Manage credentials

Most agents need an API key for their model provider. An HTTP/HTTPS proxy on your host intercepts outbound requests from the sandbox, looks up the matching credential on the host, and overwrites the auth header before forwarding. The real credential stays on the host when proxy management is active; the sandbox sees only a sentinel value. See Trust boundaries for how credential isolation fits into the broader sandbox security model.

How credential injection works

When a sandbox makes an outbound request, the host-side proxy decides three things: whether the request matches a service the kit (or built-in agent) declares, what header to write, and what value to inject. The kit declares the match and the header; you provide the value on the host. For proxy-managed credentials, the real value never enters the sandbox — the agent sees only a sentinel like proxy-managed.

A kit can set OAuth passthrough: true to opt out of sentinel masking. This sends the real token response into the sandbox and reduces credential isolation. See the oauth kit fields.

There are several ways to provide that value. When more than one source has a value for the same service, the stored secret takes precedence.

For multi-provider agents (OpenCode, Docker Agent), the proxy selects credentials based on the API endpoint being called. See individual agent pages for provider-specific details.

Stored secrets

sbx secret set stores credential values or dynamic secret sources in your OS keychain, keyed on a service identifier. Built-in agents declare a fixed set of services. Custom kits can declare their own. The same sbx secret set flow works for both.

Where secrets are stored

The store backing sbx secret set depends on your operating system:

• macOS: the system Keychain.
• Windows: the Windows Credential Manager.
• Linux: the Secret Service exposed by your desktop keyring, such as GNOME Keyring or KDE Wallet.

The Ubuntu package depends on GNOME Keyring, so a standard desktop install needs no extra setup.

On Linux hosts without a running Secret Service — headless servers and some WSL setups — sbx falls back to a file under your user config directory $XDG_CONFIG_HOME/com.docker.sandboxes, which defaults to ~/.config/com.docker.sandboxes when $XDG_CONFIG_HOME is unset. The fallback is automatic and needs no configuration. When you store a secret this way, sbx prints a notice:

No keychain detected - this secret will be stored on disk, protected by file permissions rather than a password

sbx stores the file in a directory with 0700 permissions, the same file-permission model used for ~/.docker/config.json. Any user or process that can read the file can retrieve the stored credentials, so treat the directory as sensitive. Where available, prefer a keychain, which mediates access per application.

If you start a Secret Service on the host later, sbx stores new secrets in the keychain again. For more on running sandboxes without a desktop keyring, see Can I use Docker Sandboxes on headless Linux?

Store a secret

$ sbx secret set anthropic

This prompts you for the secret value interactively. Service secrets are global by default, so the secret is available to all sandboxes. To scope a secret to a specific sandbox instead:

$ sbx secret set openai --sandbox my-sandbox

Note

A sandbox-scoped secret takes effect immediately, even if the sandbox is running. A global secret only applies when a sandbox is created. If you set or change a global secret while a sandbox is running, recreate the sandbox for the new value to take effect.

Use a dynamic secret source

Dynamic secret sources let sbx retrieve a credential from an authenticated host tool when the proxy needs it. The secret store contains the reference or command instead of the credential value. Resolution and caching happen on the host, and the sandbox still receives only the proxy-managed placeholder.

Use --ref with a 1Password secret reference or an AWS Secrets Manager ARN:

$ sbx secret set anthropic --ref 'op://Work/Anthropic/credential'
$ sbx secret set openai \
    --ref 'arn:aws:secretsmanager:us-west-2:123456789012:secret:openai-api-key'

The corresponding op or aws CLI must be installed and authenticated on the host.

Note

To resolve a reference with a specific 1Password account or AWS profile, set OP_ACCOUNT or AWS_PROFILE when you run sbx secret set. sbx uses that account or profile whenever it resolves the secret. If neither variable is set, the provider CLI uses its default.

Use --command for another host tool that prints a secret to standard output:

$ sbx secret set github --command 'gh auth token'

sbx runs the command through the host shell and trims its output. The command text is stored and replayed by the daemon. Don't embed a secret directly in the command because the text can appear in shell history and process listings.

By default, sbx verifies the source when you register it and reports an error without exposing the resolver's standard error. Use --no-verify to store a source that can't be resolved during registration. To troubleshoot an initial verification failure, use --show-error. Provider error output can contain sensitive information. You can't combine --show-error with --no-verify.

Resolved service secrets are cached for 55 minutes by default. To change the cache duration, use --refresh <duration>. To resolve the source for every credential use instead of caching it, pass --refresh on-demand:

$ sbx secret set anthropic \
    --ref 'op://Work/Anthropic/credential' \
    --refresh 30m
$ sbx secret set github --command 'gh auth token' --refresh on-demand

--ref and --command are mutually exclusive. They can't be combined with --token, --oauth, or --registry.

Import from environment variables

If you already have API keys set in your shell, sbx secret import reads them and stores them in the keychain without typing each value manually:

$ sbx secret import

This scans your current session for the environment variables in the built-in services table below and prompts you to confirm each one before writing. To import a single service:

$ sbx secret import openai

Pass --all to import everything without prompting (new entries only; existing entries are left unchanged), or --force to overwrite existing entries:

$ sbx secret import --all
$ sbx secret import openai --force

Pass --dry-run to preview what would be imported without writing anything. Run sbx secret ls afterwards to confirm what's stored. For setting up credentials in CI, see CI and headless use.

Built-in services

Each built-in service name maps to the environment variables sbx secret import reads and the API domains the proxy injects credentials into:

When you store a secret with sbx secret set <service>, the proxy injects it into requests to the listed API domains.

Services declared by kits

Custom kits can declare their own service identifiers in spec.yaml. In schemaVersion: "2", credentials are declared under the credentials: list:

credentials:
  - service: my-service
    apiKey:
      name: MY_SERVICE_TOKEN
      proxyManaged: true
      inject:
        - domain: api.my-service.com
          scheme: bearer

Each service declares apiKey, oauth, or both. When both resolve at runtime, the API key takes precedence and OAuth acts as the fallback. To provide the credential value, run sbx secret set with the same identifier the kit declares:

$ sbx secret set my-service

There's no separate registration step; the keychain entry is keyed on the identifier the kit already uses. See Authenticate to external services for the full kit-side wiring.

List and remove secrets

List all stored secrets:

$ sbx secret ls
SCOPE      TYPE      NAME      SECRET
(global)   service   github    gho_GCaw4o****...****43qy

Remove a secret:

$ sbx secret rm github

Note

Running sbx reset deletes all stored secrets along with all sandbox state. You'll need to re-add your secrets after a reset.

GitHub token

The github service gives the agent access to the gh CLI inside the sandbox. Resolve your existing GitHub CLI token on the host:

$ sbx secret set github --command 'gh auth token'

The daemon refreshes the token from the host command after the default cache period. This is useful for agents that create pull requests, open issues, or interact with GitHub APIs on your behalf.

SSH agent

If your host has an SSH agent and SSH_AUTH_SOCK is set, Docker Sandboxes forwards the agent into the sandbox and sets SSH_AUTH_SOCK there. The private keys stay on your host. Processes inside the sandbox can request signatures from the forwarded agent, but they can't read or copy the private key.

Use SSH agent forwarding for Git operations over SSH and SSH-based commit signing. The signing key must be loaded in the host SSH agent for sandboxed commit signing to work. Outbound SSH connections are still subject to sandbox network policy. For details, see Commit signing.

Custom secrets

Important

Custom secrets are experimental. Behavior, flags, and the placeholder format may change without notice.

For credentials that don't fit the service-identifier model — for example, when an agent validates the environment variable format at boot, or when the credential lands in a request body rather than a header — use sbx secret set-custom. The secret is keyed on one or more target domains, an environment variable name, and an optional placeholder string, instead of a service identifier. Custom secrets are global by default. Pass --sandbox to scope one to a specific sandbox.

$ sbx secret set-custom \
    --host api.example.com \
    --env API_KEY \
    --value <secret>

Repeat --host to cover multiple domains with the same secret — useful when an API is split across related hostnames or when two unrelated endpoints share a credential:

$ sbx secret set-custom \
    --host api.example.com \
    --host uploads.example.com \
    --env API_KEY \
    --value <secret>

A --host value can also use wildcards, with the same syntax as network rules: * matches a single label (*.example.com covers api.example.com) and ** matches any number (**.example.com covers api.example.com and v2.api.example.com).

Custom secrets also accept dynamic secret sources. Replace --value with either --ref or --command:

$ sbx secret set-custom \
    --host api.example.com \
    --env API_KEY \
    --ref 'op://Work/Example/credential'

Dynamic custom secrets resolve on demand by default. Pass --refresh with a duration to cache the resolved value. The verification and error-output flags work the same as they do for service secrets. --ref and --command can't be combined with --value or --token.

Warning

Passing the secret as --value <secret> records it in your shell history and exposes it to other processes running as your user. Avoid pasting real credentials inline — read the value from a variable that's already in your environment, and clear shell history if a real secret was passed on the command line.

Inside the sandbox, API_KEY is set to a generated placeholder (for example, sbx-cs-<rand>). When a sandboxed process sends a request to any of the configured hosts and the placeholder appears anywhere in the request, the proxy replaces it with the real value. The agent never sees the real secret.

Prefer the service-based flow whenever it's an option — the kit handles the wiring; you only provide the value.

Credential bindings

A credential bindings file records which credential mechanisms and domains you've approved for each service. It lives at ~/.config/sbx/credentials.yaml, or %APPDATA%\sbx\credentials.yaml on Windows.

Third-party kits that declare schemaVersion: "2" require an approved binding for each credential they use. sbx creates one interactively the first time you run such a kit (see First-run approval); you can also write entries by hand. Credentials declared only by embedded, built-in kits are authorized by provenance and don't need a binding.

Each entry under bindings is keyed by a service identifier and approves one or both credential mechanisms:

• apiKey — approves injecting the service's stored API key. The value comes from the secret store (sbx secret set <service>); the binding records approval, it doesn't hold or locate the value.
• oauth — approves the OAuth flow for the service. You sign in on the host, and the proxy handles token refresh and routing. OAuth domains include the token endpoint host and any resource hosts declared by the kit.

Each mechanism takes a domains list that records the domains you approved. sbx asks for approval when a kit requests domains that the existing binding doesn't cover.

bindings:
  anthropic:
    apiKey:
      domains: [api.anthropic.com]
  github:
    apiKey:
      domains: [api.github.com, github.com]

A binding is only an approval record: the presence of apiKey or oauth authorizes that mechanism. Declining a credential writes no entry at all. The real credential isn't stored in this file.

First-run approval

When a third-party kit needs a credential that has no binding, sbx walks you through approving one. For an API key, you can use a value already in the secret store or enter one at the prompt. For OAuth, you approve the sign-in flow. In both cases, you approve the domains declared by the kit. sbx writes the entry to credentials.yaml.

In non-interactive contexts (CI or --detached), there's no one to answer the prompt. Without a binding, the sandbox starts with the credential withheld. If the kit marks the credential as required: true, sbx also prints a warning. Pre-create the binding by running the kit interactively once or by writing credentials.yaml directly before running unattended.

The bindings file gates whether a third-party v2 kit can use a service credential. The kit's credential injection rules and network permissions still constrain which requests can carry the credential.

Kits that require a binding

Only third-party kits that declare schemaVersion: "2" require a binding. Built-in agents also use schemaVersion: "2", but credentials declared only by embedded kits are authorized by provenance and inject automatically. A third-party kit that extends a built-in agent inherits its credentials, but not its built-in provenance. The inherited credentials therefore require approval. If a third-party kit declares the same service itself, that service also requires approval. Kits on schemaVersion: "1" inject their declared credentials without a binding.

Warning

Proxy-managed OAuth isn't supported for third-party sandbox agents, including kits that extend a built-in agent. Repeating the parent's OAuth declaration in the child kit doesn't activate OAuth interception. Use a stored API key when the service supports one. Otherwise, an OAuth login performed inside the sandbox stores the real token there.

Registry credentials

Registry credentials authenticate to private OCI registries when pulling templates or kits, and can also let the agent pull and push images from inside the sandbox through the host-side proxy. Use sbx secret set --registry <host> to store them. For Docker Hub, sbx reuses your sbx login session — no registry secret needed. For other registries (GitHub Container Registry, ECR, ACR, self-hosted Nexus, and so on), store credentials with sbx secret set --registry.

Choose the scope by adding --all-sandboxes, adding --sandbox SANDBOX, or passing neither:

sbx secret set [--all-sandboxes | --sandbox SANDBOX] --registry HOST

• Host-only (no scope flag): the sbx CLI uses it to pull templates and kits when creating a sandbox. The credential stays on the host and is never available inside the sandbox.
• All sandboxes (--all-sandboxes): same as host-only, plus the proxy authenticates registry login requests from sandboxes. The credential stays on the host and is never written to the sandbox filesystem. Use it when agents build and publish container images.
• Sandbox-scoped (--sandbox SANDBOX): same proxy behavior as --all-sandboxes, but only for the named sandbox. Use it when only one sandbox needs registry access.

Store registry credentials

Pipe a token from stdin and target the registry hostname:

$ gh auth token | sbx secret set --registry ghcr.io --password-stdin

For registries that require a username (for example, ACR with an admin account), add --username:

$ echo "$ACR_PASSWORD" | sbx secret set \
    --registry myregistry.azurecr.io \
    --username myuser \
    --password-stdin

Add --all-sandboxes to make the credential available to every new sandbox:

$ gh auth token | sbx secret set --all-sandboxes --registry ghcr.io --password-stdin
$ sbx run claude

Store all-sandboxes registry credentials before creating a sandbox. Existing sandboxes don't pick up all-sandboxes registry credentials added later. To add registry access to an existing sandbox, use a sandbox-scoped credential instead.

To scope the credential to a single sandbox, store it under that sandbox's name:

$ gh auth token | sbx secret set --sandbox my-app --registry ghcr.io --password-stdin

sbx kit pull also uses these credentials, with the Docker credential store as a fallback. sbx kit push uses only the Docker credential store — push targets still require a prior docker login.

Remove registry credentials

Remove both the host-only and all-sandboxes entries for a registry:

$ sbx secret rm --registry ghcr.io -f

To remove only the all-sandboxes entry and leave the host-only credential in place, pass --all-sandboxes:

$ sbx secret rm --all-sandboxes --registry ghcr.io -f

To remove a sandbox-scoped credential, pass the sandbox name:

$ sbx secret rm --sandbox my-sandbox --registry ghcr.io -f

Best practices

• Use stored secrets to provide credentials. The OS keychain protects them at rest; on Linux hosts without a keychain they are held in a permission-protected file instead. See Where secrets are stored.
• Don't set API keys manually inside the sandbox. Sandbox agents are pre-configured to use proxy-managed credentials.
• Registry credentials stay on the host and are injected by the proxy when a sandbox authenticates to the registry. Reserve them for sandboxes that need registry access, and prefer sandbox scope over --all-sandboxes to limit exposure.
• Several agents support OAuth as another secure option: the flow runs on the host, so the token is never exposed inside the sandbox. If you haven't stored a credential, the agent prompts you to authenticate — Codex prompts on the host from sbx run codex, while Claude Code, Cursor, and Droid prompt interactively inside the sandbox. To authenticate ahead of time, run sbx secret set openai --oauth for Codex or use /login inside Claude Code; Cursor and Droid have no ahead-of-time option, so their sign-in prompt appears when the agent starts. See the individual agent pages for each agent's flow.
• If you store credentials in 1Password or AWS Secrets Manager, see Sourcing credentials from 1Password and Sourcing credentials from AWS Secrets Manager.

Custom templates and placeholder values

When building custom templates or installing agents manually in a shell sandbox, some agents require environment variables like OPENAI_API_KEY to be set before they start. Set these to placeholder values (e.g. proxy-managed) if needed. The proxy injects actual credentials regardless of the environment variable value.