Last Updated: 28 October 2026 · Toolkit 3.15.23

A plugin for Codex and Claude Code that gives AI coding agents a shared set of skills and working rules, so several agents can build, test and review real projects with a clear owner and a check at each step.

<aside> 🔗

GitHub Repository

https://github.com/weijunswj/ai-agent-toolkit

</aside>

💡 Project Overview

I started this in May 2026 to keep my agent rules, guides and helper scripts in one place. It has since grown into the process I use to run AI coding agents on live projects, including Swooshz and the automation work at my day job.

Today it does three things:

• Installs a set of skills and global rules into Codex and Claude Code as a native plugin, and keeps them up to date from the repo's main branch.
• Sets out how a change moves from idea to merge: discovery, build, then independent review.
• Adds safety checks for the risky parts of agent work: live systems, credentials, personal data and n8n.

⚠️ Problem

Running AI coding agents on real projects goes wrong in predictable ways:

• Agents start building before anyone understands the problem.
• The same agent writes and approves its own work.
• Rules get copied between repos and drift apart.
• An agent touches a live system, a credential or customer data without anyone checking first.