The idea
Niaz Morshed introduces jev-review, an experimental local-first MCP plugin that gives coding agents quality feedback across several criteria. The agent can use the scores in another revision.
Inside the decision loop
Work in progress + quality criteria
Score separate quality dimensions
Revise and check again
What to take from it
A scoring loop becomes more useful when each dimension maps to a change the developer can inspect. Set a revision budget and keep tests as a separate source of evidence.
The source calls the project experimental. A rising score does not by itself demonstrate improved correctness, and repeated optimization can exploit a poorly specified rubric.
Follow the original work
This case is an editorial interpretation of a public community demonstration. It has not been reproduced or benchmarked by this publication.
Original post by @niazmorshed_