Skip to content

Security: JiaHuiRed/RedCode

Security

SECURITY.md

Security

RedCode is a single-maintainer personal fork of opencode, developed and tested only on Windows 10/11. What follows describes how the tool actually behaves — it is not a support commitment.

Threat Model

Overview

RedCode is an AI-powered coding assistant that runs locally on your machine. It provides an agent system with access to powerful tools including shell execution, file operations, and web access.

No Sandbox

RedCode does not sandbox the agent. The permission system exists as a UX feature to help users stay aware of what actions the agent is taking - it prompts for confirmation before executing commands, writing files, etc. However, it is not designed to provide security isolation.

If you need true isolation, run RedCode inside a Docker container or VM - see containerization.md for concrete recipes and their trade-offs (in short: both protect your host, neither protects API keys you pass into the sandbox from a compromised session).

Trust Model

  • Local users are considered trusted
  • Agent is considered untrusted but operating under user supervision
  • Remote APIs (LLM providers) may observe code context sent in requests
  • Third-party plugins may have access to the same capabilities as the agent

Tool Permissions

Tool Requires Approval Notes
Read
Write ⚠️ Configurable
Edit ⚠️ Configurable
Grep/Glob
Shell
Web ⚠️ Configurable
MCP Tools ⚠️ Configurable Depends on MCP server

Reporting a Vulnerability

Open a GitHub issue at JiaHuiRed/RedCode. There is one maintainer and no triage automation, so response time varies.

If the issue affects upstream opencode rather than this fork specifically, report it there instead — that is where it can actually be fixed for everyone.

There aren't any published security advisories