OpenWork Enterprise Security and Outbound Access Requirements: Complete Guide

OpenWork Enterprise enforces security through a repository-wide security policy and strictly controls outbound network access via a machine-readable allowlist.

Enterprise deployments of the different-ai/openwork repository must satisfy two complementary requirement sets: a comprehensive security policy that governs secret handling and vulnerability management, and an outbound-network allowlist that restricts external connections to pre-approved hosts only. This guide breaks down both requirement categories with specific implementation details from the source code.

---

Security Policy Requirements

Every OpenWork Enterprise deployment must comply with the repository-wide security policy defined in SECURITY.md. This policy establishes three mandatory practices that harden the codebase against common attack vectors.

No Secrets in Source Code or Logs

The policy strictly prohibits exposure of secrets, API keys, or credentials in any source code, configuration files, or log output. This applies to:

• Hardcoded tokens in TypeScript/JavaScript files
• Environment variable dumps in error handlers
• Debug logging that might capture sensitive headers

CI-Based Security Checks

All pull requests and releases must pass automated security validations. The CI pipeline runs:

• Secret-scanning to detect accidentally committed credentials
• Dependency-vulnerability audits to flag known CVEs in npm packages
• Enterprise-readiness gate validation through mock-server tests

Enterprise-Ready MCP Tool Validation

The most distinctive security requirement is the "Enterprise-ready" readiness gate. This gate validates that every published MCP (Model Context Protocol) tool set meets enterprise criteria defined in the mock-server test suite.

In packages/enterprise-mcp-mock-server/test/probe-boundaries.test.ts, the test suite enforces this gate:

// packages/enterprise-mcp-mock-server/test/probe-boundaries.test.ts
test("bounded tool schemas support nested enterprise payloads and reject hostile schema keys", () => {
  // …test logic…
  // If the tool catalog is empty, the probe fails with an enterprise‑readiness error
  if (toolNames.size === 0) {
    throw new ProbeFailure("MCP_TOOL_DISCOVERY", "catalog_empty", "Enterprise readiness not satisfied");
  }
});

Tools that fail this probe cannot be published for enterprise use, ensuring only hardened, bounded tool schemas reach production environments.

---

Outbound Network Access Requirements

OpenWork Enterprise operates under a default-deny network posture. Instances may only initiate outbound connections to hosts explicitly listed in the machine-readable allowlist.

The Allowlist Structure

The canonical source of truth is docs/enterprise/outbound-access.json, a JSON file containing an array of permitted host patterns. The human-readable companion docs/enterprise/outbound-access.md documents the policy rationale and update procedures.

Example structure of the allowlist:

{
  "hosts": [
    "api.openai.com",
    "*.anthropic.com",
    "registry.npmjs.org"
  ],
  "version": "2024.06.01",
  "last_reviewed": "2024-06-01T00:00:00Z"
}

CI Enforcement via check-outbound-access.mjs

The script scripts/check-outbound-access.mjs parses the allowlist and validates that no code attempts connections to unapproved hosts. This script runs in CI and fails the build on violations:

// scripts/check-outbound-access.mjs (excerpt)
import fs from "fs";
import path from "path";

const manifestPath = path.join(repoRoot, "docs/enterprise/outbound-access.json");
const json = JSON.parse(fs.readFileSync(manifestPath, "utf‑8"));
if (!Array.isArray(json.hosts)) {
  throw new Error("docs/enterprise/outbound-access.json must contain a hosts array");
}

The script typically integrates with:

• Static analysis of HTTP client calls (fetch, axios, node-fetch)
• Runtime network interception in test environments
• Container egress policies for deployed applications

Violation Consequences

Any attempt to contact a host not present in outbound-access.json causes:

• Immediate CI check failure
• Blocked merge of the offending pull request
• Audit log entry for security review

---

Key Implementation Files

File	Purpose
SECURITY.md	Repository-wide security policy: secrets handling, vulnerability checks, enterprise-readiness criteria
docs/enterprise/outbound-access.md	Human-readable outbound access policy documentation
docs/enterprise/outbound-access.json	Machine-readable allowlist of permitted external hosts
scripts/check-outbound-access.mjs	CI validation script that enforces the outbound allowlist
packages/enterprise-mcp-mock-server/test/probe-boundaries.test.ts	Test suite implementing the enterprise-ready MCP tool gate

---

Deployment Checklist

Use this checklist when preparing an OpenWork Enterprise deployment:

1. Verify SECURITY.md compliance across all custom code
2. Confirm no secrets in environment variables beyond secure secret management
3. Review docs/enterprise/outbound-access.json against your infrastructure needs
4. Submit pull request to add any required hosts to the allowlist with business justification
5. Ensure CI passes check-outbound-access.mjs and probe-boundaries.test.ts
6. Document any exceptions with security team approval

---

Summary

• Security policy in SECURITY.md mandates secret-free code, CI security checks, and enterprise-ready MCP tool validation
• Outbound access is restricted to hosts listed in docs/enterprise/outbound-access.json, enforced by scripts/check-outbound-access.mjs
• Enterprise readiness gate in the mock-server test suite ensures only bounded, secure MCP tools reach production
• Both requirements operate as default-deny policies: violations block deployment

---

Frequently Asked Questions

What happens if my code needs to connect to a host not in the allowlist?

Submit a pull request modifying docs/enterprise/outbound-access.json to add the required host. Include business justification and security review approval. The change takes effect after CI passes and maintainers merge. Until then, scripts/check-outbound-access.mjs will fail your build.

How does the enterprise-readiness gate differ from standard security scanning?

Standard security scanning detects known vulnerabilities and secrets. The enterprise-readiness gate specifically validates MCP tool schemas—ensuring they support nested enterprise payloads, reject hostile keys, and maintain bounded operation scopes. This is implemented in packages/enterprise-mcp-mock-server/test/probe-boundaries.test.ts as custom test logic, not generic tooling.

Can we disable outbound access restrictions for development environments?

No. The check-outbound-access.mjs script runs in all CI contexts. For local development, you can modify docs/enterprise/outbound-access.json temporarily, but such changes cannot be merged without security review. The policy intentionally prevents environment-based variance that could mask production issues.

Where are security incidents reported for OpenWork Enterprise?

Follow the disclosure process in SECURITY.md. The policy defines contact methods for reporting vulnerabilities and the timeline for maintainer response. Enterprise deployments should also monitor the repository for security advisories and update promptly when patches release.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}