompt for agents with memory.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_COWORK_MEMORY_GUIDELINES ◇ When auto memory is enabled and this is non-empty, the standard auto memory section of the system prompt is replaced by # auto memory followed by this text. When set and no memory mode was chosen explicitly, it also makes Claude Code use file-based memory (files) over the memory tools, and it changes how memory index files are capped when read.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_COWORK_MEMORY_INDEX_CONTENT ◇ Adding pinned: true to a memory file's frontmatter would apply it to every session automatically.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_COWORK_MEMORY_PATH_OVERRIDE ◇ Directory to use for auto memory, taking precedence over a memory directory from settings and the default per-project location. A ~/ prefix is not expanded, and a path that fails its checks (for example shorter than 3 characters, a bare drive letter, or containing NUL) is ignored. When set, it also keeps auto memory from being turned off in CLAUDE_CODE_REMOTE sessions that have no CLAUDE_CODE_REMOTE_MEMORY_DIR.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_DEBUG ◇ When true (1, true, yes or on), each Node process warning event is also written to the debug log at warn level as [Warning] Name: message, or [Internal Warning] ... for warnings matching Claude Code's internal patterns. The tengu_node_warning analytics event is sent either way.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_DISABLE_ADOPT ◇ Artifact comment watchers now survive backgrounding and exit, and tell you when one couldn't be resumed.v2.1.195 → v2.1.308 ◆in the build CLAUDE_ENABLE_BYTE_WATCHDOG ◇ 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 CLAUDE_ENABLE_BYTE_WATCHDOG_BEDROCK ◇ Stalled streams are now detected across direct, gateway and Bedrock connections, and you can force it with an env var.v2.1.143 → v2.1.308 ◆in the build CLAUDE_ENABLE_STREAM_WATCHDOG ◇ Five more environment variables, including stream and byte watchdogs, can be set in settings.json.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_ENV_FILE ◇ Session environment files can be read and written through a storage layer instead of directly on disk.v2.1.138 or earlier → v2.1.308 ◆in the build CLAUDE_FORCE_DISPLAY_SURVEY ◇ The feedback survey checks for itself whether you're mid-prompt, so it interrupts you less.v2.1.138 or earlier → v2.1.308 ◆in the build