Skip to content

Security: rune-oai/t3code-proxy

Security

SECURITY.md

Security policy

This page explains how to report a vulnerability, what a hub holds that an attacker would want, and how to expose a hub so that only your own devices can reach it.

Report a vulnerability

Report vulnerabilities privately through GitHub security advisories: report a vulnerability. Do not open a public issue, pull request or discussion for a security problem.

Include what you found, the version (t3code-proxy -h prints it on the first line), steps to reproduce, and the impact you expect. Remove real keys, tokens and account data from anything you send. You get a reply in the advisory thread. Once a fix is released, the advisory is published with credit to you unless you ask otherwise.

Supported versions

Security fixes go into the latest release. Upgrade to it before you report, if you can.

What a hub holds

A hub is a high-value target. Anyone who can read its files or reach its management API can use every subscription on it.

Item Where What it allows
Account logins the auth directory, ~/.t3code-proxy by default OAuth access and refresh tokens for every Claude and Codex account
Codex refresh journal .codex-refresh/ in the auth directory Refresh receipts, which contain credentials
Client access keys access.api-keys in config.yaml, plaintext Sending requests through the hub with your subscriptions
Management secret management.secret-key in config.yaml, bcrypt hash Full control: accounts, config, logs, the api-call proxy
Quota history quota-history.db next to the config Account labels and usage over time
Logs stdout, or logs/ when logging-to-file is on Request metadata; request bodies when request-log is on

The installers write config.yaml with owner-only permissions, and the service definitions set a 0077 umask on Linux and macOS. Keep these files out of backups that others can read and out of version control.

Recommended exposure

  • Bind to loopback: server.host: "127.0.0.1". In a container, bind to all interfaces inside the container and publish the port on the host's loopback only, as docker-compose.yml does.
  • Publish the port to your tailnet with tailscale serve. Do not use tailscale funnel, port forwarding on your router, or a public reverse proxy.
  • Set server.trusted-proxies to loopback only, so forwarded client addresses are accepted only from tailscale serve.
  • Use a long random management secret and long random access keys. The installers generate both.
  • Use Tailscale ACLs if devices you do not control share your tailnet.
  • Set management.secret-key to "" if you do not need the panel or T3 Code limits; that turns the management API off.

Scope

In scope: the t3code-proxy binary, the web panel, the installers, the service definitions and the container image published from this repository.

Out of scope: weaknesses in Claude Code, Codex, T3 Code, Tailscale or the providers' services, and setups that ignore the exposure guidance above, such as a hub bound to a public address.

There aren't any published security advisories