I built a more restrained alternative to giant AI skill bundles
I’ve just shipped agent-harness v1.0.0.
This project exists because I kept running into the same problem with large AI skills/prompts/context bundles: they were often impressive, but too broad for real day-to-day product work.
I don’t always want “everything.”
I want the right agent assets for the right repo on the right host without blowing up context windows.
So I built agent-harness as a Node/TypeScript CLI that tries to do that more selectively across VS Code / GitHub Copilot, Cursor, OpenCode, Zed, Claude Code, and Pi.
This is not a polished grand-unified AI platform pitch.
It’s a builder tool, mostly vibe-coded at the start, then forced through real release testing and issue cleanup.
The main thing I want now is feedback:
- Is this a real problem?
- Is the lifecycle model useful?
- Is the host support too broad or not broad enough?
- Is the product too complex for what it gives back?
- What should be cut?
If you try it, please open issues — especially negative ones.
Repo: https://github.com/ar27111994/agent-harness
Release: https://github.com/ar27111994/agent-harness/releases/tag/v1.0.0
npm: https://www.npmjs.com/package/@ar27111994/agent-harness
