The steering report
How often people correct their agents, where, and why, from the session logs already on your machines.
Casebox
Stop steering your coding agents. Casebox turns every correction your team makes into evidence, every revert into a test case, and every harness change into a measured experiment on your own history.
This site documents Casebox 0.1.0: how to install it, use it on your team’s repositories, and operate it.
Cases from your history
Merged pull requests, reverts and corrected sessions become replayable tasks with tests that decide them.
Measured comparisons
A model, agent or harness change runs on your cases and returns better, worse, equivalent or inconclusive, with an interval.
Proposals as pull requests
Recurring corrections become small harness edits, tested on held-out cases before a person merges them.