Skip to content

Security: AH64-dll/OpenEdit

Security

SECURITY.md

Security Policy

Open Edit is an experimental, local-first tool. It runs as a local MCP server on your own machine and never phones home — no cloud accounts, no telemetry, no uploads. Security-sensitive surface: the MCP stdio server, open_edit serve (HTTP review UI), the run_script sandbox (bwrap on Linux / dev on Windows), and the render subprocesses.

Reporting a vulnerability

Please report privately via GitHub private vulnerability reporting:

  1. Go to https://github.com/AH64-dll/OpenEdit
  2. Security → Report a vulnerability (or open a draft advisory)

Do not open a public issue for security findings. Include:

  • The affected component and version (release tag or commit)
  • A minimal reproduction (commands, project files if small)
  • Impact and any suggested fix

What to expect

  • Acknowledgment within 3 business days.
  • A fix or a clear mitigation plan within 14 days for confirmed issues.
  • Credit in the release notes for the fix (if you want it).

Scope

In scope: the Python package (open_edit/), installers (install.sh, install.ps1), and the GitHub Pages guide. Out of scope: dependencies reported through their own channels, and the optional whisper models (faster-whisper handles model loading).

Safe defaults

  • MCP runs over local stdio only; bind serve to 127.0.0.1 unless you know why not (OPEN_EDIT_SERVE_HOST).
  • serve enforces a bearer token (OPEN_EDIT_TOKEN) for non-localhost clients.
  • run_script uses bwrap sandboxing on Linux when available.
  • No secrets are stored by the project; credentials live only in your own host configuration (e.g. ~/.config/gh/hosts.yml).

There aren't any published security advisories