Open-source tool Tura cuts coding-agent turns using macro execution layer

A developer has open-sourced Tura, a macro execution layer designed to reduce the number of model turns coding agents require during routine workflows. Instead of handling each step—such as editing files, running tests, and linting—as a separate turn, Tura bundles them into a single macro executed by a runtime. On a 60-task DeepSWE benchmark, the Macro plus backward reasoning configuration achieved an 80% pass rate while using fewer rounds than comparable Codex CLI settings. The developer cautions that fewer turns do not automatically translate to lower overall costs, as retries, failures, and cache behavior also affect total expenditure. The implementation, benchmark data, and methodology are publicly available on GitHub.