agent-loop-workflow: A ready-made collaboration protocol skeleton for DSH multi-agent teams

# Introduction When building multi-agent systems in DeepSeek Harness (DSH), the common practice is to hardcode workflow rules into each agent's instructions: one for the executor, one for the reviewer, and one for the lead. As the rules become numerous, maintaining consistency becomes difficult, requiring changes to be synchronized across multiple locations. A more pressing issue is uncontrolled loops—agents repeatedly retry the same error, there is no mechanism to determine when to stop, and handoff information is lost when passed between roles. The agent-loop-workflow plugin introduced below serves as a workflow skeleton for these issues: it extracts the collaboration protocol so that any agent team only needs to load it once, rather than copying it everywhere.