> ## Documentation Index
> Fetch the complete documentation index at: https://docs.grantex.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# OpenClaw

> Install Grantex Agent Skills in an OpenClaw workspace and use the JSON CLI for delegated authorization.

OpenClaw can load the same portable Grantex skills used by other Agent Skills-compatible hosts. No OpenClaw-specific Grantex SDK is required.

Published `@grantex/cli@0.3.0` includes the one-command OpenClaw skill installer.

## Install in a workspace

Run from the configured OpenClaw agent workspace:

```bash theme={null}
npm install -g @grantex/cli@0.3.0
grantex agent install --target openclaw
openclaw skills check
```

The installer writes both skills to `<workspace>/skills`. Start a new OpenClaw turn after installation.

<Info>
  OpenClaw loads Agent Skills-compatible folders from its workspace and managed skill roots. See the [OpenClaw skills reference](https://docs.openclaw.ai/skills).
</Info>

To install for all local OpenClaw agents instead, select the managed skill root explicitly:

<CodeGroup>
  ```bash Bash theme={null}
  grantex agent install --dir "$HOME/.openclaw/skills"
  openclaw skills check
  ```

  ```powershell PowerShell theme={null}
  grantex agent install --dir "$HOME\.openclaw\skills"
  openclaw skills check
  ```
</CodeGroup>

## Configure

Expose credentials to the OpenClaw host with its secret or environment configuration:

<CodeGroup>
  ```bash Bash theme={null}
  export GRANTEX_URL=https://api.grantex.dev
  export GRANTEX_KEY=YOUR_API_KEY
  grantex --json me
  ```

  ```powershell PowerShell theme={null}
  $env:GRANTEX_URL = "https://api.grantex.dev"
  $env:GRANTEX_KEY = "YOUR_API_KEY"
  grantex --json me
  ```
</CodeGroup>

Never embed API keys or grant tokens in `SKILL.md`, workspace instructions, prompts, or logs.

## Use

Ask OpenClaw to use `use-grantex-cli` for operational tasks or `integrate-grantex` for code changes. The skills instruct it to:

* request least-privilege scopes;
* preserve the human consent step;
* consume JSON instead of human tables;
* keep tokens out of process arguments; and
* distinguish a CLI preflight from protected-service enforcement.

Example shell calls:

```text theme={null}
grantex --json grants list --status active
grantex verify --env GRANTEX_GRANT_TOKEN --json
grantex --json enforce test --token-env GRANTEX_GRANT_TOKEN --connector salesforce --tool create_lead
```

## Trust boundary

An autonomous OpenClaw process must not be its own final authorization boundary. Verify the grant token, exact scope, and required current state inside the service, connector, or gateway before the side effect occurs.

See [Agent CLIs & Skills](/integrations/agent-cli) for the full shell contract and [Scope Enforcement](/guides/scope-enforcement) for production enforcement.