Skip to content

Security: YunzhouLi-hub/GInstall-OneClick

Security

SECURITY.md

Security policy — GInstall agent integrations

This repository contains Markdown skills, slash-command docs, and JSON plugin manifests only. There is no runnable installer code here; execution happens when users install a separate GInstall CLI and agents invoke ginstall per these instructions.

Supported versions

Security-sensitive fixes belong primarily in the CLI repository. This integrations repo should stay in sync when behaviour affecting secrecy or trust boundaries changes (for example how URLs are validated or how tokens are handled).

Threat model (what we worry about)

Risk Notes
Leaked secrets Patches must never embed GitHub PATs, OAuth tokens, or private URLs with embedded credentials. Use GITHUB-UPLOAD.md hygiene checks before push.
Untrusted repository URLs Treat user-supplied URLs as untrusted. Malicious repos can ship hostile package.json scripts or native payloads. Prefer manual / plan-first flows for unknown sources; confirm behaviour of install --ignore-scripts (or equivalent) in your CLI version.
Token exposure via agents GITHUB_TOKEN in the environment may be inherited by child processes and echoed in logs. Do not paste tokens into chat; use host-supported secret stores or short-lived CI credentials where possible.
Prompt-driven misuse Skills encourage running shell commands. Host misconfiguration or overly broad agent permissions could amplify mistakes. Restrict agent filesystem/network access per vendor guidance.

Responsible disclosure

If you find a security issue in content shipped in this repo (for example instructions that systematically leak secrets), please open a private security advisory with the maintainer or email the contact listed on the publisher’s GitHub profile.

Issues in the GInstall CLI itself should be reported to that repository’s maintainers.


简体中文摘要

本仓库仅为说明文档与清单文件,不包含可执行安装代码。风险主要来自:① 切勿将 PAT / Token 写入文档或聊天记录;② 对用户提供的仓库 URL 视为不可信,未知来源优先只做规划或人工审查后再执行安装脚本;③ Token 会通过环境变量传给子进程,注意日志与 CI 泄露;④ Agent 自动执行 Shell 时需遵循宿主权限最小化原则。CLI 自身的漏洞请向 CLI 上游仓库 报告。

There aren't any published security advisories