During my work at Brokk last year we developed a task-slicing system. The idea was to have a bunch of tasks where each task can be implemented and verified in isolation. So I can mentally focus only on one task at a time with checking the evidence it works as expected. Most follow-up tasks are based on their predecessor(s). After my time at Brokk I picked up this idea and developed Casefile. To avoid loading the entire previous session context, the agent creates a task log. So a follow-up task o...