Skip to content

Security: Blaise1030/tabsh

Security

SECURITY.md

Security

tabsh runs shells, so a hole in the daemon is a hole on your machine. Please report suspected vulnerabilities privately, not in a public issue.

Reporting

Use GitHub's private vulnerability reporting:

https://github.com/Blaise1030/tabsh/security/advisories/new

Include the version, your OS, the steps, and what an attacker gains. A proof of concept is welcome; a description is enough to start. I aim to acknowledge within 72 hours and to ship a fix or a mitigation as soon as it is understood.

Please do not open a public issue, post the details on social media, or test against a daemon you don't own. If you need a channel other than GitHub, say so in the advisory and we'll find one.

What is in scope

  • Authentication or authorization bypass in the daemon.
  • Reading or writing files outside the intended folders.
  • Command execution beyond what the paired user asked for.
  • Leaking the pairing token (logs, URLs sent to a server, referrers).
  • Cross-site requests that reach the daemon from another origin.
  • XSS or HTML injection in the app that can reach the daemon.
  • Dependency vulnerabilities that are actually reachable.

What is not in scope

  • Someone who already has a shell as your user, or can read ~/.tabsh/token.
  • A daemon deliberately exposed on 0.0.0.0 or a LAN address. It is designed for loopback; use Tailscale or an SSH tunnel if you need another device.
  • An agent or MCP server you chose to run. It runs with your permissions by design.
  • Denial of service from your own shells or a local process.

The model, in one paragraph

The daemon binds 127.0.0.1 by default, creates a random token in ~/.tabsh/token, and accepts API and WebSocket requests only with that token. It also checks the Origin/Host of browser requests so another site cannot drive it. The token is placed in the URL fragment, which browsers do not send to servers, and the hosted app at tabsh.cc only talks to the daemon over loopback. There is no account, no relay, and no telemetry.

If you find a gap between that paragraph and the code, that's a vulnerability worth reporting.

There aren't any published security advisories