Reducing coding-agent round trips with macro execution: a 60-task benchmark

I maintain Tura, an open-source execution layer for coding agents. The idea came from a repetitive pattern in long coding tasks: the agent often spends a separate model turn on each operational step.