AI Agent Hub
Back to plugins
⚙️

dsh-preset-run

Workflow Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install YiGeSama/dsh-preset-run

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

Run dsh plugin install YiGeSama/dsh-preset-run to add the plugin to the current dsh profile (GitHub repo: https://github.com/YiGeSama/dsh-preset-run), then restart the web process so preset_run becomes available in all sessions.

About this plugin

In DeepSeek Harness, the Web frontend can open a new session, pick an agent preset, and send a message — but that pipeline has never been exposed as a programmatic entry point. If you want to let another preset-driven agent run a sub-task inside your current conversation and hand back the result, you used to have to open a manual window or wire up low-level APIs yourself. dsh-preset-run closes that gap: it registers a preset_run(preset, task) host-plane tool so a single call completes the full cycle of creating an isolated session, mounting the chosen preset, executing the task, and returning the final text — all through the official agentPresets.mount path, identical to the Web backend's composeAgent implementation.

Each sub-session is ephemeral: once it finishes, the agent is disposed, the session is removed from the registry, and no UI artifacts remain. The preset list is not hard-coded; whatever is present in the deployed agentPresets roster is available out of the box — the built-in standard, code, minimal, and cordis presets work immediately, and third-party routing presets appear automatically once installed. Model routing, timeout handling, and error cleanup are all built in, so callers only need to specify which preset to use and what task to run.

This plugin is ideal for agent workflows that require cross-preset collaboration: for example, probing with a router-spec preset that narrows tools on the first turn, then executing a full task under the standard preset; or debugging the cordis preset without interrupting your current session. Multi-step orchestration, parallel sub-task dispatch, and agent-to-agent relay are all natural fits.

Use Cases

  • Invoke a different preset agent to run a sub-task and collect the final text within the current session
  • Dispatch isolated one-shot sub-sessions with different presets across a multi-step workflow
  • Test specific preset behavior without interrupting the active working session

Best For

  • Developers orchestrating multi-agent workflows in DeepSeek Harness
  • Agent workflow builders needing cross-preset collaboration or batch sub-task dispatch
  • Automation engineers who prefer programmatic session management over manual window switching