Skip to content

Security: khasky/git-account-manager

SECURITY.md

Security policy

Reporting a vulnerability

Report privately through GitHub, not in a public issue:

Open a private security advisory

Do not open a public issue, pull request, or discussion for a security problem, and do not post it anywhere else until a fix has shipped.

Useful in a report:

  • what an attacker can do, and what they need first (local access, a malicious repository, a hostile network, another app on the machine);
  • where it lives — a file, or an app surface (OAuth flow, token storage, SSH key handling, ~/.ssh/config or .gitconfig writes, the pre-push guard, the updater);
  • the OS and version, the app version, and how it was installed (MSI, NSIS, .dmg, .AppImage, .deb, or built from source);
  • the smallest reproduction you have — a snippet, a screen recording, or numbered steps;
  • if it involves a repository or account you cannot share, an equivalent public one.

Never include real credentials, OAuth tokens, private keys, or another person's data. A redacted excerpt is enough.

What to expect

  • Acknowledgement within 7 working days.
  • A first assessment (accepted / not a vulnerability / needs more detail) within 14 working days.
  • Fixes ship in a normal release; a critical one gets an out-of-band release.
  • Public disclosure is coordinated with you once the fix ships, or 90 days after the report — whichever comes first.
  • Credit in the release notes and the advisory if you want it — say so in the report.

This is a solo, unpaid project: there is no bug bounty.

Scope

In scope: this repository — the desktop app and everything it writes or reads on your machine (the OS credential store entries, profiles.json, ~/.ssh/config, global and per-repository Git config, the installed pre-push hook), the OAuth flows it drives, and the signed-update path.

Out of scope: vulnerabilities in GitHub, GitLab, Bitbucket, Git itself, OpenSSH, Tauri, or the OS credential store — report those to their own programs. Also out of scope: findings that require physical access to an unlocked device, an attacker who already has code execution as your OS user, or social engineering of the maintainer.

Safe harbor

Research done in good faith under this policy is welcome. Stay within your own accounts, machines, and repositories, do not degrade anything for anyone else, and give a reasonable window before disclosing. Under those conditions there will be no legal action from this project.

Supported versions

Only the latest published release receives security fixes. It is listed on the releases page; the Installation section of the README lists every way to get it.

There aren't any published security advisories