Skip to content

Security: RayrunOS/Rayrun-Bot

Security

SECURITY.md

Security Policy

Rayrun Bot is pre-alpha. No release is published yet, so no released version is currently supported with security fixes. The first release is planned as v0.1.x; the table below will be updated when it ships.

Version Supported
v0.1.x planned, not released yet
main best effort, no guarantees

Reporting a vulnerability

Please report vulnerabilities privately through GitHub Security Advisories on the repository's Security tab. Do not open a public issue for a vulnerability, and never include live tokens, API keys, or credentials in a report.

This is a volunteer-maintained pre-alpha project. We will acknowledge reports as soon as we can, but we do not promise a response time, a fix SLA, or an audit.

What is in scope

  • Rayrun credential handling — how the bot obtains and uses Rayrun cloud computer credentials, and whether they can leak to another bot, a log, or a third party.
  • DeepSeek API key handling — the user's DeepSeek key stays with the user; anything that exposes it outside the user's own Harness or cloud computer.
  • Plugin RPC — the DSH web plugin talks to the local Harness over same-origin RPC; bypasses, CSRF, or cross-origin exposure matter.
  • Approval flow — actions that a user must approve should not be executable without that approval.
  • Cloud computer isolation — the Rayrun cloud image currently runs as root without bubblewrap, so the worker profile runs with danger-full-access. This is a deliberate workaround for that environment, not a sandbox guarantee: treat a compromised bot as having full access to its own cloud computer.

Handling practices

  • This repository ships no telemetry. Once the computer package moves in, its only outbound request is a version check against the npm registry; anything beyond that must be opt-in and documented before it ships.
  • The Rayrun token lives in the Harness vault, host side, and must never reach a client module, an RPC response, or a log.
  • The DeepSeek key is injected into the bot's agent process as a Rayrun credential; it must never be written into the workspace, the repository, fixtures, prompts, or chat.
  • Log redaction is still being verified — see the status note in README.md.

Out of scope: bugs in DeepSeek Harness itself, in Rayrun's platform or cloud images, or in third-party dependencies. Report those to the relevant upstream project.

There aren't any published security advisories