Skip to content

Security: jaymunshi/patchwing

Security

SECURITY.md

Security Policy

Scope

PatchWing is a defensive-only bug-fixing tool. It fixes defects in code the operator owns or is authorised to fix, and its reproducers stop at Definition A — an observable boundary violation ("the bad outcome happened"), never a weaponised exploit chain. It does not scan, hunt, or probe third-party systems.

The CVEs referenced in the documentation and examples/ are already public and already fixed upstream. They are evaluation targets, not disclosures, and contain no novel exploit information.

Reporting a vulnerability in PatchWing itself

If you find a security issue in PatchWing's own code (for example, a sandbox escape, a way to make the verifier certify a patch that does not close the bug, or a way to poison the evidence bundle), please report it privately:

  • Do not open a public GitHub issue for it.
  • Email the maintainer at the address published on the repository owner's GitHub profile, or use GitHub's private "Report a vulnerability" advisory flow on this repository.
  • Include the version/commit, a minimal reproduction, and the impact you observed.

You will get an acknowledgement, and a fix or a written decision, as fast as a single maintainer reasonably can. Please allow time to remediate before any public disclosure.

Handling the tool's outputs safely

  • A PatchWing evidence bundle may embed a reproducer that triggers the very bug it documents. Treat reproducer inputs as untrusted: run them only inside the sandbox the bundle names (network: none), never against production.
  • Every "green" verdict is decided by the container, not by the model that wrote the patch. A patch verified green proves the frozen reproducer is closed and the suite stays green — not that the patch is globally correct. Read the advisory-review section of each bundle, and have a human review before merging anything.

There aren't any published security advisories