DSH Quality
Workflow & Automation
NanmiCoder/dsh-agent-teams

dsh-agent-teams: from solo to squad

Turn the current session into a captain, spawn a persistent team of sub-agents, split tasks with dependencies, and let a shared scheduler dispatch.

Config fragment

bash
dsh plugin --profile web add github:NanmiCoder/dsh-agent-teams

The plugin elevates the current session into captain mode; the Web UI renders each member's live state (working / idle / paused).

Core code

The orchestration loop worth copying: decompose -> dispatch -> aggregate.

ts
const team = await ctx.agent.spawnTeam({
  roles: ['researcher', 'writer', 'reviewer'],
  persistent: true,
});
const tasks = splitIntoDependentTasks(goal, team);
const results = await scheduler.dispatch(tasks, {
  assign: (t) => team.idleMember(), // by availability
  dependencies: true,
});
return captain.summarize(results);

A shared scheduler assigns work to idle members, honours task dependencies, and the captain aggregates the archive. It is the same pattern as parallel "research N competitors, write one page each" flows.

Why it scores

  • Pattern: persistent sub-agent teams with a shared scheduler (429 stars).
  • UX: live member states in the Web UI make orchestration visible.
  • Lesson: multi-agent is not chaos if dispatch is deterministic.