4 4
5# Configure the sandboxed Bash tool5# Configure the sandboxed Bash tool
6 6
7> Learn how Claude Code's sandboxed Bash tool provides filesystem and network isolation for safer, more autonomous agent execution.7> Restrict the files and network hosts Claude Code's shell commands can reach with the built-in sandbox. Turn it on, set the boundary, and fix what it breaks.
8 8
9The Bash sandbox lets Claude run most shell commands without stopping to ask permission. Instead of approving each command, you define which files and network domains commands can touch, and the operating system enforces that boundary for every Bash, PowerShell, or Monitor command and its child processes.9The Bash sandbox is a boundary that the operating system enforces around the shell commands Claude runs on your machine. You set which files and network domains those commands can reach, and the limits apply to Bash, PowerShell, and Monitor commands and the processes they start. Because the operating system applies the limits while a command runs, Claude Code can [run sandboxed commands without asking you](#sandbox-modes) to approve each one.
10
11The sandbox covers shell commands only. Claude's file tools, MCP servers, and hooks [run outside it](#what-runs-outside-the-sandbox).
12
13The sandbox runs on macOS, Linux, and WSL2. On native Windows, Claude Code runs commands unsandboxed. To use the sandbox on a Windows machine, run Claude Code inside a WSL2 distribution.
10 14
11<Note>15<Note>
12 To compare other isolation approaches such as dev containers, custom containers, and virtual machines, see [Sandbox environments](/docs/en/sandbox-environments). To reduce permission prompts for tools other than Bash, see [permission modes](/docs/en/permission-modes).16 This page covers the sandbox around shell commands on your own machine. Other pages cover related questions:
17
18 * For how a cloud session is isolated, see [Security and isolation](/docs/en/claude-code-on-the-web#security-and-isolation)
19 * To compare other isolation approaches such as dev containers, custom containers, and virtual machines, see [Sandbox environments](/docs/en/sandbox-environments)
20 * To reduce permission prompts for tools other than Bash, see [permission modes](/docs/en/permission-modes)
13</Note>21</Note>
14 22
23## What the sandbox restricts
24
25While the sandbox is on, the shell commands Claude runs start inside its boundary, and so do the processes they start. The sandbox is off by default. To turn it on, run `/sandbox` in a session, as [Get started](#get-started) shows, or set [`sandbox.enabled`](/docs/en/settings-reference#sandbox-enabled) to `true` in a [settings file](/docs/en/settings) such as `~/.claude/settings.json`.
26
27The table shows what a sandboxed command can reach by default and the settings that change each default.
28
29| Access | Default | Change it with |
30| :- | :- | :- |
31| Writes | The working directory, a per-user temp directory, and [directories you've added](/docs/en/permissions#additional-directories-grant-file-access-not-configuration). [Protected paths](#protected-paths) stay write-denied | [`filesystem.allowWrite`](/docs/en/settings-reference#sandbox-filesystem-allowwrite), [`filesystem.denyWrite`](/docs/en/settings-reference#sandbox-filesystem-denywrite) |
32| Reads | Most of the machine, including credential files such as `~/.ssh` and `~/.aws/credentials` | [`filesystem.denyRead`](/docs/en/settings-reference#sandbox-filesystem-denyread), [`credentials`](#protect-credentials) |
33| Network | No direct route out. Connections go through a proxy on your machine that checks each host against your allowed domains, which start empty. Your permission mode decides [what happens to other hosts](#hosts-outside-your-allowed-domains) | [`network.allowedDomains`](/docs/en/settings-reference#sandbox-network-alloweddomains), [`network.deniedDomains`](/docs/en/settings-reference#sandbox-network-denieddomains) |
34| Environment variables | Inherited from Claude Code, including any secrets in its environment | [`credentials`](#protect-credentials), [`CLAUDE_CODE_SUBPROCESS_ENV_SCRUB`](/docs/en/env-vars) |
35
36Claude Code builds the sandbox on the open source [`@anthropic-ai/sandbox-runtime`](https://github.com/anthropics/sandbox-runtime) package.
37
38### What runs outside the sandbox
39
40The sandbox wraps shell commands. These tools and processes run outside it:
41
42* **Built-in file and web tools**: tools such as Read, Edit, Write, WebFetch, and WebSearch follow [permission rules](/docs/en/permissions) instead. A `denyRead` entry doesn't stop the Read tool, and `allowedDomains` doesn't limit WebFetch
43* **Other processes Claude Code starts**: command [hooks](/docs/en/hooks), local [MCP servers](/docs/en/mcp), [plugin monitors](/docs/en/plugins/components#monitors), [LSP servers](/docs/en/tools-reference#lsp-tool-behavior), and helper commands such as your [status line](/docs/en/statusline) command and `apiKeyHelper` run with your full access
44
45Some shell commands also run outside the sandbox, depending on your settings:
46
47* **Commands you type yourself**: a command you enter at the [`!` shell-mode prompt](/docs/en/interactive-mode#shell-mode-with-prefix) runs unsandboxed in most sessions. [Strict sandbox mode](#turn-off-the-retry-with-strict-sandbox-mode) lists the sessions where a command you type runs sandboxed
48* **Excluded commands**: commands that match [`excludedCommands`](#run-commands-outside-the-sandbox-with-excludedcommands) run unsandboxed
49* **Unsandboxed retries**: Claude can [ask to run a command unsandboxed](#the-unsandboxed-retry-escape-hatch), usually after it fails in the sandbox
50
51To put the tools, processes, and commands in this section behind one boundary, run the Claude Code process itself in a [container, virtual machine, or the sandbox runtime](/docs/en/sandbox-environments).
52
15## Get started53## Get started
16 54
17The sandbox is built into Claude Code and runs on macOS, Linux, and WSL2. Native Windows is not supported. On Windows, run Claude Code inside a WSL2 distribution.55The sandbox is built into Claude Code. What you install depends on your platform:
18 56
19On macOS, there is nothing to install: sandboxing uses the built-in Seatbelt framework. On Linux and WSL2, the sandbox relies on two packages, covered in [Set up Linux and WSL2](#set-up-linux-and-wsl2). Even if you haven't installed them yet, you can start with `/sandbox`, because its panel shows whether anything is missing.57* **macOS**: sandboxing uses the built-in Seatbelt framework, so you can go straight to the steps
58* **Linux and WSL2**: the sandbox relies on `bubblewrap` and `socat`, covered in [Set up Linux and WSL2](#set-up-linux-and-wsl2). Even if you haven't installed them yet, you can start with `/sandbox`, because its panel shows whether anything is missing
20 59
21<Steps>60<Steps>
22 <Step title="Run /sandbox">61 <Step title="Run /sandbox">
44 83
45 The first time a command needs a new network domain, Claude Code prompts for approval; in [auto mode](/docs/en/permission-modes#eliminate-prompts-with-auto-mode), Claude instead names the hosts a command needs [on the command itself](#per-command-allowed-domains-in-auto-mode) for the classifier to review with it.84 The first time a command needs a new network domain, Claude Code prompts for approval; in [auto mode](/docs/en/permission-modes#eliminate-prompts-with-auto-mode), Claude instead names the hosts a command needs [on the command itself](#per-command-allowed-domains-in-auto-mode) for the classifier to review with it.
46 85
47 Commands that can't run sandboxed fall back to the regular permission flow. Claude Code titles their permission prompt "Bash command (unsandboxed)" instead of "Bash command", so you can tell which commands ran outside the sandbox. To widen or narrow what the sandbox allows, see [Configure sandboxing](#configure-sandboxing).86 To widen or narrow what the sandbox allows, see [Configure sandboxing](#configure-sandboxing).
48 87
49 If sandboxed commands fail with `Operation not permitted` inside a container, see the Bubblewrap entry under [Troubleshooting](#troubleshooting).88 If sandboxed commands fail with `Operation not permitted` inside a container, see [Bubblewrap fails to start inside a container](#bubblewrap-fails-to-start-inside-a-container).
50 </Step>89 </Step>
51</Steps>90</Steps>
52 91
59```98```
60 99
61<Warning>100<Warning>
62 By default, if the sandbox cannot start because dependencies are missing or the platform is unsupported, Claude Code shows a warning and runs commands without sandboxing. To make this a hard failure instead, set [`sandbox.failIfUnavailable`](/docs/en/settings-reference#sandbox-failifunavailable) to `true`. This is intended for managed deployments that require sandboxing as a security gate.101 By default, if the sandbox can't start because a dependency is missing or the platform is unsupported, Claude Code runs commands without sandboxing. To make Claude Code exit at startup instead, set [`sandbox.failIfUnavailable`](/docs/en/settings-reference#sandbox-failifunavailable) to `true`. Managed deployments that require sandboxing as a security gate can use this setting.
63</Warning>102</Warning>
64 103
104### Confirm commands run inside the sandbox
105
106To check that the sandbox is working, ask Claude to run each line in the table. What you type at the [`!` prompt](#what-runs-outside-the-sandbox) usually runs outside the sandbox, so typing a line yourself doesn't test it.
107
108| Command | Result inside the sandbox |
109| :- | :- |
110| `touch ~/sandbox-probe` | Fails with `Operation not permitted` on macOS, or `Read-only file system` on Linux and WSL2 |
111| `curl --noproxy '*' https://example.com` | Fails with `Could not resolve host`, because the command has no route around the sandbox proxy |
112
113If Claude asks to retry a failed command outside the sandbox, decline the retry. If `touch` succeeds and your home directory isn't one of the directories the sandbox lets commands write to, delete `~/sandbox-probe`. Then run `/sandbox` to check that the sandbox is on and its dependencies are installed.
114
65### Set up Linux and WSL2115### Set up Linux and WSL2
66 116
67On Linux and WSL2, the sandbox relies on two packages:117On Linux and WSL2, the sandbox relies on these packages:
68 118
69* [`bubblewrap`](https://github.com/containers/bubblewrap): the unprivileged sandboxing tool that enforces filesystem isolation119* [`bubblewrap`](https://github.com/containers/bubblewrap): the unprivileged sandboxing tool that enforces filesystem isolation
70* [`socat`](http://www.dest-unreach.org/socat/): the relay used to route network traffic through the sandbox proxy120* [`socat`](http://www.dest-unreach.org/socat/): the relay used to route network traffic through the sandbox proxy
119 <Accordion title="WSL2 notes">169 <Accordion title="WSL2 notes">
120 Check your WSL version with `wsl -l -v` from PowerShell. If you see `Sandboxing requires WSL2`, your distribution is running WSL1. Upgrade it to WSL2 or run Claude Code without sandboxing.170 Check your WSL version with `wsl -l -v` from PowerShell. If you see `Sandboxing requires WSL2`, your distribution is running WSL1. Upgrade it to WSL2 or run Claude Code without sandboxing.
121 171
122 On WSL2, WSL hands a launch of a Windows binary such as `cmd.exe`, `powershell.exe`, or anything under `/mnt/c/` to the Windows host over a Unix socket, so whether a sandboxed command can launch one follows the sandbox's [Unix-socket settings](/docs/en/settings-reference#sandbox-network-allowunixsockets): the optional seccomp filter has to be installed to block the socket in the first place. To allow these launches, set `allowAllUnixSockets`; to keep them out of the sandbox entirely, add the command to [`excludedCommands`](/docs/en/settings-reference#sandbox-excludedcommands).172 On WSL2, WSL hands a launch of a Windows binary such as `cmd.exe`, `powershell.exe`, or anything under `/mnt/c/` to the Windows host over a Unix socket, so whether a sandboxed command can launch one follows the sandbox's [Unix-socket settings](/docs/en/settings-reference#sandbox-network-allowunixsockets): the optional seccomp filter has to be installed to block the socket in the first place. To allow these launches, set `allowAllUnixSockets`, which opens every Unix socket to sandboxed commands.
123 </Accordion>173 </Accordion>
124</AccordionGroup>174</AccordionGroup>
125 175
129 179
130#### Auto-allow mode180#### Auto-allow mode
131 181
132When a command can be sandboxed, Claude Code runs it inside the sandbox and approves it automatically, without asking your permission. Commands that cannot be sandboxed, such as those needing network access to non-allowed hosts, fall back to the regular permission flow, where Claude Code checks your [permission rules](/docs/en/permissions) and gates any command those rules do not already allow, with a prompt in Manual mode.182Claude Code approves a command automatically, with no prompt, when the command runs inside the sandbox. A command goes through the regular [permission flow](/docs/en/permissions) when it runs outside the sandbox because it matches [`excludedCommands`](#run-commands-outside-the-sandbox-with-excludedcommands) or because Claude [retries it unsandboxed](#the-unsandboxed-retry-escape-hatch).
183
184A sandboxed command that connects to a host you haven't allowed stays in the sandbox. [Hosts outside your allowed domains](#hosts-outside-your-allowed-domains) covers who decides whether the connection goes through.
133 185
134Even in auto-allow mode, the following still apply:186Even in auto-allow mode, the following still apply:
135 187
136* Explicit [deny rules](/docs/en/permissions) are always respected188* Explicit [deny rules](/docs/en/permissions) are always respected
137* `rm` or `rmdir` commands that target a [critical path](/docs/en/permission-modes#critical-paths) still go through the regular permission flow189* `rm` or `rmdir` commands that target a [critical path](/docs/en/permission-modes#critical-paths) still go through the regular permission flow
138* Content-scoped [ask rules](/docs/en/permissions) like `Bash(git push *)` still force a prompt even for sandboxed commands190* Content-scoped [ask rules](/docs/en/permissions) like `Bash(git push *)` still force a prompt even for sandboxed commands
139* A bare `Bash` ask rule, or the equivalent `Bash(*)` form, is skipped for commands that run sandboxed; it still applies to commands that fall back to the regular permission flow. In [plan mode](/docs/en/permission-modes#analyze-before-you-edit-with-plan-mode), the rule isn't skipped: it prompts for sandboxed commands too, including read-only ones. Before v2.1.212, the skip applied in plan mode as well191* A bare `Bash` ask rule, or the equivalent `Bash(*)` form, is skipped for commands that run sandboxed; it still applies to commands that fall back to the regular permission flow. In [plan mode](/docs/en/permission-modes#analyze-before-you-edit-with-plan-mode), the rule isn't skipped: it prompts for sandboxed commands too, including read-only ones
140 192
141<Info>193<Info>
142 Auto-allow mode works independently of your permission mode setting, with three exceptions: [plan mode](/docs/en/permission-modes#analyze-before-you-edit-with-plan-mode), an auto mode command that carries [per-command allowed domains](#per-command-allowed-domains-in-auto-mode), and [server-side classifier review](/docs/en/permission-modes#how-the-classifier-evaluates-actions) of sandboxed commands in auto mode. Even if you're not in "accept edits" mode, sandboxed Bash commands run automatically when auto-allow is enabled. This means Bash commands that modify files within the sandbox boundaries execute without prompting, even in Manual mode, where the file edit tools would prompt.194 Auto-allow mode works independently of your permission mode setting, with three exceptions: [plan mode](/docs/en/permission-modes#analyze-before-you-edit-with-plan-mode), an auto mode command that carries [per-command allowed domains](#per-command-allowed-domains-in-auto-mode), and [server-side classifier review](/docs/en/permission-modes#how-the-classifier-evaluates-actions) of sandboxed commands in auto mode. Even if you're not in "accept edits" mode, sandboxed Bash commands run automatically when auto-allow is enabled. This means Bash commands that modify files within the sandbox boundaries execute without prompting, even in Manual mode, where the file edit tools would prompt.
143 195
144 In plan mode, auto-allow doesn't widen approvals; see [plan mode](/docs/en/permission-modes#analyze-before-you-edit-with-plan-mode) for how Claude Code gates commands while you plan. Before v2.1.212, auto-allow ran sandboxed commands without a prompt in plan mode too.196 In plan mode, auto-allow doesn't widen approvals; see [plan mode](/docs/en/permission-modes#analyze-before-you-edit-with-plan-mode) for how Claude Code gates commands while you plan.
145</Info>197</Info>
146 198
147#### Regular permissions mode199#### Regular permissions mode
150 202
151#### The unsandboxed retry escape hatch203#### The unsandboxed retry escape hatch
152 204
153Some commands can't run inside the sandbox at all, such as tools that are incompatible with it or that need a host you haven't allowed. Claude Code reports sandbox violations in the blocked command's result, naming the path or host the sandbox denied, so Claude sees what the sandbox blocked. Rather than failing the task or requiring you to turn sandboxing off, Claude Code includes an escape hatch: Claude analyzes the violation and may retry the command with the `dangerouslyDisableSandbox` parameter.205The unsandboxed retry is an escape hatch for commands that fail inside the sandbox, such as tools that are incompatible with it. When the sandbox blocks a network connection, Claude Code names the denied host in the command's result, so Claude sees what was blocked. Claude analyzes the failure and may retry the command with the `dangerouslyDisableSandbox` parameter.
206
207The retried command runs unsandboxed. In an interactive terminal session, who approves it depends on your permission mode:
208
209* **`bypassPermissions` mode**: the retry runs without a prompt
210* **Manual mode and `acceptEdits` mode**: you get a prompt titled "Bash command (unsandboxed)"
211* **[Auto mode](/docs/en/permission-modes#eliminate-prompts-with-auto-mode)**: a separate classifier model evaluates the underlying command
212* **`dontAsk` mode**: Claude Code denies the retry
213* **Plan mode**: see [how Claude Code gates commands while you plan](/docs/en/permission-modes#analyze-before-you-edit-with-plan-mode)
214
215These rules and settings change who approves the retry:
216
217* **A matching allow rule**: if an allow rule such as `Bash(curl *)` matches the command, it also approves the retry, so the command runs outside the sandbox with no prompt
218* **An ask rule for the parameter**: add an [ask rule](/docs/en/permissions#match-by-input-parameter) for `Bash(dangerouslyDisableSandbox:true)` to be prompted on Bash retries. You get the prompt in auto mode and `bypassPermissions` mode too, and the rule takes precedence over a matching allow rule
219* **[`permissions.blockReadsOutsideWorkingDirectories`](/docs/en/settings-reference#permissions-blockreadsoutsideworkingdirectories)**: [Actions no mode auto-approves](/docs/en/permission-modes#actions-no-mode-auto-approves) covers the retries that prompt while it's on
220
221#### Turn off the retry with strict sandbox mode
222
223You can disable the unsandboxed retry by setting `"allowUnsandboxedCommands": false` in your [sandbox settings](/docs/en/settings-reference#sandbox-settings). With the retry disabled, Claude Code ignores the `dangerouslyDisableSandbox` parameter. While the sandbox is running, commands Claude runs are then sandboxed unless they match an `excludedCommands` entry. To keep Claude Code from running commands unsandboxed when the sandbox can't start, also set [`failIfUnavailable`](/docs/en/settings-reference#sandbox-failifunavailable). The `/sandbox` **Overrides** tab shows this setting as **Strict sandbox mode**.
154 224
155The retried command runs outside the sandbox, so it goes through the regular permission flow. In Manual mode you get a confirmation prompt. In [auto mode](/docs/en/permission-modes#eliminate-prompts-with-auto-mode), the classifier evaluates the underlying command. While [`permissions.blockReadsOutsideWorkingDirectories`](/docs/en/settings-reference#permissions-blockreadsoutsideworkingdirectories) is on, a retry that needs approval to run outside the sandbox prompts you instead. To be prompted on every unsandboxed retry even in auto mode, add an [ask rule](/docs/en/permissions#match-by-input-parameter) for `Bash(dangerouslyDisableSandbox:true)`.225A `false` in your user settings, `--settings`, or managed settings holds even when a project's settings set `true`. A `false` in your user settings doesn't make the sandbox admin-required, so a project's other sandbox settings still apply. Before v2.1.285, a project's `true` overrode a `false` in your user settings.
156 226
157You can disable this escape hatch by setting `"allowUnsandboxedCommands": false` in your [sandbox settings](/docs/en/settings-reference#sandbox-settings). With the escape hatch disabled, Claude Code ignores the `dangerouslyDisableSandbox` parameter, and every command Claude runs must run sandboxed unless you've listed it in `excludedCommands`. The `/sandbox` **Overrides** tab shows this setting as **Strict sandbox mode**.227If you or your administrator disable the retry in managed settings or with the `--settings` flag, the sandbox becomes admin-required. Claude Code then ignores the settings in a repository's files that loosen the sandbox, `excludedCommands` entries included. [Repository settings under an admin-required sandbox](#repository-settings-under-an-admin-required-sandbox) lists them.
158 228
159Strict sandbox mode applies to the commands Claude runs. Commands you type yourself at the [`!` shell-mode prompt](/docs/en/interactive-mode#shell-mode-with-prefix) run outside the sandbox unless the session is one of these:229Strict sandbox mode applies to the commands Claude runs. Commands you type yourself at the [`!` shell-mode prompt](/docs/en/interactive-mode#shell-mode-with-prefix) run outside the sandbox unless the session is one of these:
160 230
188 258
189These paths are enforced at the OS level, so all commands running inside the sandbox, including their child processes, respect them. This is the recommended approach when a tool needs write access to a specific location, rather than excluding the tool from the sandbox entirely with `excludedCommands`.259These paths are enforced at the OS level, so all commands running inside the sandbox, including their child processes, respect them. This is the recommended approach when a tool needs write access to a specific location, rather than excluding the tool from the sandbox entirely with `excludedCommands`.
190 260
191When you define the same filesystem array in multiple [settings scopes](/docs/en/settings#settings-precedence), Claude Code merges them, combining paths from every scope rather than replacing one scope's array with another's.261When you define the same filesystem array in multiple [settings scopes](/docs/en/settings#settings-precedence), Claude Code merges them, combining the paths rather than replacing one scope's array with another's. Claude Code leaves an entry out of the merge when a lock under [Keep developers from widening the policy](#keep-developers-from-widening-the-policy) covers it.
192 262
193If you exclude a source with [`--setting-sources`](/docs/en/cli-reference) on the CLI or [`settingSources`](/docs/en/agent-sdk/claude-code-features#control-filesystem-settings-with-settingsources) in the Agent SDK, Claude Code ignores its `sandbox.filesystem` entries, its `Edit` permission rules, and its `Read` deny rules when building the sandbox configuration. Requires Claude Code v2.1.246 or later.263If you exclude a source with [`--setting-sources`](/docs/en/cli-reference) on the CLI or [`settingSources`](/docs/en/agent-sdk/claude-code-features#control-filesystem-settings-with-settingsources) in the Agent SDK, Claude Code ignores its `sandbox.filesystem` entries, its `Edit` permission rules, and its `Read` deny rules when building the sandbox configuration. Requires Claude Code v2.1.246 or later.
194 264
195When you edit these filesystem lists during a session, Claude Code [applies the change to the running session](/docs/en/settings#when-edits-take-effect), so the next sandboxed command runs under the new paths.265When you edit these filesystem lists during a session, Claude Code [applies the change to the running session](/docs/en/settings#when-edits-take-effect), so the next sandboxed command runs under the new paths.
196 266
197Path prefixes control how paths are resolved:267Sandbox filesystem paths use standard conventions: `/tmp/build` is absolute and `~/.kube` is relative to your home directory. This differs from [Read and Edit permission rules](/docs/en/permissions#read-and-edit), which use `//path` for absolute and `/path` for project-relative. For relative paths, trailing slashes, and wildcards, see [Sandbox path prefixes](/docs/en/settings-reference#sandbox-path-prefixes).
198
199| Prefix | Meaning | Example |
200| :- | :- | :- |
201| `/` | Absolute path from filesystem root | `/tmp/build` stays `/tmp/build` |
202| `~/` | Relative to home directory | `~/.kube` becomes `$HOME/.kube` |
203| `./` or no prefix | Relative to the project root for project settings, or to `~/.claude` for user settings | `./output` in `.claude/settings.json` resolves to `<project-root>/output` |
204
205This syntax differs from [Read and Edit permission rules](/docs/en/permissions#read-and-edit), which use `//path` for absolute and `/path` for project-relative. Sandbox filesystem paths use standard conventions: `/tmp/build` is absolute. For how Claude Code treats a trailing slash or a wildcard in these paths, see [Sandbox path prefixes](/docs/en/settings-reference#sandbox-path-prefixes).
206 268
207You can also deny write or read access using `sandbox.filesystem.denyWrite` and `sandbox.filesystem.denyRead`, and re-allow specific paths within a denied region using `sandbox.filesystem.allowRead`. When read rules overlap, the rule with the narrower path applies:269You can also deny write or read access using `sandbox.filesystem.denyWrite` and `sandbox.filesystem.denyRead`, and re-allow specific paths within a denied region using `sandbox.filesystem.allowRead`. When read rules overlap, the rule with the narrower path applies:
208 270
212| `"allowRead": ["~/"]` with `"denyRead": ["~/.env"]` | `~/.env` stays blocked and the rest of the home directory is readable. The deny holds inside a wider allow, so a broad allow can't silently re-expose a secret |274| `"allowRead": ["~/"]` with `"denyRead": ["~/.env"]` | `~/.env` stays blocked and the rest of the home directory is readable. The deny holds inside a wider allow, so a broad allow can't silently re-expose a secret |
213| `"allowRead": ["~/"]` with `"denyRead": ["~/**/.env"]` | Every `.env` under the home directory stays blocked and the rest is readable. A [wildcard deny](/docs/en/settings-reference#sandbox-path-prefixes) holds inside a wider allow the same way an exact path does |275| `"allowRead": ["~/"]` with `"denyRead": ["~/**/.env"]` | Every `.env` under the home directory stays blocked and the rest is readable. A [wildcard deny](/docs/en/settings-reference#sandbox-path-prefixes) holds inside a wider allow the same way an exact path does |
214 276
215The example below blocks reading from the entire home directory while still allowing reads from the current project. Place it in your project's `.claude/settings.json`, because the relative path `.` resolves to the project root only when the configuration lives in project settings:277The example below blocks reading from the entire home directory while still allowing reads from the current project. Place it in your project's `.claude/settings.json`, because the relative path `.` resolves to the project root when the configuration lives in project settings:
216 278
217```json theme={null}279```json theme={null}
218{280{
230 292
231To deny sandboxed commands read access to home directories and mounted volumes while keeping the working directories readable, set [`permissions.blockReadsOutsideWorkingDirectories`](/docs/en/settings-reference#permissions-blockreadsoutsideworkingdirectories) instead of writing path rules.293To deny sandboxed commands read access to home directories and mounted volumes while keeping the working directories readable, set [`permissions.blockReadsOutsideWorkingDirectories`](/docs/en/settings-reference#permissions-blockreadsoutsideworkingdirectories) instead of writing path rules.
232 294
295### Run commands outside the sandbox with `excludedCommands`
296
297List a command pattern in [`sandbox.excludedCommands`](/docs/en/settings-reference#sandbox-excludedcommands) to run matching commands outside the sandbox, which means no filesystem restrictions and no network proxy. Use it for a tool that can't work inside the sandbox and that you trust with your full access. A tool that needs one more directory or one more host may work with `allowWrite` or `allowedDomains`, which keep the command sandboxed.
298
299This example takes `docker compose` commands out of the sandbox. Save it in `~/.claude/settings.json` to apply it to all of your projects:
300
301```json theme={null}
302{
303 "sandbox": {
304 "enabled": true,
305 "excludedCommands": ["docker compose *"]
306 }
307}
308```
309
310Claude Code checks your entries against each Bash and Monitor call. A call is the whole command line Claude sends, which can chain several commands. The following rules decide whether a call leaves the sandbox:
311
312* **End the pattern with ` *`**: entries use the same syntax as a `Bash(...)` [permission rule](/docs/en/permissions#permission-rule-syntax), where a pattern with no wildcard is an exact match. `docker` matches only `docker` with no arguments. `docker *` matches `docker` with or without arguments
313* **Every command in the call has to match**: `npm ci && docker compose build` stays sandboxed unless another entry covers `npm ci`
314* **Claude Code matches the text of the call**: a script or `make` target that calls `docker` internally doesn't match, and neither does `/usr/local/bin/docker`
315* **Some calls stay sandboxed**: a redirect to a file, a `cd`, or a command substitution such as `$(...)` keeps the whole call sandboxed. The [reference entry](/docs/en/settings-reference#sandbox-excludedcommands) lists more calls that stay sandboxed
316* **Where you save the entry can matter**: while the sandbox is [admin-required](#repository-settings-under-an-admin-required-sandbox), Claude Code ignores entries in `.claude/settings.json` and `.claude/settings.local.json`
317
318An excluded command goes through the regular permission flow:
319
320* [Read-only commands](/docs/en/permissions#read-only-commands) and commands your allow rules cover run without a prompt
321* In auto mode, the classifier reviews other excluded commands
322* In `bypassPermissions` mode, an excluded command runs without a prompt unless an ask rule matches it
323
324To confirm an entry matches, switch to Manual mode and ask Claude to run a matching command that changes something, such as `docker compose up -d`. The permission prompt is titled "Bash command (unsandboxed)".
325
326<Warning>
327 An excluded command runs with your full access. A broad entry such as `docker *` covers everything that tool can do. If you write a pattern that covers an interpreter, a script inside your working directory, or a tool that acts on a file there, as `docker compose` does with its compose file, Claude can write that file and then run it outside the sandbox. A narrower pattern leaves less that Claude can run outside the sandbox.
328</Warning>
329
233### Disable filesystem isolation330### Disable filesystem isolation
234 331
235Set `sandbox.filesystem.disabled` to `true` to skip filesystem isolation while keeping network isolation. The example below turns off filesystem isolation while keeping an allowlist of network domains:332Set `sandbox.filesystem.disabled` to `true` to skip filesystem isolation while keeping network isolation. The example below turns off filesystem isolation while keeping an allowlist of network domains:
250 347
251The sandbox has two independent layers: [filesystem isolation](#filesystem-isolation) controls which paths sandboxed commands can read and write, and [network isolation](#network-isolation) controls which domains they can reach. With the filesystem layer off, sandboxed commands get unrestricted read and write access to the host filesystem, while their network egress stays confined to your allowed domains. Turn the layer off when you sandbox to control where commands connect rather than what they write.348The sandbox has two independent layers: [filesystem isolation](#filesystem-isolation) controls which paths sandboxed commands can read and write, and [network isolation](#network-isolation) controls which domains they can reach. With the filesystem layer off, sandboxed commands get unrestricted read and write access to the host filesystem, while their network egress stays confined to your allowed domains. Turn the layer off when you sandbox to control where commands connect rather than what they write.
252 349
253The setting is off by default and applies on the platforms where the sandbox runs: macOS, Linux, and WSL2. Requires Claude Code v2.1.216 or later.350`sandbox.filesystem.disabled` defaults to `false`. Requires Claude Code v2.1.216 or later.
254 351
255<Warning>352<Warning>
256 With filesystem isolation off and commands auto-allowed, a sandboxed command can write files that later commands run or read, such as shell startup files, executables on `$PATH`, or `~/.claude/settings.json`, and use them to widen its own access on the next run. Set `filesystem.disabled` to `true` only for workloads you trust not to escalate their own access. Locking network domains with [`allowManagedDomainsOnly`](#keep-developers-from-widening-the-policy) narrows the risk but doesn't remove it, since that lock applies only to commands running inside the sandbox.353 With filesystem isolation off and commands auto-allowed, a sandboxed command can write files that later commands run or read, such as shell startup files, executables on `$PATH`, or `~/.claude/settings.json`, and use them to widen its own access on the next run. Set `filesystem.disabled` to `true` only for workloads you trust not to escalate their own access. Locking network domains with [`allowManagedDomainsOnly`](#keep-developers-from-widening-the-policy) narrows the risk but doesn't remove it, since that lock applies only to commands running inside the sandbox.
264* When managed settings configure `sandbox.filesystem` at all, or list any `sandbox.credentials.files` entry with `"mode": "deny"`, only managed settings can set the key. This keeps administrator-deployed filesystem restrictions in force; to relax such a deployment, set `"disabled": true` in managed settings.361* When managed settings configure `sandbox.filesystem` at all, or list any `sandbox.credentials.files` entry with `"mode": "deny"`, only managed settings can set the key. This keeps administrator-deployed filesystem restrictions in force; to relax such a deployment, set `"disabled": true` in managed settings.
265* When [`CLAUDE_CODE_SUBPROCESS_ENV_SCRUB`](/docs/en/env-vars) is set, Claude Code ignores `filesystem.disabled` from every source, including managed settings, and keeps filesystem isolation on.362* When [`CLAUDE_CODE_SUBPROCESS_ENV_SCRUB`](/docs/en/env-vars) is set, Claude Code ignores `filesystem.disabled` from every source, including managed settings, and keeps filesystem isolation on.
266 363
267Whether a managed `credentials.files` entry pins `filesystem.disabled`, locking the key to managed settings so developers can't turn filesystem isolation off, depends on the entry's `mode` and what happens to the entry when the sandbox starts:364A [valid](/docs/en/settings-reference#invalid-credential-entries-in-managed-settings) `mask` entry doesn't lock the key, even when Claude Code [falls back to `deny`](#mask-credential-files) for it at startup. List a path that can't be masked, such as a credential directory, as an explicit `deny` entry in managed settings, which locks the key.
268
269| Managed entry | Pins `filesystem.disabled` | What protects the file when isolation is off |
270| - | - | - |
271| `"mode": "deny"` | Yes | Nothing: the read block is part of the filesystem layer |
272| `"mode": "mask"`, applied as a mask | No | Masking itself: the [sentinel copy and proxy](#mask-credential-files) on Linux and WSL2, the sandbox's own read rules on macOS |
273| `"mode": "mask"`, [fallen back to `deny`](#mask-credential-files) at setup | No | Nothing, same as `deny`. List a path that can't be masked, such as a directory, as an explicit `deny` entry, which pins the key |
274| `"mode": "mask"`, [degraded to `deny` by validation](/docs/en/managed-settings#invalid-entries-in-managed-settings) | Yes, like an explicit `deny` | Nothing, same as `deny` |
275
276A fallback happens when the sandbox starts, after Claude Code has already read the settings the pin check runs on, so a fallen-back entry never pins. Validation rewrites an invalid entry to `deny` while settings load, so a degraded entry pins like one you wrote as `deny`.
277 365
278#### What changes when filesystem isolation is off366#### What changes when filesystem isolation is off
279 367
327When you [exclude a settings source](#configure-sandboxing):415When you [exclude a settings source](#configure-sandboxing):
328 416
329* **Project or local settings**: Claude Code applies none of their `credentials` entries. Requires Claude Code v2.1.246 or later.417* **Project or local settings**: Claude Code applies none of their `credentials` entries. Requires Claude Code v2.1.246 or later.
330* **User settings**: Claude Code still applies the `deny` entries in `~/.claude/settings.json` and keeps its [file `mask` entries](#mask-credential-files) as restrictions, but drops its [environment variable `mask` entries](#mask-environment-variables).418* **User settings**: Claude Code still applies the `deny` entries in `~/.claude/settings.json` and keeps its [file `mask` entries](#mask-credential-files) as restrictions that no longer authorize the proxy to substitute the real value, but drops its [environment variable `mask` entries](#mask-environment-variables).
331 419
332There is no built-in credential deny list, so only the files and variables you list are restricted.420There is no built-in credential deny list, so only the files and variables you list are restricted.
333 421
335 423
336### Mask credentials424### Mask credentials
337 425
338Masking goes further than a `deny` entry under [Protect credentials](#protect-credentials). Instead of blocking a credential, Claude Code shows sandboxed commands a placeholder, the sentinel, and the [sandbox proxy](#network-isolation) swaps in the real value on outbound requests to hosts you allow. For files, the substitution is Linux and WSL2 behavior; [macOS blocks the file instead](#mask-credential-files).426When you mask a credential, Claude Code shows sandboxed commands a per-session placeholder called the sentinel, and the [sandbox proxy](#network-isolation) substitutes the real value on outbound requests to hosts you allow. A `deny` entry under [Protect credentials](#protect-credentials) blocks the credential instead. For files on macOS, Claude Code [blocks the file instead](#mask-credential-files) of masking it.
339
340#### Mask environment variables
341
342`"mode": "mask"` protects a credential while keeping the tools that authenticate with it working. `deny` removes the variable entirely, which also breaks tools that need it, such as `gh` or `npm`. Requires Claude Code v2.1.199 or later.
343 427
344With `mask`, the sandboxed command sees a per-session sentinel value instead of the real one. Each `mask` entry can list `injectHosts`, the hosts the real value is allowed to reach. When a request leaves the sandbox for one of them, the [sandbox proxy](#network-isolation) replaces the sentinel with the real value. The command and anything it logs never hold the real credential, but its requests still authenticate.428Masking environment variables requires Claude Code v2.1.199 or later. The [`sandbox.credentials`](/docs/en/settings-reference#sandbox-credentials) reference lists every field.
345 429
346The proxy substitutes the credential inside request contents, so it has to see them. Set [`network.tlsTerminate`](/docs/en/settings-reference#sandbox-network-tlsterminate) so the proxy terminates TLS itself.430Masking requires the following:
347 431
348Without it, masking fails without exposing anything: the command still sees only the sentinel, but the sentinel reaches the server unchanged and authentication fails. Claude Code reports this misconfiguration at startup.432* **TLS termination**: the proxy substitutes the real value inside request contents, so it has to see them. Set [`network.tlsTerminate`](/docs/en/settings-reference#sandbox-network-tlsterminate) so the proxy terminates TLS itself. Without it, masking fails without exposing anything: the command still sees only the sentinel, but the sentinel reaches the server unchanged and authentication fails. Claude Code reports this misconfiguration at startup.
433* **An allowed destination**: each `mask` entry can list `injectHosts`, the hosts the real value is allowed to reach. The proxy injects only on connections the [domain allowlist](#network-isolation) admits, so each `injectHosts` host must also be reachable through `network.allowedDomains`. For a `mask` entry with no `injectHosts`, the proxy substitutes the real value on requests to every host in `network.allowedDomains`.
434* **A trusted settings scope**: masking authorizes the proxy to send your real credential somewhere, so Claude Code honors `mask` entries, `network.tlsTerminate`, [`credentials.allowPlaintextInject`](/docs/en/settings-reference#sandbox-credentials-allowplaintextinject), `awsPairs`, and `sigv4` only from user settings, managed settings, and the `--settings` flag. It ignores them in a repository's `.claude/settings.json` or `.claude/settings.local.json`. When your administrator delivers `mask` entries, `network.tlsTerminate`, or `credentials.allowPlaintextInject` through server-managed settings, they count as [settings that need approval](/docs/en/server-managed-settings#security-approval-dialogs).
349 435
350Substitution covers headers and request bodies. Requests that authenticate with a signature derived from the credential, rather than the credential itself, need re-signing at the proxy; [Re-sign AWS requests](#re-sign-aws-requests) covers how that works for AWS.436#### Mask environment variables
351 437
352The proxy injects only on connections the [domain allowlist](#network-isolation) admits, so each `injectHosts` destination must also be reachable through `network.allowedDomains`.438To mask an environment variable, set `"mode": "mask"` on its `credentials.envVars` entry. The command and anything it logs never hold the real credential, but its requests still authenticate. When the same variable is listed with `deny` in any scope, `deny` takes precedence.
353 439
354The example below masks two tokens. `GH_TOKEN` is substituted only on requests to `api.github.com`, while `NPM_TOKEN` has no `injectHosts` and is substituted on requests to every host in `network.allowedDomains`.440This example masks two tokens. `GH_TOKEN` is substituted only on requests to `api.github.com`, while `NPM_TOKEN` has no `injectHosts` and is substituted on requests to every host in `network.allowedDomains`:
355 441
356```json theme={null}442```json theme={null}
357{443{
371}457}
372```458```
373 459
374<span id="ipv6-destinations-in-injecthosts" />Spell an IPv6 destination differently in the two lists, because each list has its own matcher:460Masking replaces the whole value by default. For a value with structure, such as a `DATABASE_URL` connection string or a JWT, use the [`extract`, `decode`, `maskClaims`, and `onExtractNoMatch` fields](/docs/en/settings-reference#sandbox-credentials-envvars) so tools that parse the value keep working.
375
376* **`network.allowedDomains`**: the [bracketed form domain lists use](#ipv6-addresses-in-domain-lists), such as `"[::1]"`. The proxy checks this list to admit the connection.
377* **`injectHosts`**: the bare address in its canonical compressed form, such as `"::1"` or `"2001:db8::1"`. The proxy matches each entry against the connection's bare destination address, ignoring ports, so a bracketed, zone-ID, or differently compressed spelling never matches and the proxy never injects the credential there.
378
379`claude doctor` flags `injectHosts` entries that can never match with the warning `Sandbox credential injectHosts entries can never match their destination`. This check requires Claude Code v2.1.229 or later.
380
381Unlike `deny`, masking authorizes the proxy to send your real credential to the listed hosts, so Claude Code honors it only from settings you or your administrator control: user settings, managed settings, and the `--settings` CLI flag. Claude Code ignores `mask` entries in a repository's `.claude/settings.json` or `.claude/settings.local.json`. In those files it also ignores `network.tlsTerminate` and [`credentials.allowPlaintextInject`](/docs/en/settings-reference#sandbox-credentials-allowplaintextinject), the setting that lets the proxy inject credentials into unencrypted requests. If you [exclude user settings](#configure-sandboxing), Claude Code drops the environment variable `mask` entries in `~/.claude/settings.json` too.
382 461
383When your administrator delivers `mask` entries, `network.tlsTerminate`, or `credentials.allowPlaintextInject` through server-managed settings, they count as [settings that need approval](/docs/en/server-managed-settings#security-approval-dialogs).462<span id="ipv6-destinations-in-injecthosts" />For an IPv6 destination, spell the address differently in the two lists:
384 463
385When the same variable is listed with `deny` in any scope, `deny` takes precedence.464* **`network.allowedDomains`**: the bracketed form, such as `"[::1]"`
465* **`injectHosts`**: the bare address in its canonical compressed form, such as `"::1"`
386 466
387Masking replaces the variable's entire value by default, which suits a bare token. Optional entry fields, which require Claude Code v2.1.224 or later, handle values with structure:467The proxy matches each `injectHosts` entry against the connection's bare destination address, ignoring ports, so a bracketed, zone-ID, or differently compressed spelling never matches. `claude doctor` flags entries that can never match with the warning `Sandbox credential injectHosts entries can never match their destination`. This check requires Claude Code v2.1.229 or later.
388
389* `extract`: a regular expression Claude Code applies across the value, replacing only the text captured by group 1 of each match, so a tool that parses the value, such as a `DATABASE_URL` connection string, still works inside the sandbox. The pattern must contain at least one capturing group.
390* `onExtractNoMatch` controls what happens when the pattern matches nothing:
391 * `warn`, the default, warns and passes the variable through unmasked
392 * `deny` unsets the variable inside the sandbox
393 * `error` stops sandbox setup until you fix the configuration
394* `decode: "jwt"`: for a variable holding a JSON Web Token (JWT). Claude Code verifies the value is a JWT and replaces it with a structurally valid fake token, so code inside the sandbox that decodes the token keeps working. Add `maskClaims` to list top-level payload claims to mask individually instead of replacing the whole token; the other claims stay readable. When the value doesn't verify as a JWT, or no listed claim matches, Claude Code passes the variable through unmasked with a warning. `decode` can't be combined with `extract`.
395
396See the [`credentials.envVars[]` rows in the settings reference](/docs/en/settings-reference#sandbox-settings) for the full field list.
397 468
398#### Re-sign AWS requests469#### Re-sign AWS requests
399 470
400AWS requests carry SigV4 signatures over the request contents, so mask `AWS_ACCESS_KEY_ID` and `AWS_SECRET_ACCESS_KEY` together. The proxy detects a SigV4 request by the access key's sentinel and re-signs it after substituting the real values. Masking the secret alone leaves requests signed with the placeholder, which the proxy can't detect, so they fail at AWS; Claude Code warns about this case at startup, but not when only the access key ID is masked. A detected request the proxy can't re-sign, such as one missing its `x-amz-date` header, fails with a proxy error instead of reaching the server with a broken signature.471AWS requests carry SigV4 signatures over the request contents, so mask `AWS_ACCESS_KEY_ID` and `AWS_SECRET_ACCESS_KEY` together. The proxy detects a SigV4 request by the access key's [sentinel](#mask-credentials) and re-signs the request with the real values, which requires Claude Code v2.1.221 or later. If you mask only the secret, requests are signed with a placeholder the proxy can't detect, so they fail at AWS.
401 472
402Claude Code links the conventional `AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY`, and `AWS_SESSION_TOKEN` variables into one credential automatically when you mask their whole values. If your AWS credential lives in variables with other names, group them yourself with [`credentials.awsPairs`](/docs/en/settings-reference#sandbox-credentials-awspairs), which requires Claude Code v2.1.224 or later. This example adds the pairing to a configuration that already masks `MY_KEY_ID`, `MY_SECRET_KEY`, and `MY_SESSION_TOKEN` whole-value, as in the [masking configuration above](#mask-environment-variables):473Claude Code links the conventional `AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY`, and `AWS_SESSION_TOKEN` variables into one credential automatically when you mask their whole values. If your AWS credential is in variables with other names, group them with [`credentials.awsPairs`](/docs/en/settings-reference#sandbox-credentials-awspairs), which requires Claude Code v2.1.224 or later.
403 474
404```json theme={null}475Streaming uploads, presigned URLs, and SigV4A requests carry signatures the proxy can't recompute. When one of these requests is signed with a masked pair's placeholder, the proxy fails it rather than forward a broken signature. Requests signed with unmasked credentials aren't affected. Use [`credentials.sigv4`](/docs/en/settings-reference#sandbox-credentials-sigv4), which requires Claude Code v2.1.224 or later, to forward one of these request forms instead. AWS still rejects the request, so the calling tool receives AWS's own rejection response instead of a proxy error.
405{
406 "sandbox": {
407 "credentials": {
408 "awsPairs": [
409 {
410 "accessKeyIdVar": "MY_KEY_ID",
411 "secretAccessKeyVar": "MY_SECRET_KEY",
412 "sessionTokenVar": "MY_SESSION_TOKEN"
413 }
414 ]
415 }
416 }
417}
418```
419
420Each entry follows these rules:
421
422* `accessKeyIdVar` and `secretAccessKeyVar` name the masked `envVars` entries holding the access key ID and the secret key. The optional `sessionTokenVar` names the entry holding the session token for temporary credentials; when set, the proxy sends the real token as `x-amz-security-token` on re-signed requests.
423* Each named variable must be a `mask` entry that masks its entire value, without `extract` or `decode`.
424* The proxy re-signs requests on the hosts listed in the access key ID entry's `injectHosts`.
425* Naming any of the conventional variables in a pair replaces the automatic pairing.
426
427Like `mask` entries, `awsPairs` is honored only from user settings, managed settings, and the `--settings` CLI flag.
428
429Three AWS request forms carry signatures the proxy can't recompute. When such a request is signed with a masked pair's placeholder, the proxy fails it rather than forward a broken signature; requests signed with unmasked credentials are never affected. The [`credentials.sigv4`](/docs/en/settings-reference#sandbox-credentials-sigv4) setting, which requires Claude Code v2.1.224 or later, relaxes this per form: setting a form's key to `passthrough` forwards the request with its placeholder-derived signature, so the calling tool receives AWS's own rejection response instead of a proxy error. Like `awsPairs`, `sigv4` is honored only from user settings, managed settings, and the `--settings` CLI flag.
430
431| Request form | `sigv4` key | Why the proxy can't re-sign it |
432| :- | :- | :- |
433| aws-chunked streaming uploads | `streaming` | Per-chunk signatures chain off the seed signature, so re-signing would require rewriting the body |
434| Presigned URLs | `presigned` | The signature lives in the URL itself, with no `Authorization` header |
435| SigV4A asymmetric signatures | `sigv4a` | There is no shared-key HMAC to recompute |
436 476
437#### Mask credential files477#### Mask credential files
438 478
439File entries also accept `"mode": "mask"`, which requires Claude Code v2.1.221 or later. What a sandboxed command sees depends on the platform:479To mask a credential file, set `"mode": "mask"` on its `credentials.files` entry. Masking files requires Claude Code v2.1.221 or later. What a sandboxed command sees depends on the platform:
440
441* **Linux and WSL2**: sandboxed commands read a sentinel copy of the file, a stand-in whose secret is replaced with a placeholder value, and the [sandbox proxy](#network-isolation) substitutes the real value on egress.
442* **macOS**: sandboxed commands can't read the listed file at all. Claude Code builds no sentinel copy and substitutes nothing on egress, so tools that authenticate with the file don't work inside the sandbox, the same effect as `deny`. Unlike a `deny` entry, the read block holds even when you [disable filesystem isolation](#disable-filesystem-isolation).
443 480
444On every platform, Claude Code applies the [`network.tlsTerminate`](/docs/en/settings-reference#sandbox-network-tlsterminate) requirement and `injectHosts` the same way as for [masked environment variables](#mask-environment-variables), and ignores repository settings the same way. If you [exclude user settings](#configure-sandboxing), Claude Code keeps the file `mask` entries in `~/.claude/settings.json` as restrictions, but the entries no longer authorize the proxy to substitute the real value.481* **Linux and WSL2**: sandboxed commands read a [sentinel](#mask-credentials) copy of the file, and the proxy substitutes the real value on outbound requests.
482* **macOS**: sandboxed commands can't read the file at all. Claude Code builds no sentinel copy, so tools that authenticate with the file don't work inside the sandbox, the same effect as `deny`. The read block holds even when you [disable filesystem isolation](#disable-filesystem-isolation).
445 483
446The example below masks a GitHub token stored in `~/.config/gh/hosts.yml`; the `extract` pattern, covered below, tells Claude Code which part of the file is the secret. On Linux and WSL2, sandboxed commands that read the file get a sentinel in place of the token, and the proxy substitutes the real token on requests to `api.github.com`:484This example masks a GitHub token stored in `~/.config/gh/hosts.yml`. The `extract` pattern marks which part of the file is the secret, so on Linux and WSL2 `gh` still parses the rest of its config:
447 485
448```json theme={null}486```json theme={null}
449{487{
467}505}
468```506```
469 507
470To confirm the mask is active, ask Claude to run `cat ~/.config/gh/hosts.yml` in a sandboxed command: on Linux and WSL2 the output shows a sentinel value in place of the token, and on macOS the read fails instead.508To confirm the mask is active, ask Claude to run `cat ~/.config/gh/hosts.yml` in a sandboxed command. On Linux and WSL2 the output shows a sentinel in place of the token, and on macOS the read fails.
471
472On Linux and WSL2, the `extract` pattern is what keeps the rest of `hosts.yml` readable. Claude Code applies the regular expression across the whole file and replaces only the text captured by group 1 of each match, so `gh` still parses its config and only the token is a placeholder. Use `extract` for any structured file that tools parse, such as `.netrc`, JSON, or YAML; the pattern must contain at least one capturing group. Without `extract`, Claude Code replaces the entire file content with one sentinel value, which suits a file that holds a single bare secret and nothing else.
473
474For a file that holds a JSON Web Token (JWT), set `decode: "jwt"` instead of, or together with, `extract`. `decode` requires Claude Code v2.1.224 or later. Claude Code finds JWT candidates with a built-in pattern, or with your `extract` pattern when set, verifies each candidate is a JWT, and replaces it with a structurally valid fake token, so code that decodes the token inside the sandbox keeps working. Add `maskClaims` to mask only the named top-level payload claims inside each verified token and leave the other claims readable. When no candidate verifies, or no named claim matches, the `onExtractNoMatch` field below governs the outcome, as it does for a pattern that matches nothing.
475
476Two optional fields refine how matching behaves. Both apply only when `mode` is `mask` and `extract` or `decode` is set. On macOS, Claude Code applies `mask` entries as `deny` before the pattern runs whenever filesystem isolation is on, so these fields, and the no-match outcomes below, take effect there only when [filesystem isolation is off](#disable-filesystem-isolation):
477 509
478* `onExtractNoMatch` controls what happens when matching finds nothing to mask in the file:510Without `extract` or `decode`, Claude Code replaces the entire file with one sentinel, which suits a file holding a single bare secret. Use the [`extract`, `decode`, `maskClaims`, `onExtractNoMatch`, and `maskDuplicates` fields](/docs/en/settings-reference#sandbox-credentials-files) to control partial masking and what happens when the pattern matches nothing.
479 511
480 * `warn`, the default, warns and skips the entry, so sandboxed commands can read the real file unmasked. The default suits credentials that may be legitimately absent; if the secret might be present but the pattern might miss it, use `deny`512<Warning>
481 * `deny` makes the file unreadable instead513 When matching finds nothing to mask, the default `onExtractNoMatch` value, `warn`, skips the entry, so sandboxed commands can read the real file unmasked. On macOS, Claude Code applies `mask` entries as `deny` before the pattern runs whenever filesystem isolation is on, so the no-match outcomes take effect there only when [filesystem isolation is off](#disable-filesystem-isolation). The default suits credentials that may be legitimately absent. If the secret might be present but the pattern might miss it, use [`deny`](/docs/en/settings-reference#mask-fields-for-files).
482 * `error` stops sandbox setup until you fix the configuration514</Warning>
483
484 Claude Code treats `deny` as `error` whenever the read block wouldn't be enforced: when you [disable filesystem isolation](#disable-filesystem-isolation), and when a `filesystem.allowRead` entry from any settings source re-opens the file's path.
485* `maskDuplicates` also replaces verbatim copies of each masked credential value, an `extract` capture or a `decode`-verified token, found outside the matched spans, for a secret repeated where matching doesn't reach. It matches raw substrings, so a short or common value would be replaced everywhere it appears; reserve it for long, high-entropy secrets. Default: false.
486 515
487`mask` applies to a single file, so list each credential file individually. Claude Code falls back to `deny` for a `mask` entry it can't mask safely: a directory path, a glob pattern, a file larger than 8 MiB, or a file that isn't UTF-8 text. Write directories as explicit `deny` entries instead; the table under [Which settings can disable it](#which-settings-can-disable-it) covers whether each form pins `filesystem.disabled` and how it behaves with filesystem isolation off.516`mask` applies to a single file, so list each credential file individually. Claude Code falls back to `deny` for a `mask` entry it can't mask safely: a directory path, a glob pattern, a file larger than 8 MiB, or a file that isn't UTF-8 text.
488 517
489## How sandboxing works518## How sandboxing works
490 519
493The sandboxed Bash tool restricts file system access to specific directories:522The sandboxed Bash tool restricts file system access to specific directories:
494 523
495* **Default write behavior**: read and write access to the current working directory and its subdirectories, any directories you've added with `--add-dir`, `/add-dir`, or [`permissions.additionalDirectories`](/docs/en/settings-reference#permissions-additionaldirectories), plus the per-user temp directory that `$TMPDIR` points to524* **Default write behavior**: read and write access to the current working directory and its subdirectories, any directories you've added with `--add-dir`, `/add-dir`, or [`permissions.additionalDirectories`](/docs/en/settings-reference#permissions-additionaldirectories), plus the per-user temp directory that `$TMPDIR` points to
496* **Default read behavior**: read access to the entire computer, except certain denied directories. Note that this default still allows reading credential files such as `~/.aws/credentials` and `~/.ssh/`. Use [`sandbox.credentials`](#protect-credentials) to block reads of these files and unset secret environment variables, or add the paths to `denyRead`.525* **Default read behavior**: read access to the entire computer, except certain denied directories. This default still allows reading credential files, so [protect credentials](#protect-credentials) you don't want commands to read.
497* **Read block**: with [`permissions.blockReadsOutsideWorkingDirectories`](/docs/en/settings-reference#permissions-blockreadsoutsideworkingdirectories) on, sandboxed commands also lose read access to your home directory and the other directories that hold user files, apart from the paths that [Sandboxed commands under the block](/docs/en/settings-reference#sandboxed-commands-under-the-block) lists. That section also says when this part of the block doesn't apply.526* **Read block**: with [`permissions.blockReadsOutsideWorkingDirectories`](/docs/en/settings-reference#permissions-blockreadsoutsideworkingdirectories) on, sandboxed commands also lose read access to your home directory and the other directories that hold user files, apart from the paths that [Sandboxed commands under the block](/docs/en/settings-reference#sandboxed-commands-under-the-block) lists. That section also says when this part of the block doesn't apply.
498* **Blocked access**: cannot modify files outside the working directory, added directories, and the per-user temp directory without explicit permission, including shell configuration files such as `~/.bashrc` and system binaries in `/bin/`
499* **Git worktrees**: when the working directory is a [linked git worktree](/docs/en/worktrees), the sandbox also allows writes to the main repository's shared `.git` directory so commands such as `git commit` can update refs and the index. Writes to `hooks/` and `config` inside that directory remain denied.527* **Git worktrees**: when the working directory is a [linked git worktree](/docs/en/worktrees), the sandbox also allows writes to the main repository's shared `.git` directory so commands such as `git commit` can update refs and the index. Writes to `hooks/` and `config` inside that directory remain denied.
500* **Configurable**: define custom allowed and denied paths through settings
501 528
502To skip filesystem isolation entirely while keeping network isolation, set [`sandbox.filesystem.disabled`](#disable-filesystem-isolation).529To skip filesystem isolation entirely while keeping network isolation, set [`sandbox.filesystem.disabled`](#disable-filesystem-isolation).
503 530
514 541
515There is no way to exempt one of these paths: an `allowWrite` entry or an `Edit` allow rule that covers the path doesn't lift the protection. The only way to turn the protection off is [`filesystem.disabled`](#disable-filesystem-isolation), which turns off filesystem isolation for every path. To see most of these paths resolved for your machine, run `/sandbox` and open the **Config** tab, which lists them under **Denied within allowed**, mixed in with your own `denyWrite` entries.542There is no way to exempt one of these paths: an `allowWrite` entry or an `Edit` allow rule that covers the path doesn't lift the protection. The only way to turn the protection off is [`filesystem.disabled`](#disable-filesystem-isolation), which turns off filesystem isolation for every path. To see most of these paths resolved for your machine, run `/sandbox` and open the **Config** tab, which lists them under **Denied within allowed**, mixed in with your own `denyWrite` entries.
516 543
517If `git merge` or `git checkout` fails with `unable to unlink old` on one of these paths, see [Troubleshooting](#troubleshooting).544If `git merge` or `git checkout` fails with `unable to unlink old` on one of these paths, see [A git command fails with `unable to unlink old`](#a-git-command-fails-with-unable-to-unlink-old).
518 545
519### Network isolation546### Network isolation
520 547
521Network access is controlled through a proxy server running outside the sandbox:548A sandboxed command has no direct route to the network:
549
550* **Linux and WSL2**: the command runs in a separate network namespace that has no connection to your network
551* **macOS**: the Seatbelt sandbox framework by default blocks connections other than the one to the sandbox proxy
552
553Claude Code runs the sandbox proxy on your machine, outside the sandbox, and directs commands to it with `HTTP_PROXY`, `HTTPS_PROXY`, `ALL_PROXY`, and related environment variables. The proxy checks the hostname of each connection against your allowed and denied domains.
554
555What a tool can reach depends on whether it uses the proxy:
556
557* **Tools that read the proxy variables**: `curl`, `npm`, `git` over HTTPS, and similar tools connect once their host is allowed. An `allowedDomains` entry with no port allows every port on that host
558* **Tools that ignore the proxy variables**: plain `ssh`, most database drivers, and similar tools can't connect, even to an allowed host. See [A database client or other non-HTTP tool fails to reach an allowed host](#a-database-client-or-other-non-http-tool-fails-to-reach-an-allowed-host)
559* **Anything that isn't TCP**: UDP, HTTP/3 over QUIC, and ICMP tools such as `ping` can't leave the sandbox
522 560
523* **Domain restrictions**: Claude Code pre-allows no domains by default. The first time a command needs a new domain, Claude Code prompts for approval; in [auto mode](/docs/en/permission-modes#eliminate-prompts-with-auto-mode), Claude instead names the hosts a command needs on the command itself, per [Per-command allowed domains](#per-command-allowed-domains-in-auto-mode).561The following settings and behaviors control which hosts the proxy allows:
524* **Approval choices**: if you choose Yes when prompted, Claude Code allows the host for the rest of the current session and doesn't prompt again for later connections to the same host. If you choose "Yes, and don't ask again", Claude Code saves a `WebFetch(domain:...)` allow rule to your [local settings](/docs/en/permissions#permission-system), so the host stays allowed in future sessions.562
563* **Domain restrictions**: your allowed domains start empty. [Hosts outside your allowed domains](#hosts-outside-your-allowed-domains) covers what happens the first time a command needs a new domain.
564* **Approval choices**: if you choose Yes when prompted, Claude Code allows the host for the rest of the current session. If you choose "Yes, and don't ask again", Claude Code saves a `WebFetch(domain:...)` allow rule to your [local settings](/docs/en/permissions#permission-system), so the host stays allowed in future sessions. While the sandbox is [admin-required](#repository-settings-under-an-admin-required-sandbox), Claude Code saves the rule to your user settings, where it applies in every project.
525* **Pre-allowed domains**: pre-allow domains with [`allowedDomains`](/docs/en/settings-reference#sandbox-network-alloweddomains) to avoid the prompt entirely. Claude Code also pre-allows domains from `WebFetch(domain:...)` allow rules, as described in [Permission rules](#permission-rules).565* **Pre-allowed domains**: pre-allow domains with [`allowedDomains`](/docs/en/settings-reference#sandbox-network-alloweddomains) to avoid the prompt entirely. Claude Code also pre-allows domains from `WebFetch(domain:...)` allow rules, as described in [Permission rules](#permission-rules).
526* **Strict allowlist**: if you set [`strictAllowlist`](/docs/en/settings-reference#sandbox-network-strictallowlist) to `true` in user, managed, or CLI `--settings` settings, Claude Code denies sandboxed commands access to any host outside the allowlist instead of prompting. The allowlist is the same one the sandbox otherwise prompts against: `allowedDomains` plus domains from `WebFetch(domain:...)` allow rules, or only the managed settings entries when `allowManagedDomainsOnly` is set. Claude Code enforces this for sandboxed commands only; in-process tools such as `WebFetch` still follow their [permission rules](#permission-rules). Setting it in a repository's `.claude/settings.json` or `.claude/settings.local.json` has no effect. Requires Claude Code v2.1.219 or later.566* **Strict allowlist**: if you set [`strictAllowlist`](/docs/en/settings-reference#sandbox-network-strictallowlist) to `true` in user, managed, or CLI `--settings` settings, Claude Code denies sandboxed commands access to any host outside the allowlist instead of prompting. The allowlist is `allowedDomains` plus domains from `WebFetch(domain:...)` allow rules, or only the managed settings entries when `allowManagedDomainsOnly` is set. [Locks that apply without an admin-required sandbox](#locks-that-apply-without-an-admin-required-sandbox) covers a repository's entries. Claude Code enforces this for sandboxed commands only; in-process tools such as `WebFetch` still follow their [permission rules](#permission-rules). Setting it in a repository's `.claude/settings.json` or `.claude/settings.local.json` has no effect. Requires Claude Code v2.1.219 or later.
527* **Managed lockdown**: if [`allowManagedDomainsOnly`](/docs/en/settings-reference#sandbox-network-allowmanageddomainsonly) is set in managed settings, non-allowed domains are blocked automatically instead of prompting, and only `allowedDomains` and `WebFetch(domain:...)` allow rules from managed settings are honored.567* **Managed lockdown**: if [`allowManagedDomainsOnly`](/docs/en/settings-reference#sandbox-network-allowmanageddomainsonly) is set in managed settings, non-allowed domains are blocked automatically instead of prompting, and only `allowedDomains` and `WebFetch(domain:...)` allow rules from managed settings are honored.
528* **Corporate proxy**: when your network requires outbound traffic to go through a corporate proxy, set `HTTPS_PROXY`, `HTTP_PROXY`, and `NO_PROXY` as [proxy configuration](/docs/en/network-config#proxy-configuration) describes, in the `env` block of your settings so that [background agents](/docs/en/network-config#set-network-variables-in-settings-not-the-shell) get them too, or in the environment you launch Claude Code from. Claude Code enforces the domain allowlist and then tunnels allowed connections through that upstream proxy.568* **Corporate proxy**: when your network requires outbound traffic to go through a corporate proxy, set `HTTPS_PROXY`, `HTTP_PROXY`, and `NO_PROXY` as [proxy configuration](/docs/en/network-config#proxy-configuration) describes, in the `env` block of your settings so that [background agents](/docs/en/network-config#set-network-variables-in-settings-not-the-shell) get them too, or in the environment you launch Claude Code from. Claude Code enforces the domain allowlist and then tunnels allowed connections through that upstream proxy. `http://` and `https://` proxy URLs work, with basic authentication in the URL if you need it.
529* **Custom proxy support**: advanced users can implement custom rules on outgoing traffic
530* **Comprehensive coverage**: restrictions apply to all scripts, programs, and subprocesses spawned by commands
531 569
532In a `WebFetch(domain:...)` rule, the sandbox honors two wildcard forms: a leading `*.`, such as `*.example.com`, and a bare `*`. The bare `*` form requires Claude Code v2.1.186 or later. A wildcard in any other position, such as `WebFetch(domain:example.*)`, still matches fetches but has no effect on sandboxed commands.570In a `WebFetch(domain:...)` rule, the sandbox honors two wildcard forms: a leading `*.`, such as `*.example.com`, and a bare `*`. The bare `*` form requires Claude Code v2.1.186 or later. A wildcard in any other position, such as `WebFetch(domain:example.*)`, still matches fetches but has no effect on sandboxed commands.
533 571
535 The built-in proxy enforces the allowlist based on the requested hostname and, by default, does not terminate or inspect TLS traffic. The experimental [`network.tlsTerminate`](/docs/en/settings-reference#sandbox-network-tlsterminate) setting, available in Claude Code v2.1.199 and later, makes the built-in proxy terminate TLS itself, which [`mask` credential entries](#mask-credentials) require. See [Security limitations](#security-limitations) for the implications of the default, and [Custom proxy configuration](#custom-proxy-configuration) if your threat model requires TLS inspection.573 The built-in proxy enforces the allowlist based on the requested hostname and, by default, does not terminate or inspect TLS traffic. The experimental [`network.tlsTerminate`](/docs/en/settings-reference#sandbox-network-tlsterminate) setting, available in Claude Code v2.1.199 and later, makes the built-in proxy terminate TLS itself, which [`mask` credential entries](#mask-credentials) require. See [Security limitations](#security-limitations) for the implications of the default, and [Custom proxy configuration](#custom-proxy-configuration) if your threat model requires TLS inspection.
536</Note>574</Note>
537 575
576#### Hosts outside your allowed domains
577
578When a sandboxed command connects to a host that isn't in your allowed domains, the command stays in the sandbox and waits for a decision. In an interactive terminal session, the decision depends on your permission mode:
579
580| Permission mode | What happens to the connection |
581| :- | :- |
582| `bypassPermissions` mode, and plan mode with [bypass permissions available](/docs/en/permission-modes#skip-all-checks-with-bypasspermissions-mode) | Allowed without a prompt |
583| Manual mode, `acceptEdits` mode, and plan mode otherwise | You get a prompt |
584| Auto mode | Refused unless the command [listed the host](#per-command-allowed-domains-in-auto-mode) and the classifier approved the list |
585| `dontAsk` mode | Refused |
586
587With [`strictAllowlist`](/docs/en/settings-reference#sandbox-network-strictallowlist) or [`allowManagedDomainsOnly`](/docs/en/settings-reference#sandbox-network-allowmanageddomainsonly) on, the built-in sandbox proxy refuses the connection in every permission mode. In `bypassPermissions` mode, hosts outside your allowed domains are allowed unless one of them is on. [The unsandboxed retry escape hatch](#the-unsandboxed-retry-escape-hatch) covers when a command can leave the sandbox in that mode. A connection to a host in [`deniedDomains`](/docs/en/settings-reference#sandbox-network-denieddomains) is refused in every permission mode too.
588
589#### Hostnames that resolve to local addresses
590
591After a hostname passes the allowlist, the sandbox proxy resolves it and refuses the connection when the name resolves only to local addresses. Local addresses include loopback addresses such as `127.0.0.1`, link-local addresses such as the `169.254.169.254` cloud metadata endpoint, and addresses assigned to your own machine. The names `localhost` and `*.localhost` are allowed to resolve to loopback.
592
593An allowed intranet hostname that resolves to a private range such as `10.0.0.0/8` connects. To let a name resolve to a refused address, add that IP address to `allowedDomains`, such as `"127.0.0.1:8080"`.
594
595The check applies to hostnames. Your allowed domains and permission mode decide a connection to an IP address. The proxy also skips the check for connections it sends through an upstream corporate proxy, because that proxy resolves the name.
596
538#### Per-command allowed domains in auto mode597#### Per-command allowed domains in auto mode
539 598
540In [auto mode](/docs/en/permission-modes#eliminate-prompts-with-auto-mode) with sandboxing on, Claude names the hosts a command needs on the command itself instead of triggering a network approval for each connection. Each Bash, PowerShell, or [Monitor](/docs/en/tools-reference#monitor-tool) command that runs in the sandbox can carry a list of hosts beyond the sandbox's allowlist: a domain such as `registry.npmjs.org`, a wildcard such as `*.pythonhosted.org`, or an IP address, each with an optional `:port`. The classifier reviews the hosts together with the command. Requires Claude Code v2.1.271 or later.599In [auto mode](/docs/en/permission-modes#eliminate-prompts-with-auto-mode) with sandboxing on, Claude names the hosts a command needs on the command itself instead of triggering a network approval for each connection. Each Bash, PowerShell, or [Monitor](/docs/en/tools-reference#monitor-tool) command that runs in the sandbox can carry a list of hosts beyond the sandbox's allowlist: a domain such as `registry.npmjs.org`, a wildcard such as `*.pythonhosted.org`, or an IP address, each with an optional `:port`. The classifier reviews the hosts together with the command. Requires Claude Code v2.1.271 or later.
549 608
550#### IPv6 addresses in domain lists609#### IPv6 addresses in domain lists
551 610
552The sandbox's domain lists are `allowedDomains`, `deniedDomains`, and the `WebFetch(domain:...)` rules that feed them. To match an IPv6 address in any of them, write the literal in brackets: `"[::1]"` matches that address on every port, and `"[::1]:443"` matches it on port 443 only. Write the port as a number from 1 to 65535 with no leading zeros. The bracketed form requires Claude Code v2.1.229 or later. Before v2.1.229, when the text after an unbracketed entry's last colon was a port number, Claude Code read it as one, so `::1:443` named the address `::1` on port 443.611To match an IPv6 address in `allowedDomains`, `deniedDomains`, or a `WebFetch(domain:...)` rule, write the address in brackets: `"[::1]"` matches that address on every port, and `"[::1]:443"` matches it on port 443 only. The bracketed form requires Claude Code v2.1.229 or later.
553
554When you choose "Yes, and don't ask again" at the network approval prompt for an IPv6 address, Claude Code saves the `WebFetch(domain:...)` rule with the address bracketed, so the rule keeps matching the address in future sessions.
555 612
556An unbracketed entry with two or more colons is ambiguous: `::1:443` is both a complete IPv6 address and an address followed by a port. Claude Code enforces ambiguous spellings conservatively instead of guessing which reading you meant:613An unbracketed entry such as `::1:443` is ambiguous between an address and an address with a port:
557 614
558* **Deny lists**: Claude Code denies every reading the entry parses as, so whichever reading you meant is blocked. For an entry with no parseable reading, Claude Code blocks nothing.615* **Deny lists**: Claude Code denies every reading the entry parses as, so whichever reading you meant is blocked. For an entry with no parseable reading, Claude Code blocks nothing
559* **Allow lists**: Claude Code never allows more than you wrote. It rewrites an ambiguous entry to its host-and-port reading when that reading parses cleanly, and may drop the entry entirely rather than widen the allowlist.616* **Allow lists**: Claude Code never allows more than you wrote. It rewrites an ambiguous entry to its host-and-port reading when that reading parses cleanly, and may drop the entry entirely rather than widen the allowlist
560 617
561Run `claude doctor` in your terminal to find the affected entries: the `Sandbox network domain entries have unreliable spellings` warning names up to three of them and counts the rest. Rewrite each one in the bracketed form to clear the warning. The warning also names entries whose spelling is unreliable for other reasons, such as `@`, path or query characters, or wildcards inside brackets.618To find ambiguous entries, run `claude doctor` in your terminal and look for the `Sandbox network domain entries have unreliable spellings` warning. Rewrite each ambiguous entry in the bracketed form.
562 619
563### OS-level enforcement620### OS-level enforcement
564 621
568* **Linux**: uses [bubblewrap](https://github.com/containers/bubblewrap) for isolation625* **Linux**: uses [bubblewrap](https://github.com/containers/bubblewrap) for isolation
569* **WSL2**: uses bubblewrap, same as Linux626* **WSL2**: uses bubblewrap, same as Linux
570 627
571WSL1 is not supported because bubblewrap requires kernel features only available in WSL2.628You can also run the [`@anthropic-ai/sandbox-runtime`](https://github.com/anthropics/sandbox-runtime) package on its own to wrap the Claude Code process. See [Sandbox runtime](/docs/en/sandbox-environments#sandbox-runtime).
572
573These same primitives are available as the standalone [`@anthropic-ai/sandbox-runtime`](https://github.com/anthropic-experimental/sandbox-runtime) package, which the [Sandbox environments](/docs/en/sandbox-environments#sandbox-runtime) page covers as a separate approach for wrapping the entire Claude Code process.
574 629
575## How sandboxing relates to permissions and permission modes630## How sandboxing relates to permissions and permission modes
576 631
623 678
624To require the sandbox for every developer, deliver the `sandbox` keys through [managed settings](/docs/en/managed-settings#delivery-mechanisms), either as a file managed by your MDM or through [server-managed settings](/docs/en/server-managed-settings) on claude.ai.679To require the sandbox for every developer, deliver the `sandbox` keys through [managed settings](/docs/en/managed-settings#delivery-mechanisms), either as a file managed by your MDM or through [server-managed settings](/docs/en/server-managed-settings) on claude.ai.
625 680
626The following managed settings configuration enables the sandbox, refuses to start Claude Code if the sandbox cannot initialize, and prevents the model from retrying commands outside the sandbox:681The following managed settings configuration enables the sandbox, refuses to start Claude Code when the platform is unsupported or a dependency is missing, and prevents the model from retrying commands outside the sandbox:
627 682
628```json theme={null}683```json theme={null}
629{684{
637 692
638The two keys beyond `enabled` control what happens when the sandbox cannot run a command:693The two keys beyond `enabled` control what happens when the sandbox cannot run a command:
639 694
640* **`failIfUnavailable`**: a missing dependency such as bubblewrap on Linux blocks Claude Code from starting rather than showing a warning and falling back to unsandboxed execution695* **`failIfUnavailable`**: a missing dependency such as bubblewrap on Linux blocks Claude Code from starting rather than falling back to unsandboxed execution
641* **`allowUnsandboxedCommands: false`**: Claude Code ignores the `dangerouslyDisableSandbox` escape hatch, so when a command fails under the sandbox, Claude can't retry it unsandboxed696* **`allowUnsandboxedCommands: false`**: Claude Code ignores the `dangerouslyDisableSandbox` escape hatch, so when a command fails under the sandbox, Claude can't retry it unsandboxed
642 697
643Two additions are worth considering alongside them. Add `excludedCommands` for any organization-approved tools that must run without isolation. Add [`sandbox.credentials`](#protect-credentials) entries for credential directories such as `~/.aws` and `~/.ssh` and for secret environment variables, since the default read policy still allows them.698Consider these additions alongside them:
699
700* Add `excludedCommands` for any organization-approved tools that must run without isolation, because this configuration [stops a repository's settings from taking commands out of the sandbox](#repository-settings-under-an-admin-required-sandbox)
701* Add [`sandbox.credentials`](#protect-credentials) entries for credential directories such as `~/.aws` and `~/.ssh` and for secret environment variables, since the default read policy still allows them
702
703This configuration sandboxes the commands Claude runs. A developer can still type a command at the [`!` shell-mode prompt](/docs/en/interactive-mode#shell-mode-with-prefix) and run it outside the sandbox, with the same access they already have in any terminal outside Claude Code. See [strict sandbox mode](#turn-off-the-retry-with-strict-sandbox-mode) for the sessions where typed commands run sandboxed.
644 704
645This configuration sandboxes the commands Claude runs. A developer can still type a command at the [`!` shell-mode prompt](/docs/en/interactive-mode#shell-mode-with-prefix) and run it outside the sandbox, with the same access they already have in any terminal outside Claude Code. See [The unsandboxed retry escape hatch](#the-unsandboxed-retry-escape-hatch) for the sessions where typed commands run sandboxed.705The sandbox doesn't run on native Windows, so with `failIfUnavailable` set, Claude Code exits at startup on those machines. If your fleet includes Windows hosts, you can:
646 706
647The sandbox does not run on native Windows, so if your fleet includes Windows hosts, scope this configuration to macOS and Linux or have those users run Claude Code inside WSL2 or a container.707* **Deliver the configuration by operating system**: deploy it through your MDM or as a [managed settings file](/docs/en/managed-settings#delivery-mechanisms) on macOS and Linux machines only. [Server-managed settings](/docs/en/server-managed-settings#current-limitations) apply to all users in the organization
708* **Move Windows users into a supported environment**: have them run Claude Code inside WSL2 or a container
648 709
649### Keep developers from widening the policy710### Keep developers from widening the policy
650 711
651For boolean keys such as `enabled` and `failIfUnavailable`, Claude Code uses the managed value and ignores anything a developer sets locally. For array keys such as `excludedCommands` and `allowRead`, Claude Code merges entries from every scope the session loads, so a developer can append entries that widen the policy.712When managed settings set a Boolean key such as `enabled` or `failIfUnavailable`, Claude Code uses the managed value and ignores anything a developer sets locally. For array keys such as `allowRead`, Claude Code merges entries from the scopes the session loads, so a developer can append entries that widen the policy unless a lock covers that key.
713
714Unless managed settings set them, a developer's user settings or `--settings` can turn on the following keys. A repository's `.claude/settings.json` can too, unless the sandbox is [admin-required](#repository-settings-under-an-admin-required-sandbox). Each one weakens the sandbox, so set it to `false` in managed settings if you don't want it used:
715
716* [`enableWeakerNestedSandbox`](/docs/en/settings-reference#sandbox-enableweakernestedsandbox)
717* [`enableWeakerNetworkIsolation`](/docs/en/settings-reference#sandbox-enableweakernetworkisolation)
718* [`network.allowAllUnixSockets`](/docs/en/settings-reference#sandbox-network-allowallunixsockets)
719* [`network.allowLocalBinding`](/docs/en/settings-reference#sandbox-network-allowlocalbinding)
720* [`allowAppleEvents`](/docs/en/settings-reference#sandbox-allowappleevents), which a repository can't turn on
721
722Set `allowManagedReadPathsOnly` to `true` in managed settings so that only `allowRead` entries from managed settings are honored. This prevents developers from widening read access beyond the organization-approved paths.
723
724To lock network domains to the managed values the same way, set [`allowManagedDomainsOnly`](/docs/en/settings-reference#sandbox-network-allowmanageddomainsonly). With the lock on, only managed settings can set a [proxy port](#custom-proxy-configuration).
725
726When managed settings configure `sandbox.filesystem` or list any `sandbox.credentials.files` entry with `"mode": "deny"`, only managed settings can set [`filesystem.disabled`](#disable-filesystem-isolation), so developers can't switch off administrator-deployed filesystem restrictions. A [valid](/docs/en/settings-reference#invalid-credential-entries-in-managed-settings) `mask` entry doesn't lock the key. See [Which settings can disable it](#which-settings-can-disable-it).
727
728#### Repository settings under an admin-required sandbox
652 729
653Set `allowManagedReadPathsOnly` to `true` in managed settings so that only `allowRead` entries from managed settings are honored. This prevents developers from widening read access beyond the organization-approved paths. To lock network domains to the managed values the same way, set [`allowManagedDomainsOnly`](/docs/en/settings-reference#sandbox-network-allowmanageddomainsonly).730The sandbox is admin-required while one of these settings is in effect:
654 731
655When managed settings configure `sandbox.filesystem` or list any `sandbox.credentials.files` entry with `"mode": "deny"`, only managed settings can set [`filesystem.disabled`](#disable-filesystem-isolation), so developers can't switch off administrator-deployed filesystem restrictions. Whether a `mask` entry pins the key depends on how it resolves; the table under [Which settings can disable it](#which-settings-can-disable-it) covers the four cases.732* [`allowUnsandboxedCommands`](/docs/en/settings-reference#sandbox-allowunsandboxedcommands) set to `false` in managed settings, or with the `--settings` flag unless managed settings set it to `true`
733* [`allowManagedDomainsOnly`](/docs/en/settings-reference#sandbox-network-allowmanageddomainsonly) set to `true` in managed settings
656 734
657`excludedCommands` has no equivalent managed-only lockdown, so a developer can always append entries that run additional commands outside the sandbox. Keep the managed list narrow.735These settings don't turn the sandbox on, so set `enabled` as well.
736
737While the sandbox is admin-required, Claude Code takes the settings that loosen it only from managed settings, the `--settings` flag, and each developer's `~/.claude/settings.json`. It ignores these settings in a repository's `.claude/settings.json` and `.claude/settings.local.json`:
738
739| Repository setting | What Claude Code ignores |
740| :- | :- |
741| `excludedCommands`, `ignoreViolations`, `network.allowedDomains`, `network.allowUnixSockets`, `network.allowMachLookup`, `network.httpProxyPort`, `network.socksProxyPort` | Every entry |
742| `filesystem.allowWrite`, `Edit(...)` allow rules, `permissions.additionalDirectories` | The write access each entry gives sandboxed commands. Claude's file tools still follow the `Edit(...)` rules and additional directories |
743| `WebFetch(domain:...)` allow rules | The host each rule adds to the sandbox allowlist. The WebFetch tool still follows the rule |
744| `enableWeakerNestedSandbox`, `enableWeakerNetworkIsolation`, `network.allowAllUnixSockets`, `network.allowLocalBinding` | `true`. A `false` still applies |
745| `enabled`, `failIfUnavailable` | `false`, when the developer's `~/.claude/settings.json` sets `true` |
746| `filesystem.allowRead` | An entry at or under a path that managed settings, `--settings`, or user settings deny reading, or a glob that could match one |
747
748These settings still apply while the sandbox is admin-required:
749
750* **In a repository's files**: deny entries and the `autoAllowBashIfSandboxed` value. Set the key in managed settings to keep a repository from changing it
751* **In a developer's own settings**: the settings in the table still apply from `~/.claude/settings.json` or `--settings`, unless a managed-only lock such as `allowManagedDomainsOnly` covers them. Most of them, such as `excludedCommands` and `filesystem.allowWrite`, have no managed-only lock
752
753The configuration under [Enforce sandboxing with managed settings](#enforce-sandboxing-with-managed-settings) makes the sandbox admin-required. Add the `excludedCommands`, `allowWrite`, and socket entries that your approved tools need to managed settings, because a repository can't supply them.
754
755Requires Claude Code v2.1.285 or later. From v2.1.282 to v2.1.284, the same settings made Claude Code ignore a repository's `excludedCommands` entries.
756
757#### Locks that apply without an admin-required sandbox
758
759Some settings make Claude Code ignore the repository keys that directly override one restriction, even when the sandbox isn't admin-required. Each one has this effect only when you set it in a file its row names, and the repository's other sandbox settings still apply. Requires Claude Code v2.1.285 or later.
760
761| Setting | Where you set it | What Claude Code ignores in a repository's settings |
762| :- | :- | :- |
763| `network.deniedDomains` or a `WebFetch(domain:...)` deny rule | Managed settings, `--settings` | `httpProxyPort` and `socksProxyPort` |
764| `network.strictAllowlist` | Managed settings, `--settings`, user settings | The proxy ports, `allowedDomains`, and `WebFetch(domain:...)` allow rules |
765| `filesystem.denyRead`, a `Read(...)` deny rule, or a `credentials.files` entry | Managed settings, `--settings` | An `allowRead`, `allowWrite`, `Edit(...)` allow, or `additionalDirectories` entry at or under a path that managed settings, `--settings`, or user settings deny reading, or a glob that could match one |
766
767These locks change what sandboxed commands can reach. The WebFetch tool and Claude's file tools still follow a repository's rules and additional directories.
658 768
659### Custom proxy configuration769### Custom proxy configuration
660 770
661For organizations requiring advanced network security, you can implement a custom proxy to:771To inspect, filter, or log sandbox traffic with your own tooling, replace the built-in sandbox proxy with a proxy you run on the same machine.
662 772
663* Decrypt and inspect HTTPS traffic773To route sandbox traffic through a corporate proxy elsewhere on your network, set `HTTPS_PROXY` instead, as the **Corporate proxy** entry under [Network isolation](#network-isolation) describes. That way, Claude Code's allowlist still applies.
664* Apply custom filtering rules
665* Log all network requests
666* Integrate with existing security infrastructure
667 774
668To point Claude Code at your proxy, set the proxy ports in [sandbox settings](/docs/en/settings-reference#sandbox-settings):775To direct sandboxed commands to your proxy, set the localhost ports it listens on in [sandbox settings](/docs/en/settings-reference#sandbox-settings):
669 776
670```json theme={null}777```json theme={null}
671{778{
678}785}
679```786```
680 787
788If you set a port and also set `HTTPS_PROXY` or `HTTP_PROXY`, Claude Code doesn't forward what sandboxed commands send to your proxy on to the proxy those variables name. To reach a corporate proxy, configure your own proxy to forward to it.
789
790Which files can set a port depends on your other sandbox settings. The first case that matches applies:
791
792* **`allowManagedDomainsOnly` is on**: managed settings only
793* **The sandbox is [admin-required](#repository-settings-under-an-admin-required-sandbox), or a [narrower network lock](#locks-that-apply-without-an-admin-required-sandbox) applies**: managed settings, `--settings`, and user settings
794* **Otherwise**: any settings file
795
796Claude Code ignores a port set anywhere else. Before v2.1.285, any settings file could set a port.
797
798<Warning>
799 Once either port applies, your proxy is responsible for filtering everything sent to it. Claude Code's own network controls, such as `allowedDomains`, `deniedDomains`, `strictAllowlist`, approval prompts, and the [local-address check](#hostnames-that-resolve-to-local-addresses), stop applying to that traffic. A sandboxed command can connect to either proxy, so if you set only one port, Claude Code's domain lists on the other proxy don't limit what the command reaches through yours.
800</Warning>
801
681## Troubleshooting802## Troubleshooting
682 803
683Some commands fail inside the sandbox even though they work outside it. The fixes below cover the most common cases.804Some commands fail inside the sandbox even though they work outside it. Find the heading that matches your symptom or error message.
805
806If your organization's sandbox is [admin-required](#repository-settings-under-an-admin-required-sandbox), Claude Code ignores the settings these fixes name in a project's settings files, so save them in `~/.claude/settings.json`, where they apply in every project. If a fix still has no effect, your organization's managed settings may set that key.
807
808A fix that adds an `excludedCommands` pattern removes the sandbox from the commands the pattern matches. See [what an excluded command can do](#run-commands-outside-the-sandbox-with-excludedcommands).
809
810### Commands fail with a host-not-allowed error
811
812Many CLI tools need to reach specific hosts. Approve the host when prompted, or add it to [`allowedDomains`](/docs/en/settings-reference#sandbox-network-alloweddomains). If your organization locks the allowlist with `allowManagedDomainsOnly`, there's no prompt, so ask your administrator to add the host.
813
814### `jest` hangs or fails
815
816`watchman` is incompatible with the sandbox. Run `jest --no-watchman` instead.
817
818### Go-based CLIs fail TLS verification on macOS
819
820Tools such as `gh`, `gcloud`, and `terraform` may fail TLS verification under [Seatbelt](#os-level-enforcement). To run these tools outside the sandbox, add a pattern for each tool, such as `gh *`, to [`excludedCommands`](#run-commands-outside-the-sandbox-with-excludedcommands). The tool then runs with your full access and its stored credentials. If you are using `httpProxyPort` with a MITM proxy and custom CA, set [`enableWeakerNetworkIsolation`](/docs/en/settings-reference#sandbox-enableweakernetworkisolation) to `true` instead.
684 821
685* **Commands fail with a host-not-allowed error**: many CLI tools need to reach specific hosts. Granting permission when prompted adds the host to your allowed list so the tool runs inside the sandbox in future.822### `open`, `osascript`, or browser-based auth flows fail with error `-600` on macOS
686* **`jest` hangs or fails**: `watchman` is incompatible with the sandbox. Run `jest --no-watchman` instead.
687* **Go-based CLIs fail TLS verification on macOS**: tools such as `gh`, `gcloud`, and `terraform` may fail TLS verification under Seatbelt. List these tools in [`excludedCommands`](/docs/en/settings-reference#sandbox-excludedcommands). If you are using `httpProxyPort` with a MITM proxy and custom CA, set [`enableWeakerNetworkIsolation`](/docs/en/settings-reference#sandbox-enableweakernetworkisolation) to `true` instead.
688* **`open`, `osascript`, or browser-based auth flows fail with error `-600` on macOS**: the sandbox blocks Apple Events by default. Set [`allowAppleEvents`](/docs/en/settings-reference#sandbox-allowappleevents) to `true` in your user, managed, or CLI settings to allow them. Project settings are ignored for this key. Enabling it removes code-execution isolation, since sandboxed commands can then launch other applications unsandboxed with no user prompt and send AppleScript commands to running applications, subject to the macOS automation-consent prompt (TCC). Alternatively, add the command to [`excludedCommands`](/docs/en/settings-reference#sandbox-excludedcommands).
689* **`docker` commands fail**: `docker` is incompatible with the sandbox. Add `docker *` to [`excludedCommands`](/docs/en/settings-reference#sandbox-excludedcommands).
690* **`pbcopy`, `xclip`, or `wl-copy` doesn't update the clipboard**: these clipboard utilities can fail to reach the system clipboard from inside the sandbox, in which case the text piped to them doesn't arrive.
691 823
692 To put Claude's output on your clipboard, ask Claude to print it in its response, then run [`/copy`](/docs/en/commands). `/copy` writes to the clipboard from the Claude Code process rather than from a sandboxed command.824The sandbox blocks Apple Events by default. Set [`allowAppleEvents`](/docs/en/settings-reference#sandbox-allowappleevents) to `true` in your user, managed, or CLI settings to allow them. Claude Code ignores this key in project settings.
693 825
694 When Claude pipes text to one of these tools, adding the tool to [`excludedCommands`](/docs/en/settings-reference#sandbox-excludedcommands) doesn't take that call out of the sandbox on its own.826Enabling `allowAppleEvents` removes code-execution isolation, since sandboxed commands can then launch other applications unsandboxed with no user prompt, and can send AppleScript commands to running applications, subject to the macOS automation-consent prompt (TCC). Alternatively, add a pattern such as `open *` to [`excludedCommands`](#run-commands-outside-the-sandbox-with-excludedcommands). Each `open` call then goes through the permission flow, and `open` can launch any file or app, including one Claude wrote.
695* **A git command fails with `unable to unlink old`**: `git merge`, `git checkout`, and similar commands fail this way when they need to replace a file the sandbox denies writes to, whether that file is under a [protected path](#protected-paths) such as `.claude/skills`, under one of your `denyWrite` entries, or outside the directories the sandbox lets commands write to at all. On Linux and WSL2 the error ends with `Read-only file system`.
696 827
697 After the failure, Claude may [offer to rerun the command outside the sandbox](#the-unsandboxed-retry-escape-hatch); approve that retry, or run the git command yourself in another terminal. If you've set `allowUnsandboxedCommands` to `false`, Claude can't offer the retry, so run the command yourself. If the same git command fails often, add it to [`excludedCommands`](/docs/en/settings-reference#sandbox-excludedcommands).828### `docker` commands fail
698* **Bubblewrap fails to start inside a container**: in an unprivileged container, bubblewrap can't mount a fresh `/proc` filesystem, so sandboxed commands fail with a `bwrap` error such as `Can't mount proc on /newroot/proc: Operation not permitted`. Set [`enableWeakerNestedSandbox`](/docs/en/settings-reference#sandbox-enableweakernestedsandbox) to `true` so the inner sandbox bind-mounts the container's existing `/proc` instead. Only use this setting when the outer container already provides the isolation boundary you need, since it exposes process information to sandboxed commands that a fresh `/proc` mount would hide.
699* **0-byte read-only files appear at `.claude` settings paths, and "Yes, and don't ask again" doesn't save**: on Linux and WSL2, the sandbox holds a write denial on a file that doesn't exist yet by creating a 0-byte read-only placeholder there while a sandboxed command runs. The sandbox removes the placeholder afterward. If a session is killed before that cleanup runs, for example by SIGKILL, the placeholders stay behind. Later sessions bind them read-only again on every start, so a settings write such as saving a permission choice fails where one sits.
700 829
701 Run `claude doctor` to list the leftover placeholder files. The [`Stale sandbox mask files left by a killed session`](/docs/en/errors#stale-sandbox-mask-files-left-by-a-killed-session) warning names up to three of them and counts the rest. Delete each file with `rm` while no other Claude Code session is running in that project. Before v2.1.257, Claude Code left the same placeholders behind without flagging them.830`docker` is incompatible with the sandbox. Take the `docker` commands you need out of the sandbox with an `excludedCommands` pattern such as `docker compose *`. [Run commands outside the sandbox with `excludedCommands`](#run-commands-outside-the-sandbox-with-excludedcommands) explains what an excluded `docker` command can reach. A narrower pattern takes fewer commands out of the sandbox.
702* **`--dangerously-skip-permissions` fails as root**: this flag is blocked when running as root or via sudo on Linux and macOS, because root access combined with no permission prompts can modify any file or service on the system. The check is skipped automatically inside a recognized sandbox. To run autonomously in a container, use the [dev container](/docs/en/devcontainer) configuration, which runs Claude Code as a non-root user.831
832### `pbcopy`, `xclip`, or `wl-copy` doesn't update the clipboard
833
834The `pbcopy`, `xclip`, and `wl-copy` clipboard utilities can fail to reach the system clipboard from inside the sandbox, in which case the text piped to them doesn't arrive.
835
836To put Claude's output on your clipboard, ask Claude to print it in its response, then run [`/copy`](/docs/en/commands). `/copy` writes to the clipboard from the Claude Code process rather than from a sandboxed command.
837
838When Claude pipes text to one of these tools, adding the tool to [`excludedCommands`](/docs/en/settings-reference#sandbox-excludedcommands) doesn't take that call out of the sandbox on its own.
839
840### A git command fails with `unable to unlink old`
841
842`git merge`, `git checkout`, and similar commands fail with `unable to unlink old` when they need to replace a file the sandbox denies writes to. On Linux and WSL2 the error ends with `Read-only file system`. The file can be in one of these places:
843
844* Under a [protected path](#protected-paths) such as `.claude/skills`
845* Under one of your `denyWrite` entries
846* Outside the directories the sandbox lets commands write to at all
847
848After the failure, Claude may [offer to rerun the command outside the sandbox](#the-unsandboxed-retry-escape-hatch). Approve that retry, or run the git command yourself in another terminal. If you've set `allowUnsandboxedCommands` to `false`, Claude can't offer the retry, so run the command yourself.
849
850### Bubblewrap fails to start inside a container
851
852In an unprivileged container, [bubblewrap](#os-level-enforcement) can't mount a fresh `/proc` filesystem, so sandboxed commands fail with a `bwrap` error such as `Can't mount proc on /newroot/proc: Operation not permitted`. Set [`enableWeakerNestedSandbox`](/docs/en/settings-reference#sandbox-enableweakernestedsandbox) to `true` so the sandbox bind-mounts the container's existing `/proc` instead. Only use this setting when the outer container already provides the isolation boundary you need, since the setting exposes process information to sandboxed commands that a fresh `/proc` mount would hide.
853
854### 0-byte read-only files appear at `.claude` settings paths, and "Yes, and don't ask again" doesn't save
855
856On Linux and WSL2, the sandbox holds a write denial on a file that doesn't exist yet by creating a 0-byte read-only placeholder there while a sandboxed command runs. The sandbox removes the placeholder afterward. If a session is killed before that cleanup runs, for example by SIGKILL, the placeholders stay behind. Later sessions bind the placeholders read-only again on every start, so a settings write such as saving a permission choice fails at a path where a placeholder remains.
857
858Run `claude doctor` in your terminal to list the leftover placeholder files. The [`Stale sandbox mask files left by a killed session`](/docs/en/errors#stale-sandbox-mask-files-left-by-a-killed-session) warning names some of them and counts the rest. Delete each file with `rm` while no other Claude Code session is running in that project. Before v2.1.257, Claude Code left the same placeholders behind without flagging them.
859
860### `git` over SSH fails with the sandbox on
861
862On macOS, `git fetch`, `git pull`, and `git push` against an SSH remote fail inside the sandbox even when the host is allowed. On Linux and WSL2, they work once the host is allowed. Claude Code tunnels git's SSH connection through the [sandbox proxy](#network-isolation), and the macOS tunnel can't authenticate to that proxy.
863
864On Linux and WSL2, check these if the connection still fails:
865
866* **The host is allowed on port 22**: an `allowedDomains` entry with no port, such as `"git.example.com"`, covers it
867* **Your corporate proxy permits port 22**: if your network requires an upstream proxy, the tunnel goes through it too
868* **The key is readable as a file**: the sandbox can block the `ssh-agent` socket, and a `denyRead` or `credentials` entry for `~/.ssh` hides your key files
869
870On macOS, switch the remote to HTTPS, which needs HTTPS credentials such as a personal access token:
871
872```bash theme={null}
873git remote set-url origin https://git.example.com/example-org/example-repo.git
874```
875
876If you have to keep the SSH remote, take git's network commands out of the sandbox with [`excludedCommands`](#run-commands-outside-the-sandbox-with-excludedcommands):
877
878```json theme={null}
879{
880 "sandbox": {
881 "excludedCommands": ["git fetch *", "git pull *", "git push *"]
882 }
883}
884```
885
886These entries match `git push origin main`. A call that adds a `cd`, uses `git -C`, or contains a command substitution stays sandboxed. The excluded git commands can reach any host, not only the ones in `allowedDomains`.
887
888Plain `ssh`, `scp`, and `rsync` over SSH fail for the reason [the database client entry](#a-database-client-or-other-non-http-tool-fails-to-reach-an-allowed-host) gives.
889
890### A database client or other non-HTTP tool fails to reach an allowed host
891
892A tool that ignores the proxy environment variables can't connect from inside the sandbox, even to a host in `allowedDomains`. A sandboxed command has [no direct route to the network](#network-isolation), so a tool that opens its own connection fails. Most database drivers, plain `ssh`, and tools that use UDP behave this way.
893
894The failure looks like a network or name resolution error:
895
896* **macOS**: `Operation not permitted`, or a name resolution error such as `Could not resolve host`
897* **Linux and WSL2**: `Network is unreachable`, or a name resolution error such as `Temporary failure in name resolution`
898
899A tool that uses the proxy fails differently when its host isn't allowed. You get a network prompt, or the tool receives a `403` response from the proxy.
900
901To let the tool connect, run the command that needs it outside the sandbox with [`excludedCommands`](#run-commands-outside-the-sandbox-with-excludedcommands). This example excludes one script and adds an [ask rule](/docs/en/permissions) so that you approve each run:
902
903```json theme={null}
904{
905 "sandbox": {
906 "excludedCommands": ["python scripts/load_orders.py *"]
907 },
908 "permissions": {
909 "ask": ["Bash(python scripts/load_orders.py *)"]
910 }
911}
912```
913
914The script runs with your full access, and Claude can edit a script that's inside your working directory, so review it when the prompt appears.
915
916### A command fails to reach a server on localhost
917
918By default, a sandboxed command can't connect directly to a server that's running on your machine outside the sandbox, such as a dev server or a database in a container. What you can change depends on your platform:
919
920* **macOS**: set [`network.allowLocalBinding`](/docs/en/settings-reference#sandbox-network-allowlocalbinding) to `true`. Sandboxed commands can then listen on network ports and connect to any port on localhost, which includes every other service listening there. A localhost service that doesn't require authentication, such as a debugger, can then act for the command outside the sandbox, and a command that listens on a non-loopback address accepts connections from other machines
921* **Linux and WSL2**: a sandboxed command's `localhost` is private to that command. The command can listen on a port and reach servers it started itself. A direct connection to `localhost` or `127.0.0.1` doesn't reach servers on the host, and `allowLocalBinding` has no effect. Run the command that needs the host's server outside the sandbox with [`excludedCommands`](#run-commands-outside-the-sandbox-with-excludedcommands), where it has no filesystem or network limits. For connections that go through the sandbox proxy, see [Hostnames that resolve to local addresses](#hostnames-that-resolve-to-local-addresses)
922
923This example turns the setting on for macOS:
924
925```json theme={null}
926{
927 "sandbox": {
928 "network": {
929 "allowLocalBinding": true
930 }
931 }
932}
933```
934
935An `allowedDomains` entry for `localhost` applies to connections that go through the proxy, so it doesn't change a direct connection. Claude Code sets `NO_PROXY` for sandboxed commands so that they connect to `localhost` directly instead of through the proxy. The entry also exposes every port on your machine's localhost to a command that does use the proxy. For a development hostname that points at `127.0.0.1`, see [An allowed hostname is refused with `resolved to a loopback address`](#an-allowed-hostname-is-refused-with-resolved-to-a-loopback-address).
936
937### An allowed hostname is refused with `resolved to a loopback address`
938
939The sandbox proxy refuses an allowed hostname that [resolves to a local address](#hostnames-that-resolve-to-local-addresses), which affects development names such as `myapp.test` that point at `127.0.0.1`. The command sees a `403` response whose body names the kind of address, such as `Connection to myapp.test blocked: resolved to a loopback address`.
940
941Add the IP address the name resolves to alongside the hostname in `allowedDomains`, each with the port your server listens on:
942
943```json theme={null}
944{
945 "sandbox": {
946 "network": {
947 "allowedDomains": ["myapp.test:3000", "127.0.0.1:3000"]
948 }
949 }
950}
951```
952
953An IP address entry with no port lets sandboxed commands reach every service listening on that address.
954
955Before v2.1.284, the proxy connected to whatever address an allowed hostname resolved to.
956
957### `/sandbox` fails with `Sandbox settings are overridden by a higher-priority configuration`
958
959`/sandbox` prints `Error: Sandbox settings are overridden by a higher-priority configuration and cannot be changed locally.` instead of opening its panel when a higher [settings level](/docs/en/settings#settings-precedence) sets `sandbox.enabled`, `sandbox.autoAllowBashIfSandboxed`, or `sandbox.allowUnsandboxedCommands`. The panel saves your choices to `.claude/settings.local.json`, and a value saved there can't override those levels.
960
961Managed settings and `--settings` rank above local settings. To see which of them this session loaded, run `/status` and read the `Setting sources` line:
962
963* **`Command line arguments`**: if you started Claude Code with [`--settings`](/docs/en/settings#change-a-setting-for-one-session), check whether the file or JSON you passed sets one of those keys. If it does, change the value there, or start Claude Code again without those keys.
964* **`Enterprise managed settings`**: your organization's managed settings are loaded. If they set one of those keys, you can't change that key from `/sandbox` or from any settings file you control, so ask your administrator.
703 965
704## Limitations966## Limitations
705 967
715 977
716* **Privilege escalation via Unix sockets**: the `allowUnixSockets` configuration can inadvertently grant access to system services that could lead to sandbox bypasses. For example, allowing access to `/var/run/docker.sock` effectively grants access to the host system through the Docker socket. Consider carefully any Unix sockets that you allow through the sandbox.978* **Privilege escalation via Unix sockets**: the `allowUnixSockets` configuration can inadvertently grant access to system services that could lead to sandbox bypasses. For example, allowing access to `/var/run/docker.sock` effectively grants access to the host system through the Docker socket. Consider carefully any Unix sockets that you allow through the sandbox.
717* **Filesystem permission escalation**: overly broad filesystem write permissions can enable privilege escalation attacks. Allowing writes to directories containing executables in `$PATH`, system configuration directories, or user shell configuration files such as `.bashrc` or `.zshrc` can lead to code execution in different security contexts when other users or system processes access these files.979* **Filesystem permission escalation**: overly broad filesystem write permissions can enable privilege escalation attacks. Allowing writes to directories containing executables in `$PATH`, system configuration directories, or user shell configuration files such as `.bashrc` or `.zshrc` can lead to code execution in different security contexts when other users or system processes access these files.
718* **Linux sandbox strength**: the Linux implementation provides strong filesystem and network isolation but includes an `enableWeakerNestedSandbox` mode that enables it to work inside Docker environments without privileged namespaces, or on Linux hosts where unprivileged user namespaces are disabled by sysctl. This option considerably weakens security and should only be used when additional isolation is otherwise enforced.980* **Linux sandbox strength**: the Linux implementation provides strong filesystem and network isolation but includes an `enableWeakerNestedSandbox` mode that enables it to work inside Docker environments without privileged namespaces. This option considerably weakens security and should only be used when additional isolation is otherwise enforced.
719* **Apple Events on macOS**: the macOS sandbox blocks Apple Events by default. The `allowAppleEvents` setting lifts this restriction so tools such as `open` and `osascript` work, but it removes code-execution isolation: sandboxed commands can launch other applications unsandboxed with no user prompt, and can send AppleScript commands to running applications, subject to the per-app macOS automation-consent prompt (TCC). It is only honored from user, managed, or CLI settings. Project settings cannot enable it.981* **Apple Events on macOS**: the macOS sandbox blocks Apple Events by default. The `allowAppleEvents` setting lifts this restriction so tools such as `open` and `osascript` work, but it removes code-execution isolation: sandboxed commands can launch other applications unsandboxed with no user prompt, and can send AppleScript commands to running applications, subject to the per-app macOS automation-consent prompt (TCC). It is only honored from user, managed, or CLI settings. Project settings cannot enable it.
720 982
721### Platform and tool compatibility
722
723* **Platform support**: supports macOS, Linux, and WSL2. WSL1 and native Windows are not supported.
724* **Performance overhead**: minimal, but some filesystem operations may be slightly slower.
725* **Tool compatibility**: some tools that require specific system access patterns may need configuration adjustments, or may need to be run outside the sandbox.
726
727### Scope983### Scope
728 984
729The sandbox isolates Bash subprocesses. Other tools operate under different boundaries:985The sandbox isolates shell commands and their child processes. [What runs outside the sandbox](#what-runs-outside-the-sandbox) lists the tools and helper processes it doesn't cover. Computer use and subagents relate to the sandbox as follows:
730 986
731* **Built-in file tools**: Read, Edit, and Write use the permission system directly rather than running through the sandbox. See [permissions](/docs/en/permissions).
732* **Computer use**: when Claude opens apps and controls your screen, it runs on your actual desktop rather than in an isolated environment. Per-app permission prompts gate each application. See [computer use in the CLI](/docs/en/computer-use) or [computer use in Desktop](/docs/en/desktop#let-claude-use-your-computer).987* **Computer use**: when Claude opens apps and controls your screen, it runs on your actual desktop rather than in an isolated environment. Per-app permission prompts gate each application. See [computer use in the CLI](/docs/en/computer-use) or [computer use in Desktop](/docs/en/desktop#let-claude-use-your-computer).
733* **Environment variables**: sandboxed Bash commands inherit the parent process environment by default, including any credentials set there. Use [`sandbox.credentials`](#protect-credentials) to unset or mask specific variables for sandboxed commands, or set [`CLAUDE_CODE_SUBPROCESS_ENV_SCRUB`](/docs/en/env-vars) to strip credentials from all subprocesses.
734* **Subagents**: [subagents](/docs/en/sub-agents) run in the same process as the parent session and use the same sandbox configuration. Bash commands inside a subagent are sandboxed when sandboxing is enabled in the parent session.988* **Subagents**: [subagents](/docs/en/sub-agents) run in the same process as the parent session and use the same sandbox configuration. Bash commands inside a subagent are sandboxed when sandboxing is enabled in the parent session.
735* **Mods**: a [mod](/docs/en/plugins/mods/overview) is a plugin that runs its own code inside Claude Code, and a process that a mod starts runs outside the sandbox. See [What a mod can reach](/docs/en/plugins/mods/overview#what-a-mod-can-reach).989* **Mods**: a [mod](/docs/en/plugins/mods/overview) is a plugin that runs its own code inside Claude Code, and a process that a mod starts runs outside the sandbox. See [What a mod can reach](/docs/en/plugins/mods/overview#what-a-mod-can-reach).
736 990
