Tool of the Day: dsh-agent-teams for Coordinating Coding Agents

A large coding task can involve research, implementation, and review. dsh-agent-teams lets a lead agent in DeepSeek Harness divide the work among team members, track tasks, exchange messages, and bring the results together.

When it helps

Use a team when the task has parts that can be worked on independently. For a small change, the coordination overhead may outweigh the benefit.

A practical example

Ask DSH: /agent-teams Review the login system, check security and tests, then summarize the three most important issues.

The lead assigns the parts and collects the findings. Review the agents' claims and file changes yourself before accepting them.

Getting started

1. Check the project's page for supported DSH versions and installation instructions.
2. Install the specified plugin version and restart DSH.
3. Try it on a small test project first.

The project documents limits: one active team per lead agent, with team state kept in local files within a single DSH process. Two processes using the same state are not coordinated. Published code and popularity do not establish safety; inspect the source and permissions before installing.

Sources: https://dshmarket.com/p/NanmiCoder/dsh-agent-teams/ and https://github.com/NanmiCoder/dsh-agent-teams

Workflow: request → team lead → research / implementation / review → combined result.