dsh-team-task

Multi-agent workflows for long-horizon tasks often fail due to model errors, member interruptions, or coordinator crashes. Traditional team plugins optimized for short bursts cannot sustain goals that span hours, leading to stalled progress, broken coordination, and contaminated results. The team-task plugin addresses this by ensuring persistence and continuity through robust design.

At its core, team-task treats the event log as the sole source of truth, making state fully replayable and verifiable. It provides runtime settlement for member progress without relying on tool calls, fence tokens to manage dispatch rights and reject stale writes, a resident reconciler to resume tasks on timer, and review gates that enforce quality with feedback-driven rework. Progressive protocols load on demand to minimize prompt overhead.

This plugin is ideal for developers and teams executing long, complex projects such as research missions, continuous integration pipelines, or simulation experiments. By having a lead plan a reviewed DAG and durable members execute nodes, the system handles crash recovery and message delivery automatically, allowing focus to shift from infrastructure management to core business logic.