Cursor kept telling me it was done, and I had no easy way to check. Sometimes it wasn’t. In one session, the agent even read logs from other conversations and never mentioned it.
So I built Osier. It records what the agent actually does and tells you when that doesn’t match what it said. It runs locally.
npm install -g osier-dev
osier setup
osier review
NPM: https://www.npmjs.com/package/osier-dev
It’s early, so honest feedback is welcome.