dsh-swarm
Run the following command in DeepSeek Harness:
dsh plugin install wanghj040530/dsh-swarm
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install wanghj040530/dsh-swarm in your terminal to finish installation; source repository: https://github.com/wanghj040530/dsh-swarm
About this plugin
When your task is complex enough to need role-based division — research, coding, analysis, review — a single main session shuttling everything around burns tokens fast, and a crashed agent loses its task with it. dsh-swarm turns each agent into an independent, persistent process, each with its own model and provider config. One cold start, then it loops for life — no more disposable subagents.
Two hard design decisions keep collaboration efficient. Artifact reference passing writes large outputs to a shared store and ships only an art:// reference plus a 200-character summary in messages; full text is fetched on demand, slashing token spend by an order of magnitude. The task ledger, paired with a watchdog, pins state to pending, claimed, done, or failed — automatic retries (up to three) kick in on failure, and a claimed task that stalls beyond five minutes is re-dispatched, so a crashed agent never loses a task.
Built for developers who need genuine multi-role collaboration: long pipelines, parallel research and synthesis, mixed-model teams. The DSH interface ships a built-in team panel for visual agent management and ledger tracking, plus native web search — form the team and get to work.
Use Cases
- Long multi-role tasks: parallel research, coding, analysis, and review
- Token-heavy multi-turn agent collaboration needing reduced relay overhead
- Automatic task retry and re-dispatch after an agent crash, ensuring delivery
Best For
- Developers who need parallel agents with later synthesis
- Tech users building long pipelines or mixed-model agent teams
- Operations-minded users who care about crash tolerance and zero task loss
Related Plugins
A method pack that makes AI coding agents plan against your real baseline, prove completion with fresh evidence, and reduce reworks and unsafe changes.
Turns the DeepSeek Harness session into a captain that builds a durable sub-agent team, splits goals into dependency-aware tasks, and coordinates work via direct messages and a live Web UI.
Gives coding agents design judgment, letting Claude Code, Cursor, and 70+ agents generate and iterate high-quality UI, presentations, and graphics right from the terminal.
Run the Pi ecosystem's plugins on DeepSeek Harness, unmodified, via a compatibility layer that implements Pi's public extension ABI on DSH's native services.