thinkflowlab/nullius ? reverse-engineered prompt

Reverse engineered prompt

Build me a local web app for running and reviewing AI agent evaluations on my own machine.

I want to drop in benchmarks, agents, or capabilities, run them against each other, and then open each result to see exactly why it scored the way it did. The app should make it easy to start a run from the command line, open a dashboard in the browser, and browse runs, trials, traces, model calls, tool use, outputs, and judge results. It should also show clear outcomes like passed, failed, unknown, or infrastructure issue, plus simple compare views so I can see what changed between two runs.

It would be great if it can work with local job folders, support multiple machines or shards, and keep everything local by default. If you need to look up current docs online, do that. Make the first version polished enough that I can run it, inspect evidence, and understand whether a problem was in the agent or the benchmark.

Are you gonna build this?

make sure you review the code using coderabbit

Try freeSponsored — opens CodeRabbit in a new tab