@mandate-os/openclaw

MandateOS bridge, plugin bundle, and installer for OpenClaw.

This package provides:

• mandate-os-openclaw-install to configure OpenClaw for a guarded workspace
• mandate-os-openclaw-bridge for policy evaluation inside the OpenClaw plugin flow
• the bundled OpenClaw extension assets that enable the guarded workflow

Environment

The installer expects:

• MANDATE_OS_BASE_URL

Optional during install:

• MANDATE_OS_MCP_DEFAULT_MANDATE_ID
• MANDATE_OS_OPENCLAW_SANDBOX_MODE (all or off)

Required at OpenClaw runtime:

• MANDATE_OS_AGENT_TOKEN

One-Command Install

MANDATE_OS_BASE_URL=https://mandateos.example.com \
MANDATE_OS_AGENT_TOKEN='key_id.secret' \
MANDATE_OS_MCP_DEFAULT_MANDATE_ID='mdt_123' \
MANDATE_OS_OPENCLAW_WORKSPACE_PATH=/absolute/path/to/your/repo \
npx --yes --package @mandate-os/openclaw@latest mandate-os-openclaw-install install

That command:

• installs the MandateOS plugin bundle under the local OpenClaw state directory
• configures the MandateOS MCP entry for the guarded workspace
• creates or updates the mandateos_guarded agent profile

To install the guarded agent with sandboxing disabled, pass either:

MANDATE_OS_OPENCLAW_SANDBOX_MODE=off \
npx --yes --package @mandate-os/openclaw@latest mandate-os-openclaw-install install

or:

npx --yes --package @mandate-os/openclaw@latest mandate-os-openclaw-install install --sandbox-mode=off

If you omit the option, fresh installs default to all, while existing guarded agents keep their current sandbox setting.

Inspect the current install state with:

MANDATE_OS_OPENCLAW_WORKSPACE_PATH=/absolute/path/to/your/repo \
npx --yes --package @mandate-os/openclaw@latest mandate-os-openclaw-install status

Repair an existing OpenClaw install in place with a clean MandateOS asset reset:

MANDATE_OS_BASE_URL=https://mandateos.example.com \
MANDATE_OS_MCP_DEFAULT_MANDATE_ID='mdt_123' \
MANDATE_OS_OPENCLAW_WORKSPACE_PATH=/absolute/path/to/your/repo \
npx --yes --package @mandate-os/openclaw@latest mandate-os-openclaw-install repair

That repair flow removes the MandateOS-owned OpenClaw plugin and bundle directories plus local MandateOS approval/status cache files, then reinstalls the current assets and repairs the MandateOS config entries. It preserves existing guarded-agent overrides such as mandateos_guarded.sandbox.mode = off. You can also change the guarded-agent sandbox mode during repair with --sandbox-mode=all|off.

Runtime note

OpenClaw must still start with MANDATE_OS_AGENT_TOKEN in its runtime environment so the bridge can evaluate actions against MandateOS policy.

Interpreting status

status is intentionally narrower than doctor. Read these sections independently:

• installHealth: local install health for the plugin, bundle, MCP entry, bridge runtime, and guarded agent wiring
• runtimeAuthorization: whether the live OpenClaw runtime has MANDATE_OS_AGENT_TOKEN
• wrapperExposureVerification: whether a live OpenClaw session exposed the MandateOS wrapper tools
• livePolicyCapability: whether base URL, runtime token, and default mandate are all present for live-policy checks

That split matters because these states can diverge. Example: install health can be good while runtime authorization is still missing, or wrapper exposure can be unverified even though the plugin registered successfully.

Interpreting doctor

doctor adds smoke tests on top of status:

• Install status checks the local install footprint.
• Runtime token confirms runtime authorization input.
• Local bridge smoke test confirms the bridge can execute locally.
• Live policy smoke test requires MANDATE_OS_BASE_URL, MANDATE_OS_AGENT_TOKEN, and MANDATE_OS_MCP_DEFAULT_MANDATE_ID.

If you want live OpenClaw policy evaluation, the backing API key used for MANDATE_OS_AGENT_TOKEN must include simulate:write.