Conflicts aware scheduling of tasks for agents

Parallel coding agents can speed up development, but assigning them tasks independently creates avoidable problems: agents edit the same files, begin work before prerequisites are complete, or repeatedly load the same project context. These collisions waste compute and make coordination especially difficult when agents work asynchronously in the cloud. Our project combines a planning tool with a conflict-aware scheduler. Developers describe work in Markdown issues and identify the files each task may change. The planner helps refine those issues into actionable steps and can gather code evidence for review. The scheduler proposes which tasks can proceed together based on file overlap, dependencies, priority, available capacity, and work already in progress. It explains why a task was deferred, so developers can understand and adjust the plan. It can also group related tasks by cache affinity, creating opportunities to reuse context and reduce redundant agent setup. The tool is for developers and teams running multiple coding agents. They create and review issues, request a batch from the command line, assign the selected work to agents, and rerun the scheduler as task states change. By combining evidence-informed planning, explainable conflict avoidance, and cache-aware scheduling, it helps teams use parallel agents more effectively while keeping developers in control of what runs next.