Security fixes are applied to the current main branch. This repository does not currently maintain separate supported release lines.

Do not open a public issue containing credentials, session material, private prompts, terminal output, or a working exploit. Use GitHub's private Security advisories → Report a vulnerability flow for this repository. If private reporting is unavailable, open a minimal issue asking the maintainer for a private contact channel without including sensitive details.

Include the affected revision, deployment shape, reproduction prerequisites, impact, and any proposed mitigation. Remove real tokens, cookies, tmux paths, hostnames, prompt contents, and repository data from the report.

Remote Agent Console is a single-trusted-operator tool. A controlled browser can send terminal input and execute configured commands with the privileges of the Unix account running the server.

• Keep the application bound to loopback.
• Expose it only through HTTPS with the configured canonical Host and Origin.
• Do not publish port 8787 directly to an untrusted network.
• Use a long unique password, an independently generated session secret, and owner-readable configuration and persistence files.
• Treat Cloudflare Tunnel, reverse-proxy, GitHub CLI, Codex, and host tmux credentials as production secrets.
• Run the console under a dedicated, least-privileged account when practical.
• Review every configured worktree command and bind mount as trusted code.

The console is not designed for untrusted users, shared hosting, arbitrary shell access, or isolation between multiple operators.