210| `CLAUDE_CODE_ARTIFACT_COMMENTS` | Set to `0` to stop Claude reading and replying to [comments on an artifact](/docs/en/artifacts#collect-comments-on-an-artifact). Has no effect when `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC` has [turned artifacts off](/docs/en/artifacts#availability). Requires Claude Code v2.1.221 or later |210| `CLAUDE_CODE_ARTIFACT_COMMENTS` | Set to `0` to stop Claude reading and replying to [comments on an artifact](/docs/en/artifacts#collect-comments-on-an-artifact). Has no effect when `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC` has [turned artifacts off](/docs/en/artifacts#availability). Requires Claude Code v2.1.221 or later |
211| `CLAUDE_CODE_ARTIFACT_COMMENTS_AUTOREACT` | Set to `0` to stop Claude [replying on its own to comments sent to it](/docs/en/artifacts#let-claude-reply-to-comments-on-its-own). Requires Claude Code v2.1.228 or later |211| `CLAUDE_CODE_ARTIFACT_COMMENTS_AUTOREACT` | Set to `0` to stop Claude [replying on its own to comments sent to it](/docs/en/artifacts#let-claude-reply-to-comments-on-its-own). Requires Claude Code v2.1.228 or later |
212| `CLAUDE_CODE_ATTRIBUTION_HEADER` | Set to `0` to omit the [attribution block](/docs/en/llm-gateway-protocol#system-prompt-attribution-block), which carries the client version and a prompt fingerprint, from the start of the system prompt. Caching on a direct connection to the Anthropic API is unaffected either way. In some direct-connection setups, Claude Code keeps the block on [auto mode](/docs/en/permission-modes#eliminate-prompts-with-auto-mode) classifier requests even when you set `0`. In [System prompt attribution block](/docs/en/llm-gateway-protocol#system-prompt-attribution-block), check which connections and credentials this covers. Before v2.1.181 the block included a per-request token on custom base URLs and Microsoft Foundry connections, so on those versions set it to `0` when your LLM gateway caches on the request body or forwards requests to a third-party provider, or when you connect to Microsoft Foundry directly |212| `CLAUDE_CODE_ATTRIBUTION_HEADER` | Set to `0` to omit the [attribution block](/docs/en/llm-gateway-protocol#system-prompt-attribution-block), which carries the client version and a prompt fingerprint, from the start of the system prompt. Caching on a direct connection to the Anthropic API is unaffected either way. In some direct-connection setups, Claude Code keeps the block on [auto mode](/docs/en/permission-modes#eliminate-prompts-with-auto-mode) classifier requests even when you set `0`. In [System prompt attribution block](/docs/en/llm-gateway-protocol#system-prompt-attribution-block), check which connections and credentials this covers. Before v2.1.181 the block included a per-request token on custom base URLs and Microsoft Foundry connections, so on those versions set it to `0` when your LLM gateway caches on the request body or forwards requests to a third-party provider, or when you connect to Microsoft Foundry directly |
213| `CLAUDE_CODE_AUTO_BACKGROUND_WORKER_CHECKIN_SECONDS` | When `CLAUDE_AUTO_BACKGROUND_TASKS` is enabled, seconds between reminders to Claude to check on [background subagents](/docs/en/sub-agents#run-subagents-in-foreground-or-background) that are still running. Accepts a plain integer from `1` to `86400` only; any other value or spelling reads as unset. When unset, there are no check-in reminders. Requires Claude Code v2.1.248 or later |213| `CLAUDE_CODE_AUTO_BACKGROUND_WORKER_CHECKIN_SECONDS` | Removed in v2.1.283. Use `CLAUDE_CODE_WORKER_CHECKIN_SCHEDULE` instead |
214| `CLAUDE_CODE_AUTO_COMPACT_WINDOW` | Set the [auto-compact window](/docs/en/model-config#set-the-auto-compact-window) in tokens, from `100000` to `1000000`. Accepts a plain integer such as `500000` only: a value like `500k` reads as `500` and clamps to the 100K minimum. The effective window is also capped at the model's context window. Takes precedence over the `/autocompact` command, the `--autocompact` flag, and the `autoCompactWindow` setting. The status line's `used_percentage` always measures against the model's full context window, so once this variable is set, that percentage no longer indicates when compaction will run |214| `CLAUDE_CODE_AUTO_COMPACT_WINDOW` | Set the [auto-compact window](/docs/en/model-config#set-the-auto-compact-window) in tokens, from `100000` to `1000000`. Accepts a plain integer such as `500000` only: a value like `500k` reads as `500` and clamps to the 100K minimum. The effective window is also capped at the model's context window. Takes precedence over the `/autocompact` command, the `--autocompact` flag, and the `autoCompactWindow` setting. The status line's `used_percentage` always measures against the model's full context window, so once this variable is set, that percentage no longer indicates when compaction will run |
215| `CLAUDE_CODE_AUTO_CONNECT_IDE` | Override automatic [IDE connection](/docs/en/vs-code). By default, Claude Code connects automatically when launched inside a supported IDE's integrated terminal. Set to `false` to prevent this. Set to `true` to force a connection attempt when auto-detection fails, such as when tmux obscures the parent terminal. Takes precedence over the [`autoConnectIde`](/docs/en/settings-reference#autoconnectide) global config setting |215| `CLAUDE_CODE_AUTO_CONNECT_IDE` | Override automatic [IDE connection](/docs/en/vs-code). By default, Claude Code connects automatically when launched inside a supported IDE's integrated terminal. Set to `false` to prevent this. Set to `true` to force a connection attempt when auto-detection fails, such as when tmux obscures the parent terminal. Takes precedence over the [`autoConnectIde`](/docs/en/settings-reference#autoconnectide) global config setting |
216| `CLAUDE_CODE_AUTO_MODE_SERVER` | Controls whether Claude Code asks the server to [review auto mode actions](/docs/en/permission-modes#server-side-classifier-review). Set to `0` to use Claude Code's own classifier requests instead. On a direct connection to the Anthropic API, requires v2.1.281 or later. The linked section lists which sessions ask the server when the variable is unset, and from which version. Requires Claude Code v2.1.271 or later |216| `CLAUDE_CODE_AUTO_MODE_SERVER` | Controls whether Claude Code asks the server to [review auto mode actions](/docs/en/permission-modes#server-side-classifier-review). Set to `0` to use Claude Code's own classifier requests instead. On a direct connection to the Anthropic API, requires v2.1.281 or later. The linked section lists which sessions ask the server when the variable is unset, and from which version. Requires Claude Code v2.1.271 or later |
253| `CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY` | Set to `1` to disable the "How is Claude doing?" session quality surveys. Surveys are also disabled when `DISABLE_TELEMETRY`, `DO_NOT_TRACK`, or `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC` is set, unless `CLAUDE_CODE_ENABLE_FEEDBACK_SURVEY_FOR_OTEL` opts back in. To set a sample rate instead of disabling outright, use the [`feedbackSurveyRate`](/docs/en/settings-reference#feedbacksurveyrate) setting. See [Session quality surveys](/docs/en/data-usage#session-quality-surveys) |253| `CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY` | Set to `1` to disable the "How is Claude doing?" session quality surveys. Surveys are also disabled when `DISABLE_TELEMETRY`, `DO_NOT_TRACK`, or `CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC` is set, unless `CLAUDE_CODE_ENABLE_FEEDBACK_SURVEY_FOR_OTEL` opts back in. To set a sample rate instead of disabling outright, use the [`feedbackSurveyRate`](/docs/en/settings-reference#feedbacksurveyrate) setting. See [Session quality surveys](/docs/en/data-usage#session-quality-surveys) |
254| `CLAUDE_CODE_DISABLE_FILE_CHECKPOINTING` | Set to `1` to disable file [checkpointing](/docs/en/checkpointing). The `/rewind` command will not be able to restore code changes. Overrides the [`fileCheckpointingEnabled`](/docs/en/settings-reference#filecheckpointingenabled) setting |254| `CLAUDE_CODE_DISABLE_FILE_CHECKPOINTING` | Set to `1` to disable file [checkpointing](/docs/en/checkpointing). The `/rewind` command will not be able to restore code changes. Overrides the [`fileCheckpointingEnabled`](/docs/en/settings-reference#filecheckpointingenabled) setting |
255| `CLAUDE_CODE_DISABLE_GIT_INSTRUCTIONS` | Set to `1` to remove built-in commit and PR workflow instructions and the git status snapshot from Claude's context. Useful when using your own git workflow skills. Takes precedence over the [`includeGitInstructions`](/docs/en/settings-reference#includegitinstructions) setting when set |255| `CLAUDE_CODE_DISABLE_GIT_INSTRUCTIONS` | Set to `1` to remove built-in commit and PR workflow instructions and the git status snapshot from Claude's context. Useful when using your own git workflow skills. Takes precedence over the [`includeGitInstructions`](/docs/en/settings-reference#includegitinstructions) setting when set |
256| `CLAUDE_CODE_DISABLE_INLINE_SHELL_RM_PROMPT` | Set to `1` to stop Claude Code from reading scripts passed to a shell with `-c`, such as `bash -c 'rm -rf ~'`, for [critical-path](/docs/en/permission-modes#removals-inside-nested-commands-and-inline-scripts) removals. Claude Code still checks the shell variable and positional parameter targets in those scripts, and the other critical-path checks keep running. Set it in the environment that launches Claude Code, since Claude Code ignores a copy delivered through a settings `env` block. Requires Claude Code v2.1.288 or later |
256| `CLAUDE_CODE_DISABLE_LEGACY_MODEL_REMAP` | Set to `1` to prevent automatic remapping of Opus 4.0 and 4.1 to the current Opus version on the Anthropic API. Use when you intentionally want to pin an older model. The remap does not run on Amazon Bedrock, Google Cloud's Agent Platform, or Microsoft Foundry |257| `CLAUDE_CODE_DISABLE_LEGACY_MODEL_REMAP` | Set to `1` to prevent automatic remapping of Opus 4.0 and 4.1 to the current Opus version on the Anthropic API. Use when you intentionally want to pin an older model. The remap does not run on Amazon Bedrock, Google Cloud's Agent Platform, or Microsoft Foundry |
257| `CLAUDE_CODE_DISABLE_MODEL_ACCESS_FALLBACK` | Set to `1` to stop Claude Code on [Amazon Bedrock](/docs/en/amazon-bedrock#when-a-model-is-disabled-mid-session) and [Google Cloud's Agent Platform](/docs/en/google-vertex-ai#when-a-model-is-disabled-mid-session) from switching to an older model when your account loses access to a session's model mid-session; the refused request fails at once instead. A [fallback model chain](/docs/en/model-config#fallback-model-chains) you configure still switches on that refusal, and the [startup model checks](/docs/en/amazon-bedrock#startup-model-checks) still fall back at launch. Requires Claude Code v2.1.285 or later |258| `CLAUDE_CODE_DISABLE_MODEL_ACCESS_FALLBACK` | Set to `1` to stop Claude Code on [Amazon Bedrock](/docs/en/amazon-bedrock#when-a-model-is-disabled-mid-session) and [Google Cloud's Agent Platform](/docs/en/google-vertex-ai#when-a-model-is-disabled-mid-session) from switching to an older model when your account loses access to a session's model mid-session; the refused request fails at once instead. A [fallback model chain](/docs/en/model-config#fallback-model-chains) you configure still switches on that refusal, and the [startup model checks](/docs/en/amazon-bedrock#startup-model-checks) still fall back at launch. Requires Claude Code v2.1.285 or later |
258| `CLAUDE_CODE_DISABLE_MOUSE` | Set to `1` to disable mouse tracking in [fullscreen rendering](/docs/en/fullscreen). Keyboard scrolling with `PgUp` and `PgDn` still works. Use this to keep your terminal's native copy-on-select behavior |259| `CLAUDE_CODE_DISABLE_MOUSE` | Set to `1` to disable mouse tracking in [fullscreen rendering](/docs/en/fullscreen). Keyboard scrolling with `PgUp` and `PgDn` still works. Use this to keep your terminal's native copy-on-select behavior |
265| `CLAUDE_CODE_DISABLE_PERMISSION_PROMPT_NOTIFY_HOOKS` | Set to `1` to stop Claude Code from running your [`Notification` hooks for unanswered permission requests](/docs/en/hooks#notification) in sessions where Claude Code sends them to the Agent SDK's `canUseTool` callback, which is how Claude Desktop and the VS Code extension host Claude Code. Has no effect in terminal sessions. Requires Claude Code v2.1.233 or later |266| `CLAUDE_CODE_DISABLE_PERMISSION_PROMPT_NOTIFY_HOOKS` | Set to `1` to stop Claude Code from running your [`Notification` hooks for unanswered permission requests](/docs/en/hooks#notification) in sessions where Claude Code sends them to the Agent SDK's `canUseTool` callback, which is how Claude Desktop and the VS Code extension host Claude Code. Has no effect in terminal sessions. Requires Claude Code v2.1.233 or later |
266| `CLAUDE_CODE_DISABLE_POLICY_SKILLS` | Set to `1` to skip loading skills from the system-wide managed skills directory. Useful for container or CI sessions that should not load operator-provisioned skills |267| `CLAUDE_CODE_DISABLE_POLICY_SKILLS` | Set to `1` to skip loading skills from the system-wide managed skills directory. Useful for container or CI sessions that should not load operator-provisioned skills |
267| `CLAUDE_CODE_DISABLE_POWERSHELL_CMD_RM_DENY` | Set to `1` to turn off the [PowerShell tool](/docs/en/tools-reference#powershell-tool) check that denies the `cmd` built-ins `rd`, `rmdir`, `del`, and `erase` on a [system path](/docs/en/permission-modes#remove-item-in-powershell), such as a drive root or your home directory. Claude Code ignores this variable in a settings file's `env` block. Requires Claude Code v2.1.283 or later |268| `CLAUDE_CODE_DISABLE_POWERSHELL_CMD_RM_DENY` | Set to `1` to turn off the [PowerShell tool](/docs/en/tools-reference#powershell-tool) check that denies the `cmd` built-ins `rd`, `rmdir`, `del`, and `erase` on a [system path](/docs/en/permission-modes#remove-item-in-powershell), such as a drive root or your home directory. Claude Code ignores this variable in a settings file's `env` block. Requires Claude Code v2.1.283 or later |
269| `CLAUDE_CODE_DISABLE_REFUSAL_FALLBACK` | Set to `1` to turn off the [automatic model switch when a safety classifier flags a request](/docs/en/model-config#automatic-model-fallback), the behavior the [`switchModelsOnFlag`](/docs/en/settings-reference#switchmodelsonflag) setting controls |
268| `CLAUDE_CODE_DISABLE_STRUCTURED_OUTPUTS` | Set to `1` to stop Claude Code from sending the structured-output `output_config.format` field and the `anthropic-beta` value that pairs with it, for an [LLM gateway](/docs/en/llm-gateway-protocol#feature-pass-through) whose upstream rejects them. This leaves on the other pre-release capabilities that [`CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS`](/docs/en/llm-gateway-protocol#disable-pre-release-capabilities) turns off. Requires Claude Code v2.1.288 or later |270| `CLAUDE_CODE_DISABLE_STRUCTURED_OUTPUTS` | Set to `1` to stop Claude Code from sending the structured-output `output_config.format` field and the `anthropic-beta` value that pairs with it, for an [LLM gateway](/docs/en/llm-gateway-protocol#feature-pass-through) whose upstream rejects them. This leaves on the other pre-release capabilities that [`CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS`](/docs/en/llm-gateway-protocol#disable-pre-release-capabilities) turns off. Requires Claude Code v2.1.288 or later |
269| `CLAUDE_CODE_DISABLE_SUBSTITUTION_RM_PROMPT` | Set to `1` to turn off the [critical-path](/docs/en/permission-modes#critical-paths) check for a recursive `rm` whose target is entirely the output of a command substitution, such as `rm -rf "$(pwd)"`. The other critical-path checks keep running. Set it in the environment that launches Claude Code, since Claude Code ignores a copy delivered through a settings `env` block. Requires Claude Code v2.1.281 or later |271| `CLAUDE_CODE_DISABLE_SUBSTITUTION_RM_PROMPT` | Set to `1` to turn off the [critical-path](/docs/en/permission-modes#critical-paths) check for a recursive `rm` whose target is entirely the output of a command substitution, such as `rm -rf "$(pwd)"`. The other critical-path checks keep running. Set it in the environment that launches Claude Code, since Claude Code ignores a copy delivered through a settings `env` block. Requires Claude Code v2.1.281 or later |
270| `CLAUDE_CODE_DISABLE_TERMINAL_TITLE` | Set to `1` to disable automatic terminal title updates based on conversation context. This also skips the background small/fast-model request that [generates a session title](/docs/en/sessions#name-your-sessions) |272| `CLAUDE_CODE_DISABLE_TERMINAL_TITLE` | Set to `1` to disable automatic terminal title updates based on conversation context. This also skips the background small/fast-model request that [generates a session title](/docs/en/sessions#name-your-sessions) |
384| `CLAUDE_CODE_SUBAGENT_MODEL` | The default model for [subagents](/docs/en/sub-agents#choose-a-model), [agent team](/docs/en/agent-teams#specify-teammates-and-models) teammates, and [workflow](/docs/en/workflows) agents that aren't assigned a model another way. Accepts an alias such as `haiku` or a full model name. Two sources take precedence over it: a model Claude passes when it spawns the agent, and a `model` field in the agent's definition, including `inherit`. To change that, set [`CLAUDE_CODE_SUBAGENT_MODEL_FORCE`](/docs/en/sub-agents#run-every-subagent-on-one-model). See [Choose a model](/docs/en/sub-agents#choose-a-model) for the full order. Setting it to `inherit` is the same as leaving it unset. Before v2.1.251, this variable overrode both the per-invocation model and the definition's `model` field |386| `CLAUDE_CODE_SUBAGENT_MODEL` | The default model for [subagents](/docs/en/sub-agents#choose-a-model), [agent team](/docs/en/agent-teams#specify-teammates-and-models) teammates, and [workflow](/docs/en/workflows) agents that aren't assigned a model another way. Accepts an alias such as `haiku` or a full model name. Two sources take precedence over it: a model Claude passes when it spawns the agent, and a `model` field in the agent's definition, including `inherit`. To change that, set [`CLAUDE_CODE_SUBAGENT_MODEL_FORCE`](/docs/en/sub-agents#run-every-subagent-on-one-model). See [Choose a model](/docs/en/sub-agents#choose-a-model) for the full order. Setting it to `inherit` is the same as leaving it unset. Before v2.1.251, this variable overrode both the per-invocation model and the definition's `model` field |
385| `CLAUDE_CODE_SUBAGENT_MODEL_FORCE` | Set to `1` to force one model onto subagents, teammates, and workflow agents. [Run every subagent on one model](/docs/en/sub-agents#run-every-subagent-on-one-model) says which model that is. Requires Claude Code v2.1.257 or later |387| `CLAUDE_CODE_SUBAGENT_MODEL_FORCE` | Set to `1` to force one model onto subagents, teammates, and workflow agents. [Run every subagent on one model](/docs/en/sub-agents#run-every-subagent-on-one-model) says which model that is. Requires Claude Code v2.1.257 or later |
386| `CLAUDE_CODE_SUBAGENT_PROMPT_CACHE_TTL` | Set `5m` or `1h`, the only values Claude Code accepts, to choose the [prompt cache TTL](/docs/en/prompt-caching#cache-lifetime) for requests outside the main conversation, such as [subagents](/docs/en/sub-agents), workflows, and background work. Takes precedence over the `subagentPromptCacheTtl` setting and over `ENABLE_PROMPT_CACHING_1H`, and `FORCE_PROMPT_CACHING_5M` overrides it. The API bills 1-hour cache writes at a higher rate. Requires Claude Code v2.1.242 or later |388| `CLAUDE_CODE_SUBAGENT_PROMPT_CACHE_TTL` | Set `5m` or `1h`, the only values Claude Code accepts, to choose the [prompt cache TTL](/docs/en/prompt-caching#cache-lifetime) for requests outside the main conversation, such as [subagents](/docs/en/sub-agents), workflows, and background work. Takes precedence over the `subagentPromptCacheTtl` setting and over `ENABLE_PROMPT_CACHING_1H`, and `FORCE_PROMPT_CACHING_5M` overrides it. The API bills 1-hour cache writes at a higher rate. Requires Claude Code v2.1.242 or later |
387| `CLAUDE_CODE_SUBPROCESS_ENV_SCRUB` | Set to `1` to strip credentials from subprocess environments (Bash tool, hooks, MCP stdio servers): Anthropic and cloud provider credentials, any other variable that Claude Code recognizes as a credential, and credentials embedded in package registry URLs. The parent Claude process keeps these credentials for API calls, but child processes cannot read them, reducing exposure to prompt injection attacks that attempt to exfiltrate secrets via shell expansion. On v2.1.251 or later, the scrub also removes Claude Code's own configuration-store pointer variables (such as `CLAUDE_CONFIG_DIR`), so a child process cannot locate a relocated configuration directory. Leave the scrub unset if a subprocess needs these variables. On Linux, this also runs Bash subprocesses in an isolated PID namespace so they cannot read host process environments via `/proc`; as a side effect, `ps`, `pgrep`, and `kill` cannot see or signal host processes. `claude-code-action` sets this automatically when `allowed_non_write_users` is configured |389| `CLAUDE_CODE_SUBPROCESS_ENV_SCRUB` | Set to `1` to strip credentials from the environments of the subprocesses Claude Code starts, such as Bash commands, hooks, and stdio MCP servers. The scrub recognizes a credential by its variable name or its value, and it leaves GitHub tokens and proxy settings in place. See [What the subprocess environment scrub removes](#what-the-subprocess-environment-scrub-removes). `claude-code-action` sets this automatically when `allowed_non_write_users` is configured |
388| `CLAUDE_CODE_SYNC_PLUGIN_INSTALL` | Set to `1` in non-interactive mode (the `-p` flag) to wait for plugin installation to complete before the first query. Without this, plugins install in the background and may not be available on the first turn. Combine with `CLAUDE_CODE_SYNC_PLUGIN_INSTALL_TIMEOUT_MS` to bound the wait |390| `CLAUDE_CODE_SYNC_PLUGIN_INSTALL` | Set to `1` in non-interactive mode (the `-p` flag) to wait for plugin installation to complete before the first query. Without this, plugins install in the background and may not be available on the first turn. Combine with `CLAUDE_CODE_SYNC_PLUGIN_INSTALL_TIMEOUT_MS` to bound the wait |
389| `CLAUDE_CODE_SYNC_PLUGIN_INSTALL_TIMEOUT_MS` | Timeout in milliseconds for synchronous plugin installation. When exceeded, Claude Code proceeds without plugins and logs an error. No default: without this variable, synchronous installation waits until complete |391| `CLAUDE_CODE_SYNC_PLUGIN_INSTALL_TIMEOUT_MS` | Timeout in milliseconds for synchronous plugin installation. When exceeded, Claude Code proceeds without plugins and logs an error. No default: without this variable, synchronous installation waits until complete |
390| `CLAUDE_CODE_SYNC_SKILLS` | Set to `1` in non-interactive mode with the `-p` flag to make Claude Code download the skills enabled for your claude.ai account in that run and wait for the list of them, up to `CLAUDE_CODE_SYNC_SKILLS_WAIT_TIMEOUT_MS`, before it runs the first query. The downloads themselves finish in the background, and Claude waits for a skill's download when it invokes that skill. Requires claude.ai authentication. Terminal sessions where you sign in with your claude.ai account [download these skills](/docs/en/skills#where-synced-skills-load) into `~/.claude/skills/synced/` and resync about every 10 minutes without this variable, so set it only when a `-p` run needs your current skills on its first query. Before v2.1.273, terminal sessions downloaded them only in a `-p` run with this variable set. The `synced` folder name is [reserved for this download](/docs/en/skills#where-skills-live). Before v2.1.227, the skills downloaded into `~/.claude/skills/` directly. Claude Code applies [extra rules to the downloaded skills](/docs/en/skills#how-synced-skills-behave), such as not running their `!` commands on your machine |392| `CLAUDE_CODE_SYNC_SKILLS` | Set to `1` in non-interactive mode with the `-p` flag to make Claude Code download the skills enabled for your claude.ai account in that run and wait for the list of them, up to `CLAUDE_CODE_SYNC_SKILLS_WAIT_TIMEOUT_MS`, before it runs the first query. The downloads themselves finish in the background, and Claude waits for a skill's download when it invokes that skill. Requires claude.ai authentication. Terminal sessions where you sign in with your claude.ai account [download these skills](/docs/en/skills#where-synced-skills-load) into `~/.claude/skills/synced/` and resync about every 10 minutes without this variable, so set it only when a `-p` run needs your current skills on its first query. Before v2.1.273, terminal sessions downloaded them only in a `-p` run with this variable set. The `synced` folder name is [reserved for this download](/docs/en/skills#where-skills-live). Before v2.1.227, the skills downloaded into `~/.claude/skills/` directly. Claude Code applies [extra rules to the downloaded skills](/docs/en/skills#how-synced-skills-behave), such as not running their `!` commands on your machine |
397| `CLAUDE_CODE_TMUX_TRUECOLOR` | Set to any non-empty value, such as `1`, to allow 24-bit truecolor output inside tmux. **Setting it to `0` or `false` still allows truecolor**, unlike most on/off variables; unset the variable to restore the 256-color clamp. By default, Claude Code clamps to 256 colors when `$TMUX` is set because tmux does not pass through truecolor escape sequences unless configured to. Set this after adding `set -ga terminal-overrides ',*:Tc'` to your `~/.tmux.conf`. See [Terminal configuration](/docs/en/terminal-config) for other tmux settings |399| `CLAUDE_CODE_TMUX_TRUECOLOR` | Set to any non-empty value, such as `1`, to allow 24-bit truecolor output inside tmux. **Setting it to `0` or `false` still allows truecolor**, unlike most on/off variables; unset the variable to restore the 256-color clamp. By default, Claude Code clamps to 256 colors when `$TMUX` is set because tmux does not pass through truecolor escape sequences unless configured to. Set this after adding `set -ga terminal-overrides ',*:Tc'` to your `~/.tmux.conf`. See [Terminal configuration](/docs/en/terminal-config) for other tmux settings |
398| `CLAUDE_CODE_TOOL_MEMORY_CGROUP_EXCLUDE` | On Linux and WSL, set to a comma-separated list of the kinds of processes Claude Code [excludes from the tool memory cap](/docs/en/tools-reference#memory-limit-on-linux-and-wsl), such as `mcp` or `lsp`. Set `none` to cap every kind, or `all-new` to cap only Bash, PowerShell, and Monitor tool commands. Claude Code keeps Bash, PowerShell, and Monitor tool commands under the cap whatever you list. Requires Claude Code v2.1.246 or later |400| `CLAUDE_CODE_TOOL_MEMORY_CGROUP_EXCLUDE` | On Linux and WSL, set to a comma-separated list of the kinds of processes Claude Code [excludes from the tool memory cap](/docs/en/tools-reference#memory-limit-on-linux-and-wsl), such as `mcp` or `lsp`. Set `none` to cap every kind, or `all-new` to cap only Bash, PowerShell, and Monitor tool commands. Claude Code keeps Bash, PowerShell, and Monitor tool commands under the cap whatever you list. Requires Claude Code v2.1.246 or later |
399| `CLAUDE_CODE_TOOL_MEMORY_LIMIT` | On Linux and WSL, set to a size such as `4G` to [cap the memory that Bash and PowerShell tool commands can use](/docs/en/tools-reference#memory-limit-on-linux-and-wsl), and Monitor tool commands on v2.1.246 or later. Write the size in plain digits, alone for a number of bytes or with a `K`, `M`, `G`, or `T` suffix. Set `0` or `off` to turn the cap off. Once the first process Claude Code starts has turned the cap on or off, a changed value takes effect the next time you launch `claude`. Requires Claude Code v2.1.233 or later |401| `CLAUDE_CODE_TOOL_MEMORY_LIMIT` | On Linux and WSL, set to a size such as `4G` to [cap the memory that Bash and PowerShell tool commands can use](/docs/en/tools-reference#memory-limit-on-linux-and-wsl), and Monitor tool commands on v2.1.246 or later. Write the size in plain digits, alone for a number of bytes or with a `K`, `M`, `G`, or `T` suffix. Set `0` or `off` to turn the cap off. Once the first process Claude Code starts has turned the cap on or off, a changed value takes effect the next time you launch `claude`. Requires Claude Code v2.1.233 or later |
402| `CLAUDE_CODE_TRANSCRIPT_LOCAL_GC` | Set to `1` to limit how large the [transcript file](/docs/en/sessions#where-transcripts-are-stored) of a long `-p` or Agent SDK session grows. After each compaction, once the file is larger than 5 MB, Claude Code removes the history from before that compaction. Resuming the session restores the same conversation whether or not the file was trimmed. Set it in the environment you start Claude Code from, since a settings `env` block can't turn it on. Requires Claude Code v2.1.287 or later |
400| `CLAUDE_CODE_USER_DIALOG_TIMEOUT_MS` | Deadline in milliseconds before Claude Code cancels a dialog it forwards to a remote client such as a [Remote Control](/docs/en/remote-control) or SDK host, or the approval dialog for a [held cross-session message](/docs/en/cross-session-messaging#control-inbound-messages); permission prompts and `AskUserQuestion` questions use their own flows and aren't governed by it. On Claude Code v2.1.236 or later, it also bounds the mid-session [Fable usage-credits consent prompt](/docs/en/model-config#fable-and-usage-credits) in a session that may be running unattended. [Control inbound messages](/docs/en/cross-session-messaging#control-inbound-messages) and [non-interactive sessions](/docs/en/cross-session-messaging#non-interactive-sessions) cover the full held-message expiry rules, including the cases where the deadline doesn't apply. Overrides the [`dialogExpiry`](/docs/en/settings-reference#dialogexpiry) setting. `0` or a negative value disables the deadline |403| `CLAUDE_CODE_USER_DIALOG_TIMEOUT_MS` | Deadline in milliseconds before Claude Code cancels a dialog it forwards to a remote client such as a [Remote Control](/docs/en/remote-control) or SDK host, or the approval dialog for a [held cross-session message](/docs/en/cross-session-messaging#control-inbound-messages); permission prompts and `AskUserQuestion` questions use their own flows and aren't governed by it. On Claude Code v2.1.236 or later, it also bounds the mid-session [Fable usage-credits consent prompt](/docs/en/model-config#fable-and-usage-credits) in a session that may be running unattended. [Control inbound messages](/docs/en/cross-session-messaging#control-inbound-messages) and [non-interactive sessions](/docs/en/cross-session-messaging#non-interactive-sessions) cover the full held-message expiry rules, including the cases where the deadline doesn't apply. Overrides the [`dialogExpiry`](/docs/en/settings-reference#dialogexpiry) setting. `0` or a negative value disables the deadline |
401| `CLAUDE_CODE_USE_ANTHROPIC_AWS` | Use [Claude Platform on AWS](/docs/en/claude-platform-on-aws) |404| `CLAUDE_CODE_USE_ANTHROPIC_AWS` | Use [Claude Platform on AWS](/docs/en/claude-platform-on-aws) |
402| `CLAUDE_CODE_USE_BEDROCK` | Use [Amazon Bedrock](/docs/en/amazon-bedrock) |405| `CLAUDE_CODE_USE_BEDROCK` | Use [Amazon Bedrock](/docs/en/amazon-bedrock) |
407| `CLAUDE_CODE_USE_VERTEX` | Use [Google Cloud's Agent Platform](/docs/en/google-vertex-ai) |410| `CLAUDE_CODE_USE_VERTEX` | Use [Google Cloud's Agent Platform](/docs/en/google-vertex-ai) |
408| `CLAUDE_CODE_WEBFETCH_CACHE_TTL_MS` | Set to the number of milliseconds [WebFetch](/docs/en/tools-reference#webfetch-tool-behavior) keeps each fetched URL's response cached. The default is `900000`, which is 15 minutes. Takes plain digits only; `0`, a decimal, or any other spelling keeps the default. Claude Code reads the value once per launch, so a change in a settings `env` block applies when you next launch `claude`. Requires Claude Code v2.1.233 or later |411| `CLAUDE_CODE_WEBFETCH_CACHE_TTL_MS` | Set to the number of milliseconds [WebFetch](/docs/en/tools-reference#webfetch-tool-behavior) keeps each fetched URL's response cached. The default is `900000`, which is 15 minutes. Takes plain digits only; `0`, a decimal, or any other spelling keeps the default. Claude Code reads the value once per launch, so a change in a settings `env` block applies when you next launch `claude`. Requires Claude Code v2.1.233 or later |
409| `CLAUDE_CODE_WEBFETCH_DEADLINE_MS` | Upper bound in milliseconds on how long [WebFetch](/docs/en/tools-reference#webfetch-tool-behavior) waits for a page to download, including any redirects it follows. A download that hasn't completed by then fails with a deadline error. The default is `300000`, which is five minutes. Set to `0` to remove the limit. Takes plain digits only; a decimal or any other spelling keeps the default. Requires Claude Code v2.1.268 or later |412| `CLAUDE_CODE_WEBFETCH_DEADLINE_MS` | Upper bound in milliseconds on how long [WebFetch](/docs/en/tools-reference#webfetch-tool-behavior) waits for a page to download, including any redirects it follows. A download that hasn't completed by then fails with a deadline error. The default is `300000`, which is five minutes. Set to `0` to remove the limit. Takes plain digits only; a decimal or any other spelling keeps the default. Requires Claude Code v2.1.268 or later |
413| `CLAUDE_CODE_WORKER_CHECKIN_SCHEDULE` | When `CLAUDE_AUTO_BACKGROUND_TASKS` is set to `1`, how long Claude Code waits before each reminder to Claude to check on [background subagents](/docs/en/sub-agents#run-subagents-in-foreground-or-background) that are still running. Takes one or more comma-separated waits in whole seconds from `1` to `86400`, such as `600` or `600,1800,3600`. Each value is the wait before the next reminder, and the last value repeats. Takes plain digits only; any other value or spelling reads as unset. When unset, there are no reminders. Requires Claude Code v2.1.283 or later |
410| `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` | How many agents a single [workflow](/docs/en/workflows) run executes at once, from `1` to `256`. By default, a run executes up to 16 agents at once, fewer when Claude Code has fewer CPUs available; queued `agent()` calls wait for a free slot. Each running agent's transcript stays in Claude Code's memory, so higher values raise memory use. Takes plain digits only; out-of-range values and other spellings keep the default. Requires Claude Code v2.1.269 or later |414| `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` | How many agents a single [workflow](/docs/en/workflows) run executes at once, from `1` to `256`. By default, a run executes up to 16 agents at once, fewer when Claude Code has fewer CPUs available; queued `agent()` calls wait for a free slot. Each running agent's transcript stays in Claude Code's memory, so higher values raise memory use. Takes plain digits only; out-of-range values and other spellings keep the default. Requires Claude Code v2.1.269 or later |
411| `CLAUDE_CODE_WORKFLOW_PREFIX_STAGGER_MS` | Upper bound in milliseconds on how long a [workflow](/docs/en/workflows) agent waits for a same-prefix sibling's first response to begin before sending its own first request. When a fan-out starts several agents that share a [prompt-cache prefix](/docs/en/workflows#prompt-caching-in-a-fan-out), Claude Code holds all but the first agent for up to this long so the rest read the cached prefix instead of each processing it uncached. Default `5000`. Set to `0` to disable the wait. When `DISABLE_PROMPT_CACHING` is set, agents never wait. Requires Claude Code v2.1.229 or later |415| `CLAUDE_CODE_WORKFLOW_PREFIX_STAGGER_MS` | Upper bound in milliseconds on how long a [workflow](/docs/en/workflows) agent waits for a same-prefix sibling's first response to begin before sending its own first request. When a fan-out starts several agents that share a [prompt-cache prefix](/docs/en/workflows#prompt-caching-in-a-fan-out), Claude Code holds all but the first agent for up to this long so the rest read the cached prefix instead of each processing it uncached. Default `5000`. Set to `0` to disable the wait. When `DISABLE_PROMPT_CACHING` is set, agents never wait. Requires Claude Code v2.1.229 or later |
412| `CLAUDE_CONFIG_DIR` | Override the configuration directory (default: `~/.claude`). All settings, session history, and plugins are stored under this path. For credentials, see [where Claude Code stores credentials](/docs/en/authentication#credential-management). Useful for running multiple accounts side by side: for example, `alias claude-work='CLAUDE_CONFIG_DIR=~/.claude-work claude'`. Set it in your shell, user settings, or managed settings. Ignored in [project and local settings](/docs/en/settings-reference#variables-claude-code-ignores-in-env) |416| `CLAUDE_CONFIG_DIR` | Override the configuration directory (default: `~/.claude`). All settings, session history, and plugins are stored under this path. For credentials, see [where Claude Code stores credentials](/docs/en/authentication#credential-management). Useful for running multiple accounts side by side: for example, `alias claude-work='CLAUDE_CONFIG_DIR=~/.claude-work claude'`. Set it in your shell, user settings, or managed settings. Ignored in [project and local settings](/docs/en/settings-reference#variables-claude-code-ignores-in-env) |
460| `IS_DEMO` | Set to any non-empty value, such as `1`, to enable demo mode: hides your email and organization name from the header and `/status` output, and skips onboarding. **Setting it to `0` or `false` still enables demo mode**, unlike most on/off variables; unset the variable to turn it off. Useful when streaming or recording a session |464| `IS_DEMO` | Set to any non-empty value, such as `1`, to enable demo mode: hides your email and organization name from the header and `/status` output, and skips onboarding. **Setting it to `0` or `false` still enables demo mode**, unlike most on/off variables; unset the variable to turn it off. Useful when streaming or recording a session |
461| `MAX_MCP_OUTPUT_TOKENS` | Maximum number of tokens allowed in MCP tool responses. Claude Code displays a warning when output exceeds 10,000 tokens. Tools that declare [`anthropic/maxResultSizeChars`](/docs/en/mcp#raise-the-limit-for-a-specific-tool) use that character limit for text content instead, but image content from those tools is still subject to this variable (default: 25000) |465| `MAX_MCP_OUTPUT_TOKENS` | Maximum number of tokens allowed in MCP tool responses. Claude Code displays a warning when output exceeds 10,000 tokens. Tools that declare [`anthropic/maxResultSizeChars`](/docs/en/mcp#raise-the-limit-for-a-specific-tool) use that character limit for text content instead, but image content from those tools is still subject to this variable (default: 25000) |
462| `MAX_STRUCTURED_OUTPUT_RETRIES` | Number of attempts Claude Code allows when the model's response fails validation against the [`--json-schema`](/docs/en/cli-reference#cli-flags) in non-interactive mode with the `-p` flag; after that many failed attempts with no valid output, the run fails. The same cap applies when a [workflow](/docs/en/workflows) subagent's structured output fails validation. Defaults to 5, a first attempt plus four retries |466| `MAX_STRUCTURED_OUTPUT_RETRIES` | Number of attempts Claude Code allows when the model's response fails validation against the [`--json-schema`](/docs/en/cli-reference#cli-flags) in non-interactive mode with the `-p` flag; after that many failed attempts with no valid output, the run fails. The same cap applies when a [workflow](/docs/en/workflows) subagent's structured output fails validation. Defaults to 5, a first attempt plus four retries |
463| `MAX_THINKING_TOKENS` | Fixed token budget for [extended thinking](https://platform.claude.com/docs/en/build-with-claude/extended-thinking). Claude Code caps it at one token below the request's max output tokens and never below 1,024. See `CLAUDE_CODE_MAX_OUTPUT_TOKENS` for how that limit is set. When unset and thinking is enabled, models with [adaptive reasoning](/docs/en/model-config#adjust-effort-level) choose their own thinking depth, and other models use the cap. Set to `0` to disable thinking on the Anthropic API, except on Opus 5.5, Sonnet 5.5, and the Fable models, which can't have thinking turned off. On [third-party providers](/docs/en/third-party-integrations), `0` omits the `thinking` parameter instead. With thinking turned off on the Anthropic API, Claude Code sends effort `high` instead of a higher level to models it knows [don't accept that combination](/docs/en/errors#effort-isnt-available-with-thinking-turned-off), such as Opus 5. Claude Code ignores nonzero values on adaptive reasoning models, except on the models where `CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING` turns adaptive reasoning off |467| `MAX_THINKING_TOKENS` | Fixed token budget for [extended thinking](https://platform.claude.com/docs/en/build-with-claude/extended-thinking). Claude Code caps it at one token below the request's max output tokens and never below 1,024. See `CLAUDE_CODE_MAX_OUTPUT_TOKENS` for how that limit is set. When unset and thinking is enabled, models with [adaptive reasoning](/docs/en/model-config#adjust-effort-level) choose their own thinking depth, and other models use the cap. Set to `0` to disable thinking on the Anthropic API, except on Opus 5.5, Sonnet 5.5, and the Fable models, which can't have thinking turned off. On [third-party providers](/docs/en/third-party-integrations), `0` omits the `thinking` parameter instead. With thinking turned off on the Anthropic API, Claude Code sends effort `high` instead of a higher level to models it knows [don't accept that combination](/docs/en/errors#effort-isnt-available-with-thinking-turned-off), such as Opus 5. For a positive value, Claude Code ignores the number itself on adaptive reasoning models, except when `CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING` turns adaptive reasoning off |
464| `MCP_CLIENT_SECRET` | OAuth client secret for MCP servers that require [pre-configured credentials](/docs/en/mcp#use-pre-configured-oauth-credentials). Avoids the interactive prompt when adding a server with `--client-secret` |468| `MCP_CLIENT_SECRET` | OAuth client secret for MCP servers that require [pre-configured credentials](/docs/en/mcp#use-pre-configured-oauth-credentials). Avoids the interactive prompt when adding a server with `--client-secret` |
465| `MCP_CONNECTION_NONBLOCKING` | Controls whether startup waits for MCP servers to connect before the first query. MCP startup is non-blocking by default: servers connect in the background and their tools become available as they finish. Set to `0` to make Claude Code wait for servers to connect before the first query. Servers configured with [`alwaysLoad: true`](/docs/en/mcp#exempt-a-server-from-deferral) still make startup wait regardless, except when served from the [discovery cache](/docs/en/mcp#server-status-detail), since their tools must be present when the first prompt is built. In non-interactive mode (`-p`) without `--input-format stream-json`, Claude Code also waits for still-pending servers before the first turn regardless of this variable. When you pass [`--mcp-config`](/docs/en/cli-reference#cli-flags) explicitly, the wait has a longer deadline; see that flag's entry for the cached-server exception |469| `MCP_CONNECTION_NONBLOCKING` | Controls whether startup waits for MCP servers to connect before the first query. MCP startup is non-blocking by default: servers connect in the background and their tools become available as they finish. Set to `0` to make Claude Code wait for servers to connect before the first query. Servers configured with [`alwaysLoad: true`](/docs/en/mcp#exempt-a-server-from-deferral) still make startup wait regardless, except when served from the [discovery cache](/docs/en/mcp#server-status-detail), since their tools must be present when the first prompt is built. In non-interactive mode (`-p`) without `--input-format stream-json`, Claude Code also waits for still-pending servers before the first turn regardless of this variable. When you pass [`--mcp-config`](/docs/en/cli-reference#cli-flags) explicitly, the wait has a longer deadline; see that flag's entry for the cached-server exception |
466| `MCP_CONNECT_TIMEOUT_MS` | How long blocking MCP startup waits, in milliseconds, for the connection batch before snapshotting the tool list (default: 5000). Applies when `MCP_CONNECTION_NONBLOCKING=0` or for servers marked [`alwaysLoad: true`](/docs/en/mcp#exempt-a-server-from-deferral). Servers still pending at the deadline keep connecting in the background. Distinct from `MCP_TIMEOUT`, which bounds an individual server's connect attempt |470| `MCP_CONNECT_TIMEOUT_MS` | How long blocking MCP startup waits, in milliseconds, for the connection batch before snapshotting the tool list (default: 5000). Applies when `MCP_CONNECTION_NONBLOCKING=0` or for servers marked [`alwaysLoad: true`](/docs/en/mcp#exempt-a-server-from-deferral). Servers still pending at the deadline keep connecting in the background. Distinct from `MCP_TIMEOUT`, which bounds an individual server's connect attempt |
516 520
517Set `CLAUDE_CODE_ENABLE_TELEMETRY` and the OpenTelemetry variables that turn on export, choose its destination, or capture content in your shell, user settings, or managed settings. Claude Code [ignores them in project and local settings](/docs/en/settings-reference#variables-claude-code-ignores-in-env), apart from the off values that section describes. `OTEL_RESOURCE_ATTRIBUTES` and the export interval, timeout, and compression variables, such as `OTEL_METRIC_EXPORT_INTERVAL`, still apply from project and local settings.521Set `CLAUDE_CODE_ENABLE_TELEMETRY` and the OpenTelemetry variables that turn on export, choose its destination, or capture content in your shell, user settings, or managed settings. Claude Code [ignores them in project and local settings](/docs/en/settings-reference#variables-claude-code-ignores-in-env), apart from the off values that section describes. `OTEL_RESOURCE_ATTRIBUTES` and the export interval, timeout, and compression variables, such as `OTEL_METRIC_EXPORT_INTERVAL`, still apply from project and local settings.
518 522
523## What the subprocess environment scrub removes
524
525When you set [`CLAUDE_CODE_SUBPROCESS_ENV_SCRUB`](#variables) to `1`, Claude Code removes credentials from the environments of the subprocesses it starts, such as Bash commands, hooks, and stdio MCP servers. This reduces what a prompt injection attack can read through shell expansion. The Claude Code process keeps the credentials for its own API calls.
526
527The scrub recognizes a credential by its variable name or by the shape of its value, so use it as one layer alongside narrow [permission rules](/docs/en/permissions) rather than as the only control.
528
529The table shows what the scrub does to example variables:
530
531| Example variable | What the scrub does |
532| :- | :- |
533| `ANTHROPIC_API_KEY`, `AWS_SECRET_ACCESS_KEY` | Removes it |
534| `NPM_TOKEN`, `DB_PASSWORD` | Removes it, because the name looks like a credential |
535| `DATABASE_URL` that contains a password | Removes it, because the value looks like a credential |
536| `PIP_INDEX_URL` or `NPM_CONFIG_REGISTRY` that contains a password | Keeps the URL and cuts the username and password from it |
537| `CLAUDE_CONFIG_DIR` | Removes it. Requires Claude Code v2.1.251 or later |
538| `GITHUB_TOKEN`, `GH_TOKEN`, `GH_ENTERPRISE_TOKEN`, `GITHUB_ENTERPRISE_TOKEN` | Leaves it in place, so that `gh` and scripts that call the GitHub API keep working |
539| `HTTP_PROXY`, `HTTPS_PROXY` | Leaves it in place, including a [username and password in the URL](/docs/en/network-config#basic-authentication). The [sandbox](/docs/en/sandboxing#network-isolation) can set these variables itself for sandboxed commands |
540| `GIT_CONFIG_COUNT`, `GIT_CONFIG_KEY_<n>`, `GIT_CONFIG_VALUE_<n>` | Leaves it in place, whatever it holds |
541| A secret whose variable name and value don't look like a credential | Leaves it in place |
542
543Because the scrub leaves `GITHUB_TOKEN` in place, give a GitHub Actions job the narrowest `permissions` it needs. To remove a GitHub token from sandboxed Bash commands, add a `deny` entry under [`sandbox.credentials`](/docs/en/sandboxing#protect-credentials).
544
545Leave the scrub unset if a subprocess needs one of the removed variables.
546
547On Linux, the scrub also runs Bash subprocesses in an isolated PID namespace so they can't read host process environments through `/proc`. As a side effect, `ps`, `pgrep`, and `kill` can't see or signal host processes.
548
519## Features that need feature-flag fetching549## Features that need feature-flag fetching
520 550
521Claude Code turns some features on through feature flags it fetches from Anthropic. Claude Code skips that fetch in these sessions:551Claude Code turns some features on through feature flags it fetches from Anthropic. Claude Code skips that fetch in these sessions:
543 573
544### First session after an install or upgrade574### First session after an install or upgrade
545 575
546In your first session after you install Claude Code, or upgrade to a version that adds a feature, a [flag-gated feature](#features-that-need-feature-flag-fetching) can be missing. That session can also start in a different [permission mode](/docs/en/permission-modes#which-mode-a-session-starts-in) than your later sessions do. Claude Code fetches the flags during that session, so your next session has the feature and the usual starting permission mode.576In your first session after you install Claude Code, or upgrade to a version that adds a feature, a [flag-gated feature](#features-that-need-feature-flag-fetching) can be missing. That session can also start in a different [permission mode](/docs/en/permission-modes#which-mode-a-session-starts-in) than your later sessions do. When Claude Code fetches the flags during that session, it saves them on the machine, so your next session on that machine has the feature and the usual starting permission mode.
577
578After a fresh install, in a non-interactive session such as `claude -p`, the Agent SDK, or the VS Code extension, Claude Code can pick the flags up before it [chooses the starting permission mode](/docs/en/permission-modes#which-mode-a-session-starts-in), but it doesn't always wait for them.
579
580In setups such as these, sessions after the first also start without freshly fetched flags:
581
582* **A clean environment on every run**: if each run starts in a CI container, or any other environment without the flags an earlier session saved, every run is a first session
583* **A gateway token with no API key**: if you authenticate with `ANTHROPIC_AUTH_TOKEN` and no API key, and `ANTHROPIC_BASE_URL` points at a host other than Anthropic's, such as an [LLM gateway](/docs/en/llm-gateway), Claude Code has no credential to fetch the flags with
547 584
548After a fresh install, in a non-interactive session such as `claude -p`, the Agent SDK, or the VS Code extension, Claude Code can still pick the flags up before it [chooses the starting permission mode](/docs/en/permission-modes#which-mode-a-session-starts-in).585To choose the permission mode that sessions in these setups start in, see [Start in a different permission mode](/docs/en/permission-modes#start-in-a-different-mode).
549 586
550## See also587## See also
551 588
