Welcome to agent-harness Discussions!
I just released agent-harness v2.x.
This project came from a pretty specific frustration: I wanted a better and more restrained alternative to giant all-in-one AI skill/prompt/context bundles.
A lot of those bundles are impressive, but for my own varied product work, they often felt too broad, too sticky, and too expensive in context-window terms.
So agent-harness is my attempt at a different direction:
discover useful agent assets
stage and activate them in a more structured way
wire them into different hosts
keep the system selective instead of maximal
It currently supports VS Code / GitHub Copilot, Cursor, OpenCode, Zed, Claude Code, Codex and Pi.
What I want most right now is feedback, especially from people who think this approach is wrong or incomplete.
Please tell me:
What feels overbuilt
What feels missing
What host support is not worth it
What abstractions are awkward
What should be removed
Where this breaks in real use
If you test it, please open issues with concrete repros or even just criticism/product doubts.
Links:
Replies