Skip to content

Security: rayline-ai/rayline-cli

SECURITY.md

Security Policy

Supported Versions

Rayline CLI, the command-line client in the broader Rayline ecosystem, is under active pre-release development. Only the latest release and main are supported; security fixes are applied there.

Reporting a Vulnerability

Please do not open public issues for security vulnerabilities. Public reports can expose users before a fix is available.

Report privately using GitHub Private Vulnerability Reporting:

  1. Go to the Security tab of this repository.
  2. Click Report a vulnerability.
  3. Provide a clear description, affected version, and reproduction steps.

If you cannot use GitHub Private Vulnerability Reporting, email security@rayline.ai.

When reporting, please include:

  • the affected version and operating system;
  • a description of the issue and its impact;
  • step-by-step reproduction;
  • any relevant logs with secrets and tokens removed.

What to Expect

We aim to acknowledge reports promptly, confirm and assess the issue, and keep you informed as we work on a fix and coordinate disclosure.

Trust Model

This section documents what Rayline's local security mechanisms protect against, what they do not protect against, and what residual risks remain.

Threat boundary

Loopback is not a security boundary on a single-user machine. Any process running as the same OS user can connect to Rayline's local ports, read the router API key from a harness's process environment, or read key material from the filesystem (subject to file permissions). Rayline's local protections are defense-in-depth measures — they are not a hard isolation boundary.

For strong isolation (e.g., when running untrusted local code alongside Rayline), use OS-level isolation: a separate OS user account, a container, or a virtual machine.

Router API key

The router API key authenticates harness sessions to the Rayline router. It is:

  • Revocable server-side: keys are minted via POST /v1/keys and revoked via POST /v1/auth/cli/revoke (see crates/rayline-cli/src/status.rs, functions mint_router_key_at and revoke_rayline_session).
  • Withheld from harness environments: Claude Code, Codex and the other harness spawns remove RAYLINE_ROUTER_API_KEY from the inherited environment (crates/rayline-cli/src/codex.rs, withhold_router_key). Claude Code gets the key only as ANTHROPIC_AUTH_TOKEN; Codex reads it from a 0600 config.toml in its Rayline CODEX_HOME, never from its arguments or environment.
  • Residual risk: the key remains readable in Claude Code's process environment via /proc/<pid>/environ on Linux (or equivalently via ps eww <pid> / proc_pidinfo on macOS), and in Codex's config.toml, by any process running as the same UID. This is a known residual. Same-UID isolation requires OS-level controls.

Recommended isolation for untrusted workloads

If you run untrusted local code alongside Rayline (e.g., MCP servers from unknown sources, arbitrary shell commands), use OS-level isolation:

  • Run Rayline under a dedicated OS user account, or
  • Use a container or VM that separates Rayline from untrusted code.

Loopback and same-UID process boundaries are not sufficient isolation against a determined attacker with code execution on your machine.

There aren't any published security advisories