Show HN: A benchmark that diffs the whole database after an AI agent's task
Benchmark for computer-use agents on self-hosted business apps, graded by whole-database diffs: success, collateral damage, irreversible changes, false completion. - shubhamjoshipromail-svg/softwar...