I tested whether my agent actually reused tools. A plain router did better in the first comparison.

The developer of CasaJev, a local prototype where Jev selects bounded actions and Codex builds Python tools, tested whether their agent reused tools. A plain router performed better in the initial comparison. The tools are stored in a registry for future tasks. The developer is seeking 2–3 individuals with recurring CSV/JSON jobs and a method to validate the output, like a weekly export with changing headers, to understand workflow-breaking variations. An anonymized example would be more beneficial than another synthetic test.

This report uniquely provides the full validation snapshot and code for CasaJev, unlike other discussions that only describe the project.