There's something ironic about AI coding agents.
We trust them to write code, refactor entire projects, debug complex issues, and automate development workflows.
Yet we still manually maintain their environment. 😅
Skills get outdated. MCP servers get deprecated. Plugins change. Configurations quietly become obsolete.
And unlike broken code, these problems often don't throw errors. Things just slowly become outdated.
So I built Keepup, a free, open-source Agent Skill that lets coding agents audit and maintain their own setup.
How does it work?
It's a simple five-step process:
- Inventory — Discover installed skills, MCP servers, plugins, hooks, and configurations.
- Check — Verify everything against current sources.
- Report — Present findings with evidence.
- Apply — Fix only what the user explicitly approves.
- Verify — Confirm the changes actually worked.
No background daemon. No new management platform. Just an Agent Skill following the open Agent Skills specification.
It works with Claude Code, Codex CLI, Gemini CLI, Cursor, GitHub Copilot CLI, and other coding agents.
Give it a try
npx skills add lezli01/skills --skill keepup -g
The project is MIT licensed and available on GitHub:
https://github.com/lezli01/skills
I also wrote about why I built it, the implementation decisions, and its limitations:
📖 https://blog.lezli01.is-a.dev/blog/keepup-agent-setup-audit-skill/
The funny part? Keepup can even discover that Keepup itself needs updating. 😄
How do you maintain your coding agent environment? Do you actually audit it, or just wait until something breaks?
Top comments (0)