built `jev-review` TypeSafe AI it's an experimental, local-first MCP plugin that gives coding agents a score quality feedback loop across different metrics. agents call jev while they work, get scored, make improvements, and repeat the loop try below 👇

@typesafeai instead of jev being called by agents, it should be automatic feedback after every turn, on a confidence threshold level

@typesafeai classic had this idea but you shipped first lol

@typesafeai oh nicee

@typesafeai I can see a good use case: "check if unnecessary fallback bloat"

@typesafeai Did it improve performance?

@typesafeai The loop is the key detail: agents call the local MCP plugin, receive scores across metrics, improve, and repeat instead of relying on one-shot evaluation.

@typesafeai The feedback loop for coding agents is such a smart application for this kind of model. We actually went deeper on this here:

@typesafeai ok but how are u able to give massive code contexts to it? it has very small context window/limit. Or is it one file at a time?

@typesafeai Beat me to it, had the same idea! I'm deepseeking through the repo now, thanks 👏

@typesafeai The useful bit is the loop, not the score. If the agent can’t turn failures into a smaller next attempt, you’ve built a dashboard, not a feedback system.

@typesafeai nice!