Tork Guardian for OpenClaw

OpenClaw is powerful. Tork makes it safe.

Enterprise-grade security and governance layer for OpenClaw agents. Detect PII, enforce policies, generate compliance receipts, control tool access, and scan skills for vulnerabilities before installation.

Installation

npm install @torknetwork/guardian

Quick Start

import { TorkGuardian } from '@torknetwork/guardian';

const guardian = new TorkGuardian({
  apiKey: process.env.TORK_API_KEY!,
});

// Govern an LLM request before sending
const result = await guardian.governLLM({
  messages: [
    { role: 'user', content: 'Email [email protected] about the project' },
  ],
});
// PII is redacted: "Email [EMAIL_REDACTED] about the project"

// Check if a tool call is allowed
const decision = guardian.governTool({
  name: 'shell_execute',
  args: { command: 'rm -rf /' },
});
// { allowed: false, reason: 'Blocked shell command pattern: "rm -rf"' }

Regional PII Detection (v1.1)

Detect country-specific PII patterns across 13 regions and 3 industries. Configure region and industry in your Guardian config or pass them per-call.

Config-level (applies to all calls)

const guardian = new TorkGuardian({
  apiKey: process.env.TORK_API_KEY!,
  pii: {
    region: ['ae', 'sa'],        // UAE + Saudi Arabia
    industry: 'finance',          // Enable finance-specific patterns
  },
});

// All govern/redact calls now include regional detection
const result = await guardian.redactPII('Emirates ID 784-1234-1234567-1');

Per-call (standalone functions)

import { redactPII } from '@torknetwork/guardian';

const result = await redactPII('tork_...', 'Aadhaar 1234 5678 9012', {
  region: ['in'],
  industry: 'healthcare',
});

Available Regions

au us gb eu ae sa ng in jp cn kr br

Available Industries

healthcare finance legal

Network Security

Tork Guardian governs all network activity — port binds, outbound connections, and DNS lookups — with SSRF prevention, reverse shell detection, and per-skill rate limiting.

Using the network handler

const guardian = new TorkGuardian({
  apiKey: process.env.TORK_API_KEY!,
  networkPolicy: 'default',
});

const network = guardian.getNetworkHandler();

// Validate a port bind
const bind = network.validatePortBind('my-skill', 3000, 'tcp');
// { allowed: true, reason: 'Port 3000/tcp bound' }

// Validate an outbound connection
const egress = network.validateEgress('my-skill', 'api.openai.com', 443);
// { allowed: true, reason: 'Egress to api.openai.com:443 allowed' }

// Validate a DNS lookup (flags raw IPs)
const dns = network.validateDNS('my-skill', 'api.openai.com');
// { allowed: true, reason: 'DNS lookup for api.openai.com allowed' }

// Get the full activity log for compliance
const log = network.getActivityLog();

// Get a network report with anomaly detection
const report = network.getMonitor().getNetworkReport();

Standalone functions

import { validatePortBind, validateEgress, validateDNS } from '@torknetwork/guardian';

const config = { apiKey: 'tork_...', networkPolicy: 'strict' as const };

validatePortBind(config, 'my-skill', 3000, 'tcp');
validateEgress(config, 'my-skill', 'api.openai.com', 443);
validateDNS(config, 'my-skill', 'api.openai.com');

Switching policies

// Default — balanced for dev & production
const guardian = new TorkGuardian({
  apiKey: 'tork_...',
  networkPolicy: 'default',
});

// Strict — enterprise lockdown (443 only, explicit domain allowlist)
const guardian = new TorkGuardian({
  apiKey: 'tork_...',
  networkPolicy: 'strict',
});

// Custom — override any setting
const guardian = new TorkGuardian({
  apiKey: 'tork_...',
  networkPolicy: 'custom',
  allowedOutboundPorts: [443, 8443],
  allowedDomains: ['api.myservice.com'],
  maxConnectionsPerMinute: 30,
});

See docs/NETWORK-SECURITY.md for full details on threat coverage, policy comparison, and compliance receipts.

Example Configs

Pre-built configurations for common environments:

import {
  MINIMAL_CONFIG,
  DEVELOPMENT_CONFIG,
  PRODUCTION_CONFIG,
  ENTERPRISE_CONFIG,
} from '@torknetwork/guardian';

| Config | Policy | Network | Description | |--------|--------|---------|-------------| | MINIMAL_CONFIG | standard | default | Just an API key, all defaults | | DEVELOPMENT_CONFIG | minimal | default | Permissive policies, full logging | | PRODUCTION_CONFIG | standard | default | Blocked exfil domains (pastebin, ngrok, burp) | | ENTERPRISE_CONFIG | strict | strict | Explicit domain allowlist, 20 conn/min, TLS only |

import { TorkGuardian, PRODUCTION_CONFIG } from '@torknetwork/guardian';

const guardian = new TorkGuardian({
  ...PRODUCTION_CONFIG,
  apiKey: process.env.TORK_API_KEY!,
});

Configuration

const guardian = new TorkGuardian({
  // Required
  apiKey: 'tork_...',

  // Optional
  baseUrl: 'https://tork.network',   // API endpoint
  policy: 'standard',                     // 'strict' | 'standard' | 'minimal'
  redactPII: true,                        // Enable PII redaction

  // Shell command governance
  blockShellCommands: [
    'rm -rf', 'mkfs', 'dd if=', 'chmod 777',
    'shutdown', 'reboot',
  ],

  // File access control
  allowedPaths: [],                        // Empty = allow all (except blocked)
  blockedPaths: [
    '.env', '.env.local', '~/.ssh',
    '~/.aws', 'credentials.json',
  ],

  // Network governance
  networkPolicy: 'default',               // 'default' | 'strict' | 'custom'
  allowedInboundPorts: [3000, 8080],       // Ports skills may bind to
  allowedOutboundPorts: [443],             // Ports for outbound connections
  allowedDomains: ['api.openai.com'],      // If non-empty, only these domains are allowed
  blockedDomains: ['evil.com'],            // Domains always blocked
  maxConnectionsPerMinute: 60,             // Per-skill egress rate limit

  // API failure handling (see docs/FAILURE-MODES.md)
  failureMode: 'closed',                   // 'closed' (default) | 'open'
  onFailure: (error) => {},                // Called when failureMode 'open' allows a call through
});

The port, domain, and rate-limit fields apply on top of whichever base networkPolicy you pick — e.g. networkPolicy: 'strict' plus your own allowedDomains keeps the strict lockdown but uses your allowlist.

Failure Modes

Auth failures never fail open. If the API returns 401 or 403 (missing, invalid, expired, or revoked key), the SDK throws TorkAuthError — governance did not run and the SDK will not pretend it did. Other 4xx responses throw TorkRequestError. This is not configurable.

Outages are your choice. For 5xx responses, timeouts, and connection failures, the default failureMode: 'closed' throws TorkUnavailableError. Opt into failureMode: 'open' to allow requests through during Tork outages; every allowed-through call is reported to your onFailure callback.

const guardian = new TorkGuardian({
  apiKey: process.env.TORK_API_KEY!,
  failureMode: 'open',                       // opt-in; default is 'closed'
  onFailure: (error) => alerting.warn('Tork unavailable, call ungoverned', error),
});

Versions ≤ 1.0.2 failed open on all errors, including auth failures. Upgrade — see docs/FAILURE-MODES.md.

Policies

| Policy | PII | Shell | Files | Network | |--------|-----|-------|-------|---------| | strict | Deny on detection | Block all | Whitelist only | Block all | | standard | Redact | Block dangerous | Block sensitive | Allow | | minimal | Redact | Allow all | Allow all | Allow all |

Standalone Functions

import { redactPII, generateReceipt, governToolCall } from '@torknetwork/guardian';

// Redact PII from text
const result = await redactPII('tork_...', 'Call 555-123-4567');

// Generate a compliance receipt
const receipt = await generateReceipt('tork_...', 'Processed user data');

// Check a tool call against policy
const decision = governToolCall(
  { name: 'file_write', args: { path: '.env' } },
  { policy: 'standard', blockedPaths: ['.env'] }
);

Security Scanner

Scan any OpenClaw skill for vulnerabilities before installing it. The scanner checks for 14 security patterns across code and network categories.

CLI

# Scan a skill directory
npx tork-scan ./my-skill

# Full details for every finding
npx tork-scan ./my-skill --verbose

# JSON output for CI/CD
npx tork-scan ./my-skill --json

# Fail on any high or critical finding
npx tork-scan ./my-skill --strict

Programmatic

import { SkillScanner, generateBadge } from '@torknetwork/guardian';

const scanner = new SkillScanner();
const report = await scanner.scanSkill('./my-skill');

console.log(`Risk: ${report.riskScore}/100`);
console.log(`Verdict: ${report.verdict}`); // 'verified' | 'reviewed' | 'flagged'

See docs/SCANNER.md for the full rule reference, severity weights, and example output.

Tork Verified Badges

Skills that pass the security scanner receive a Tork Verified badge:

| Badge | Score | Meaning | |-------|-------|---------| | Tork Verified (green) | 0 - 29 | Safe to install | | Tork Reviewed (yellow) | 30 - 49 | Manual review recommended | | Tork Flagged (red) | 50 - 100 | Security risks detected |

import { SkillScanner, generateBadge, generateBadgeMarkdown } from '@torknetwork/guardian';

const scanner = new SkillScanner();
const report = await scanner.scanSkill('./my-skill');
const badge = generateBadge(report);

// Add to your README
console.log(generateBadgeMarkdown(badge));

Get Your API Key

Sign up at tork.network to get your API key.