Skip to content

Security: ShriD5/godspeed-code

Security

SECURITY.md

Security policy

Reporting a vulnerability

Please report vulnerabilities privately. Don't open a public issue.

  • Use Report a vulnerability under this repository's Security tab (GitHub Security Advisories). This opens a private advisory that only the maintainer can see.
  • If that option isn't available to you, contact the maintainer, @ShriD5, privately through GitHub.

Include the Godspeed version (Godspeed menu → About, or godspeed --version), your OS version, the steps to reproduce, and what an attacker could do. You should get a reply within a week. Godspeed is maintained by one person, so fixes ship in the next release rather than on a fixed schedule.

Supported versions

Only the latest release gets fixes.

How Godspeed runs

Godspeed is a local app. By default it makes no account and syncs nothing:

  • The engine listens for its UI and the godspeed CLI on a loopback WebSocket port (127.0.0.1:27654 by default) and rejects handshakes that carry a browser Origin header.
  • Chats, settings and agent memory live in ~/.godspeed (or GODSPEED_DATA_DIR).
  • Agents run as local processes under your user account, with the permissions the chat's access mode allows. Full access lets an agent run any command your user can.
  • Multi-device sync is off unless you deploy edge/ yourself and point Godspeed at it.

In scope

  • Anything that lets a web page, another user or a remote party reach the engine's IPC port or the zoomie (always-on agent) webhook endpoint (POST /hooks/agents/{id}/{token}) without the token.
  • An agent action that runs without approval when the chat's access mode should have asked.
  • Credentials or tokens leaking into logs, transcripts, error messages or exported data.
  • Path traversal or code execution through chat import, file attachments, previews or markdown rendering.
  • The edge/ sync service, when deployed as documented.

Out of scope

  • What an agent does under full access, or actions you approved.
  • Vulnerabilities in the agent CLIs themselves (Claude Code, Codex, Cursor and others). Report those to their vendors.
  • Gatekeeper warnings caused by the app not being notarized yet.
  • Attacks that need an attacker who already runs code as your user.

There aren't any published security advisories