Sentinel Plugin for OpenClaw

AI agent security platform: prompt injection defence + secret/credential scanning at the agent runtime layer. For OpenClaw agents. Monitors tool calls, enforces security policies, scans inbound content for injection attacks, scans tool outputs for leaked credentials, and sends real-time alerts.

Quick Start

1. Install the plugin:

   openclaw plugins install @sentinel-agents/sentinel

2. Restart OpenClaw, then run /sentinel to verify the plugin is active.

3. (Optional) Set your Pro licence key:

   openclaw config set plugins.entries.sentinel.config.SENTINEL_LICENCE_KEY <your-key>

   Then restart the gateway:

   openclaw gateway restart

   Alternatively, process.env.SENTINEL_LICENCE_KEY works as a fallback (e.g. in .env files or system environment).

   The plugin works fully without a key -- Pro features are disabled until a key is added.

Commands

• /sentinel -- overview and recent tool calls
• /sentinel status -- detailed status including licence and detections
• /sentinel alerts list -- show configured alert destinations
• /sentinel alerts add <type> <url> -- add an alert destination
• /sentinel alerts test -- send a test alert to all destinations
• /sentinel upgrade -- check for updates
• /sentinel config -- show all config values and sources
• /sentinel config set <key> <value> -- set a config value
• /sentinel config get <key> -- show a single config value
• /sentinel config reset <key> -- reset a config value to default
• /sentinel --blocks [N] -- show last N blocked calls with timestamp, tool, reason, and truncated params (default 5)

Configuration

There are two ways to configure Sentinel:

Interactive (recommended):

/sentinel config set <key> <value>

Environment variables (CI/Docker):

export SENTINEL_DASHBOARD=true
export SENTINEL_DASHBOARD_PORT=3099

Environment variables always take priority over the config file.

| Key | Env Var | Default | Description | |-----|---------|---------|-------------| | dashboard | SENTINEL_DASHBOARD | false | Enable web dashboard | | dashboardPort | SENTINEL_DASHBOARD_PORT | 3099 | Dashboard port | | shareLogs | SENTINEL_SHARE_LOGS | false | Share anonymised telemetry | | scanPath | SENTINEL_SCAN_PATH | (auto) | Path to sentinel-scan binary | | licenceKey | SENTINEL_LICENCE_KEY | (none) | Your licence key |

Config file location:

• Linux/macOS: ~/.config/sentinel/sentinel.json
• Windows: %APPDATA%\sentinel\sentinel.json

Dashboard

The dashboard provides a web UI for monitoring Sentinel activity.

• Disabled by default -- enable with /sentinel config set dashboard true
• Access at http://localhost:3099
• API token: auto-generated on first enable and saved to ~/.config/sentinel/dashboard-token
• Set SENTINEL_API_TOKEN env var to use a fixed token (recommended for persistent setups)

Block History API (v0.7.5+)

The dashboard exposes a paginated endpoint for retrieving blocked call history:

GET /api/sentinel/blocked?page=1&limit=50

• Default limit: 50 (max 200)
• Response: JSON array of blocked calls with timestamp, tool name, block reason, and redacted params
• Auth: requires SENTINEL_API_TOKEN (same as dashboard)

Use /sentinel --blocks [N] from the chat to quickly inspect recent blocks without opening the dashboard.

Secret Scanning (v0.8.0)

Runtime scanning of tool outputs for leaked credentials. Detects AWS keys, GitHub tokens, Stripe secrets, private keys, connection strings and 14 more patterns — redacted before they enter your LLM context.

Configuration:

{
  "secretScanning": {
    "enabled": true,
    "strictness": "standard",
    "scanPoints": { "toolOutput": true },
    "actions": { "toolOutput": "redact_and_warn" },
    "allowlist": [],
    "customPatterns": []
  }
}

Strictness levels:

• relaxed — regex only, lowest false positives
• standard — regex + entropy with context (recommended)
• strict — highest sensitivity

Commands:

• /sentinel --secrets — view recent detections + scanning status
• /sentinel --secrets stats — view 24h detection counts by type

Secret scanning is disabled by default — zero behaviour change on upgrade from 0.7.x. Outbound response scanning coming in a future release.

Inbound Scanning (optional)

For inbound content scanning, install the Python scanner:

pip install sentinel-security

The plugin works without it -- inbound scanning is simply disabled until the binary is available.

Windows

Installation

openclaw plugins install may fail with spawn EINVAL on some Windows setups. Workaround:

1. Open a terminal and navigate to the plugins directory:

   cd %APPDATA%\openclaw\plugins

2. Clone or copy the plugin manually:

   git clone https://github.com/satvoop/Sentinel.git sentinel
   cd sentinel\plugin
   npm install

3. Restart OpenClaw. The plugin will load normally from that point.

Python for inbound scanning

sentinel-scan requires Python 3.8+. The Python Scripts directory must be in your PATH:

%USERPROFILE%\AppData\Local\Programs\Python\Python3X\Scripts\

Replace Python3X with your installed version (e.g. Python312).

Setting environment variables

PowerShell (current session):

$env:SENTINEL_LICENCE_KEY = "your-key"
$env:SENTINEL_DASHBOARD = "true"

Permanent (via System Properties):

1. Open Start > "Edit the system environment variables"
2. Click "Environment Variables..."
3. Add or edit under User variables

Node.js

Node.js 18+ is required. Download from https://nodejs.org/.

iMessage alerts

iMessage alerts are macOS only and are not available on Windows.

Config file location

Sentinel stores user data (config, alert settings, dashboard token) in:

• Linux/macOS: ~/.config/sentinel/
• Windows: %APPDATA%\sentinel\
• Override: set SENTINEL_CONFIG_DIR env var to use a custom location (useful for Docker/CI)

Plugin ID mismatch warning

The plugin id is sentinel. This matches the OpenClaw config entry key.

Documentation

Full docs: https://sentinel-agents.com/docs

Free Prompt Audit

Scan your prompts for injection vulnerabilities: https://sentinel-agents.com/audit

Support

[email protected]