AI Agent Hub
Back to plugins
⚙️

dsh-autoresearch

Workflow Updated 2026.09.13

Run the following command in DeepSeek Harness:

dsh plugin install aa2246740/dsh-autoresearch

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install aa2246740/dsh-autoresearch in the DeepSeek Harness terminal; make sure the official dsh and pnpm are available on your PATH before installing from the repository at https://github.com/aa2246740/dsh-autoresearch, then restart the Host and refresh the browser to activate the plugin.

About this plugin

If you want DeepSeek Harness Web to run an autonomous modify-measure-keep-or-rollback loop without hovering over every iteration, dsh-autoresearch plugs that research pipeline straight into the official web session interface with a single /autoresearch command.

Once you set a goal and a round budget, the agent iterates: it edits project files, executes local measurement scripts, and decides whether to keep or revert each change. Nothing is uploaded or pushed. A Git baseline or a private snapshot guards every round, and protected files are restored automatically on crash, discard, or failed checks. A collapsible panel at the top of the session reports progress in real time, and the loop pauses to ask you whenever a trade-off calls for human judgement. The model saying done is not enough; the ledger must record a complete entry before the loop truly stops.

Built for developers running DeepSeek Harness 0.1.5-rc.3 on Node.js 22.19+ who want the agent to experiment freely but safely inside a local sandbox. The core loop is ported from grok-autoresearch (Copyright Tobi Lutke, David Cortes); the DSH host integration and web panel are original work in this repository, released under the MIT license.

Use Cases

  • Set a performance or quality goal and let the agent edit code, run measurement scripts, and keep or revert each change automatically
  • Run agent iterations unattended; on crash or failed checks, protected files are restored via Git baseline or private snapshot
  • Agent pauses at critical trade-offs to ask for human judgement before continuing to the next iteration

Best For

  • Developers on DeepSeek Harness Web who want the agent to experiment autonomously inside a local sandbox
  • Engineers who need a structured modify-measure-keep-or-rollback loop without monitoring every round manually
  • Project leads who require zero upload or push, keeping all experimentation strictly on the local machine