DeepSeek Super-Agent: what it is and how to set it up

What DeepSeek Super-Agent is

The repository describes dsh-superagent as “a proactive coding agent for one GitHub repository, packaged as a bundle for DeepSeek Harness.” When someone on an allowed list asks it for work, it opens a work item (a Ticket), works on it in its own agent session and directory, and answers in the original conversation, for example with a pull request. A web panel lists only what needs a person.

DeepSeek labels it a research prototype, “not an official DeepSeek product.” The public code differs from DeepSeek's internal version, and some designs from DeepSeek's write-up are future directions that the prototype doesn't implement.

• Sources: GitHub notifications, subscribed issues and pull requests, CI changes and merged or closed pull requests; Feishu chats if you add them.
• Parallel work: every Ticket gets its own executor session named worker-<ticket id> and its own directory, usually a git worktree checkout, up to a concurrency limit you set.
• Follow-ups: later comments on an issue or pull request reach the Ticket working on it, without mentioning the agent again.
• Its rule for itself: “take as little human attention as possible, and give as much value as possible.” It confirms a request once, stays quiet while it works, and writes again only with a result, a question or a failure.

Is it really proactive? The three accept modes

News coverage describes an agent that fixes bugs before you ask. The documentation is more careful: a direct request from an allowed person always works, but work nobody asked for needs a rule your team wrote, and the default mode only records what the agent would have done.

How to set up dsh-superagent

You need Node.js 22.19+ or 24+, pnpm, git 2.32+, tmux, Python 3.8+, GitHub CLI 2.48+, DeepSeek Harness 0.2.1-alpha.2 or a later 0.2 release, and a model provider configured in DeepSeek Harness. The setup is long, so the repository has your coding agent walk you through it.

1. Install DeepSeek Harness

   DeepSeek Harness (dsh) is DeepSeek's open-source agent harness, in developer preview. Start it once and configure a model provider; the Web UI opens at http://127.0.0.1:3080.

   npx @deepseek-ai/dsh web

2. Clone the repository and hand SETUP.md to your coding agent

   In the checkout, ask your coding agent to read SETUP.md. It probes your machine, asks one question per message, shows the whole plan once, and only then installs anything.

   git clone https://github.com/deepseek-ai/dsh-superagent-prototype.git
   cd dsh-superagent-prototype
   # then ask your coding agent:
   # Read SETUP.md in this repository and set up dsh-superagent with me.

3. Answer the setup questions

   Where requests come from (GitHub only, or Feishu too), which repository, which GitHub account the agent acts as (a dedicated one, never your own), which logins may ask it for work, which clone each Ticket checks out from, who its commits name as author, and whether it may enable auto-merge (the recommended answer is no).

4. Do the steps only you can do

   Create the agent's GitHub account, accept the repository invitation as that account and sign its CLI in. The agent never asks for a secret in the chat; it gives you the exact commands to run in your own terminal.

5. Check the defaults it applies

   A separate DeepSeek Harness profile named superagent on port 3082, branches prefixed agent/, accept mode observe, and an hourly check that restarts the agent if it stopped.

6. Send it the first request

   From an allowed account, comment @<agent login> <request> on an issue. A Ticket appears in the panel within about a minute; the agent acknowledges, works in its own directory and replies in the same thread, usually with a pull request.

Source: the dsh-superagent repository on GitHub

Safety: what to know before you run it

• Workers run with danger-full-access as your OS user. They can read your home directory and run any program by path; the account wrappers pick credentials but aren't a security boundary.
• For isolation, DeepSeek's docs say to run it under a dedicated OS account that holds only the agent's credentials.
• Everything it writes appears under the agent's GitHub account, so the setup recommends a dedicated account and a short allowed list to start.
• Messages and comments can't change who may ask for work: allowed lists, rules and accept mode change only in the panel.
• After each turn, background work has at most 30 minutes to finish before its shell jobs are stopped and the Ticket is marked waiting.
• It's a research prototype, and DeepSeek Harness is a developer preview with breaking changes expected.

DeepSeek Super-Agent vs ChatGPT dots vs Crewmark

All three are agents that keep working without you in the chat, but they start work differently. Super-Agent waits for a mention or a rule your team wrote; ChatGPT dots live in OpenAI's cloud; Crewmark wakes the coding agent you already use, DeepSeek Harness included, on a schedule.

Make your DeepSeek Harness, Claude Code or Codex session proactive

Super-Agent needs its own profile, accounts and setup for each repository. Crewmark works with the agent you already run, DeepSeek Harness included: it gives that session a role, a goal and working hours, and wakes it on a heartbeat to take the next issue from a board, for any kind of work, not only one repository.

1. Turn a session into an employee

   Pick a role such as software engineer or SEO specialist, write its goal as a number, and run the setup command in DeepSeek Harness, Claude Code, Codex or OpenCode.

2. Let the heartbeat start the work

   During its working hours the employee wakes every 15, 30 or 60 minutes in the same conversation and picks up where it left off; see how to make Claude Code proactive.

3. Assign issues and answer its messages

   Put work on the board; the employee takes it, comments as it goes and messages you when it needs a decision. Read what an AI employee is.

DeepSeek Super-Agent FAQ