Skip to content

Security: warterbili/auto_reverse_

Security

.github/SECURITY.md

Security Policy

auto_reverse is offensive-security and reverse-engineering tooling. This policy covers two distinct concerns: how to use the project lawfully, and how to report a vulnerability in the project itself.

Acceptable use

The capabilities here (fuzzers, scanners, exploitation/C2 integrations, anti-bot bypass, instrumentation, packer unpacking) are powerful and easily misused. By using auto_reverse you agree to use it only against systems you own or are explicitly authorized to test:

  • authorized security research,
  • penetration-testing engagements with a signed scope,
  • Capture-The-Flag (CTF) competitions,
  • interoperability and compatibility research,
  • defensive and detection-engineering work.

You are solely responsible for complying with every law and contractual term that applies to your target. The maintainers accept no liability for misuse.

Built-in guardrails (see the brain):

  • The orchestrator desensitizes by policy — real credentials, tokens, and PII must never be written into reports or case records; use placeholders.
  • The orchestrator stops and asks for a human when a target appears to be for unauthorized or illegal use.

Do not open public issues that contain real credentials, private keys, customer data, or unredacted captures of third-party traffic.

Reporting a vulnerability in auto_reverse

If you find a security issue in this project's own code — for example, the setup scripts, fetch.py/doctor.py, an adapter, or a bundled skill that could compromise the host running it (arbitrary code execution, path traversal, unsafe download, secret leakage) — please report it privately:

  1. Preferred: open a GitHub Security Advisory on this repository (Security → Report a vulnerability).
  2. Alternatively, email the maintainers (see the repository profile) with a clear subject line beginning SECURITY:.

Please include: affected file/version, a description, reproduction steps, and impact. Do not open a public issue for an undisclosed vulnerability.

What to expect

  • We aim to acknowledge a report within 5 business days.
  • We will work with you on a fix and a coordinated disclosure timeline.
  • We credit reporters in the release notes unless you ask to remain anonymous.

Supported versions

This project is pre-1.0 and moves fast. Only the latest main is supported; please reproduce any report against the current main before filing.

There aren't any published security advisories