Skip to content

Security: new-world-coder/AgentShield

Security

SECURITY.md

Security Policy

Supported Versions

We maintain and patch versions on the main branch. Please use the latest release for full protection.

Security Guidelines (Cloudflare AI Agent Best Practices)

  • Input Validation: All inputs to the agent are strictly validated and sanitized against prompt injection and malformed payloads.
  • Authentication & Authorization: Access is controlled using Cloudflare Zero Trust, enforcing least-privilege by default.
  • Traffic Proxying: All inbound/outbound agent requests are routed via Cloudflare AI Gateway with cf-aig-authorization header for session control.
  • Encryption: Sensitive states and agent comms are encrypted at rest and in transit.
  • Data Loss Prevention: Integrate with Cloudflare DLP to monitor and prevent leakage of sensitive information.
  • Browser Isolation: Protected endpoints may use Cloudflare’s Browser Isolation for high-risk operations.
  • Continuous Testing: Security test suites run via automated CI processes. All PRs run npx vitest run and npx wrangler dev --test.
  • Supply Chain Security: All dependencies are audited for vulnerabilities before and after deployment.

Vulnerability Reporting

If you discover a potential security issue, please report it via a private GitHub issue or email the maintainers.

There aren't any published security advisories