Skip to content

Security: MukundaKatta/agentguard

Security

SECURITY.md

Security policy

agentguard is a security primitive (an egress firewall). If you find a way to bypass its policy enforcement — for example a host pattern that misses, a normalization bug that lets a denied URL through, or a way to drop the fetch wrapper mid-execution — please report it privately.

Reporting a vulnerability

Use GitHub's private vulnerability reporting on this repo. That goes straight to the maintainer; no public issue is opened until a fix is ready.

If GitHub Security Advisories is not available to you, email mukunda.vjcs6@gmail.com with subject [agentguard security]. Include:

  • Affected version
  • A minimal reproducer (URL pattern + policy spec is usually enough)
  • Your proposed fix, if you have one

You will get an acknowledgement within 7 days. A patched release is the target outcome; advisories will be coordinated under a CVE if appropriate.

Scope

In scope:

  • Bypasses of policy() allow/deny matching, including unicode/IDN, IPv6, IPv4 numeric, and CIDR-equivalent forms
  • Race conditions where a request slips out during firewall() setup or teardown
  • Bugs in blockResponse or PolicyViolation that leak the original request URL or headers when the user expected them to be redacted

Out of scope:

  • Vulnerabilities in Node.js core, the user's code, or third-party fetch implementations agentguard wraps. Report those upstream.
  • Performance or feature requests — open a normal issue.

Supported versions

Only the latest published 0.1.x is patched. Earlier preview versions are not supported. Track this file's [Unreleased] section for in-flight fixes.

There aren't any published security advisories