Introduction

The DSH plugin ecosystem allows overlaying different views within a Web session. Existing trajectory views are typically used to view the raw Trajectory, but when there are many events, locating records by action type or understanding the call chain from problem to solution still requires repeated expansion and filtering. dsh-client-ui-trajectory-categories is a pure client-side DSH Web plugin, which, based on preserving the original trajectory list, organizes trajectories into a Chinese-first, zero-extra-token categories, summary, and canvas view.

What is this

dsh-client-ui-trajectory-categories is maintained by QWQ-nn and is licensed under MIT. The plugin is positioned as a client-side plugin for DSH Web, used to convert raw trajectories into a Chinese-first, zero-token summary view. It reuses the trajectory view projection provided by @deepseek-ai/dsh-client-ui-trajectory, while depending on @deepseek-ai/dsh-client-runtime and @deepseek-ai/dsh-client-ui-conversation. The main body of the plugin targets the browser side, with lib/index.js being a host-side no-op.

Core Features

New “Categories” Tab

After installation, a “Categories” tab will appear in the session header. The original Trajectory list remains unchanged.

“Summary” Panel

The “Summary” panel is located in the top-left corner, recording statistics by action type: Write/Read/Delete/Download/Command/Search/Sub-agent/Other Tools/Message/Compression. Category tags are always in Chinese and support matching Chinese and English tool names/terminology. Records are displayed by category in the panel; clicking a category chip expands those records; clicking the chip again or clicking × clears/collapses them. Clicking a record allows viewing args, result, and error.

Mind Map Canvas

The canvas rebuilds the trajectory as “Problem -> Solution -> Tool Call”. It displays one problem at a time, with other problems switched via the “Problem Canvas” tab.

Within the canvas, clicking a problem node expands/collapses its sub-process, clicking a solution pill expands/collapses the tool call chain, and clicking a tool chip opens the detail panel. The detail panel shows what was done, output/logs, modifications, args, and errors. The canvas supports drag-to-pan, zooming via Ctrl + wheel or zoom buttons, and node dragging with inertia lines.

Zero Extra Token Consumption

This plugin only performs client-side aggregation, does not initiate model requests, and does not perform LLM summaries, therefore it generates no extra token consumption. It reuses the trajectory view projection from ui-trajectory and does not re-process the event stream. It defaults to lazy loading and on-demand rendering, initially loading 40 records per category.

Installation and Enablement

Use the installation command:

dsh plugin --profile web add dsh-client-ui-trajectory-categories

After installation, restart the dsh web service and refresh the page; a “Categories” tab will appear in the session header. If you wish to manually control the patch layer, you can install as follows.

  1. Place the package into the node_modules directory of the target profile, for example (subject to the actual profile path):
~/.dsh/profiles/node_modules/dsh-client-ui-trajectory-categories
  1. Append the plugin configuration to the cordis.patch.yml of the target profile:
- insert:
  - id: ui-trajectory-categories
    name: 'dsh-client-ui-trajectory-categories'
  1. Restart dsh web and refresh the page.

Typical Usage

First complete the installation and refresh the page, then click “Categories” in the session header.

  1. Click the category chip in the “Summary” panel to expand those records; clicking the chip again or clicking × clears/collapses them.
  2. Click a record to view args, result, and error.
  3. Switch to the “Problem Canvas” tab to view the canvas for different problems.
  4. Click a problem node to expand/collapse its sub-process.
  5. Click a solution pill to expand/collapse the tool call chain.
  6. Click a tool chip to open the detail panel and view what was done, output/logs, modifications, args, and errors.
  7. Drag the canvas to pan, use Ctrl + wheel or zoom buttons to scale, and drag nodes while observing inertia lines.

Applicable Scenarios and Notes

This plugin is suitable for use when DSH Web sessions are long and there is a need to quickly view action statistics, tool call details, and the relationship between problems and solutions. It does not alter the original Trajectory list and does not rely on model requests.

It is recommended to check the source code and license before installation. The plugin runs with the permissions of the current dsh process and is a browser-only/client-side plugin, with lib/index.js being a host-side no-op. The license is MIT, and the maintainer is QWQ-nn. Dependencies include:

  • @deepseek-ai/dsh-client-ui-trajectory
  • @deepseek-ai/dsh-client-runtime
  • @deepseek-ai/dsh-client-ui-conversation

Links

GitHub Repository:

https://github.com/QWQ-nn/dsh-client-ui-trajectory-categories

Directory Page: The verified information does not provide a specific directory page URL; do not interpret the DSH Community Directory as the official app store. If you need to view the directory page, search for the plugin name dsh-client-ui-trajectory-categories in the DSH Community Directory.