Skip to content

Security: tastyeffectco/sandboxd

SECURITY.md

Security Policy

sandboxd runs untrusted code by design: each app executes inside its own Docker sandbox, and the control plane orchestrates those containers. Because of that, we take security reports seriously and ask you to disclose them privately.

Reporting a vulnerability

Do not open a public issue for security problems.

Report privately through GitHub's private vulnerability reporting: Security → Report a vulnerability. It notifies the maintainers directly and keeps the report private until a fix ships. If you can't access that form, open a minimal public issue that only says "security report — please enable a private channel" (no details) and we'll follow up.

Please include:

  • affected component — sandboxd core (control plane / runtimed / Traefik config) or the console client;
  • a description, impact, and reproduction steps or a proof of concept;
  • affected version / commit and your deployment configuration (relevant .env).

We aim to acknowledge within 72 hours, agree on a disclosure timeline, and credit you in the release notes unless you prefer to remain anonymous.

Scope & trust model

Some behaviors are intentional and are not vulnerabilities on their own — know the trust model before reporting:

  • The control-plane container mounts the host Docker socket to orchestrate sibling sandboxes. It is effectively host-root; run it on a host you control.
  • Agents run inside the throwaway sandbox container with --dangerously-skip-permissions; the containment boundary is the container, not the agent.
  • SANDBOXD_API_AUTH_DISABLED defaults to true for local use and the API is published on loopback only. Exposing it without token auth is a misconfiguration, not a bug. See docs/production-safety.md.
  • Public preview routes are intentionally framable (the edge strips X-Frame-Options so the console can embed them). Treat previews as public.

In scope: sandbox escape, cross-sandbox access, credential/secret disclosure, auth bypass, control-plane RCE, SSRF against the host/metadata, and privilege escalation beyond the documented trust model.

Supported versions

Security fixes land on the latest release line. Until 1.0, only the most recent minor (0.x) is supported — please upgrade before reporting.

There aren't any published security advisories