DeepSeek Harness 插件

dsh-client-ui-trajectory-categories

DSH Web 轨迹分类视图插件:「分类」标签页 = 左上角「合计」动作统计(写入/读取/删除/下载/命令/搜索/子代理/其他工具/消息/压缩)+ 思维导图画布(一次一个问题,问题画布标签切换,问题 → 方案 → 工具调用,节点可拖动、线条带惯性)。点击问题/方案收起展开子流程,点击工具查看「做了什么/输出日志/修改内容/参数/错误」。中文分类、中英文工具名匹配,保留原始「轨迹」流程列表。纯...(英文原文)

跳到安装方式

来源信息

GitHub 仓库
QWQ-nn/dsh-client-ui-trajectory-categories
最近更新
2026年8月19日
分类
界面增强
GitHub stars
3
载体类型
plugin
目录证据
上游声明已找到 dsh.bundle
证据路径
package.json#dsh.bundle
核对版本
0.1.0-rc.8
上游核对日期
2026-08-20

该证据由上游目录提供。本站没有安装、运行或安全审核这个插件。

安装

默认先复制一段 Prompt,让 Agent 读 GitHub 仓库和源码;需要自己装时再切到命令。

复制这段 Prompt,发给 DSH、Codex 或其他 Agent,让它先读 GitHub 仓库和源码。

请先不要安装或执行任何命令。阅读这个插件的 GitHub 仓库、README 和关键源码,然后用清楚、直接的方式回答以下问题,帮助我判断它是否适合我的需求:

1. 这个插件是什么,解决什么问题;
2. 适合哪些用户和典型使用场景;
3. 安装后如何使用,并给出一个最小使用示例;
4. 有哪些已知限制,以及隐私、安全、兼容性或维护风险;
5. 给出“推荐 / 有条件推荐 / 不推荐”的明确建议和理由。

请区分仓库明确说明、根据源码推断和未知信息。证据不足时请明确说明,不要猜测或照抄 README。

GitHub:https://github.com/QWQ-nn/dsh-client-ui-trajectory-categories
插件名:dsh-client-ui-trajectory-categories
作者:QWQ-nn

检查来源文件

安装前先看这个插件目录里的 README 和其他文件。

文件资源管理器3 个文件
README.md来源说明 · 只读预览

dsh-client-ui-trajectory-categories

A pure client-side DSH Web plugin that turns the raw trajectory into a Chinese-first, zero-token summary view.

Adds a 「分类」 (Categories) tab to every conversation session:

  • Top-left 「合计」 (Summary) panel — action-type statistics (写入/读取/删除/下载/命令/搜索/子代理/其他工具/消息/压缩) with one-click record expansion.
  • Mind-map canvas — one problem per canvas: 问题 (problem) → 方案 (attempt) → 工具调用 (tool calls), rebuilt from the trajectory. Switch between problems with the 「问题画布」 tabs.
  • The original Trajectory list stays untouched.

!Overview !License

Features

Chinese-first categorization

  • Category labels are always Chinese, regardless of UI language.
  • Classification matches both Chinese and English tool names/terms — 写入/写文件/编辑/修改, 删除/移除, 读取/查看/查找, 下载/抓取, 命令/执行/运行, 搜索/查询, 子代理/委派 … all map to the right category.
  • Arg previews prioritize the meaningful field (file path / URL / command / query), so rows read naturally in Chinese.

「合计」 summary panel (top-left)

  • Heatmap + category ranking (from community PR #1): a GitHub-style calendar heatmap of how many actions ran each day (hover a day for per-category breakdown), plus a leaderboard of the most-used categories. Click a day to filter the summary chips/records below; click again (or ×) to clear.
  • One chip per category with a count badge; click a chip to expand that category's records inline.
  • Each record shows time, name and arg preview; click to expand full 参数 (args) / 结果 (result) / 错误 (error).
  • Real-time filter by tool name/content; 全部展开 / 全部收起 buttons; “显示更多” loads more records lazily (default 40 per category).

Mind-map canvas (one problem per canvas)

  • Rebuilds the flow 问题 → 方案 1 →(报错)→ 方案 2 → … → 完成 automatically from user/assistant/tool events.
  • One problem per canvas — other problems live in the 「问题画布」 tab bar above the canvas; each tab shows a status dot (red = had errors, green = delivered).
  • Attempt pills show reply summary, 报错 badge with the error text, and status badges: 工作中 (running) / 完成 (delivered).
  • Click:

- click a problem → expand/collapse all of its sub-flows; - click an attempt pill → slide its tool-call chain in/out (collapsed pills show a ×N count); - click a tool chip → detail panel: 做了什么 (what the AI did, from args), 输出 / 日志 (result), 修改内容 (old → new diff, only when present), 参数, 错误.

  • Interactive canvas: drag the background to pan, Ctrl+wheel or the zoom buttons to zoom; hover highlights the block; the cursor gently repels nearby blocks.
  • Draggable nodes with inertia lines: drag a problem node and all its attempts/tools follow; connectors are elastic curves anchored to node edges that glide after the nodes (inertia).
  • Horizontal axis ≈ time — newer tool calls sit further right.

Zero extra token consumption

  • Pure client-side aggregation; no model requests, no LLM summarization.
  • Reuses the trajectory view projection built by ui-trajectory — no re-processing of the event stream.
  • Defaults to collapsed / render-on-demand; args and results render truncated previews until clicked; records load lazily (40 per category first).

Dependencies

  • @deepseek-ai/dsh-client-ui-trajectory — provides the trajectory view projection (data source)
  • @deepseek-ai/dsh-client-runtime, @deepseek-ai/dsh-client-ui-conversation

Installation

One-liner (official bundle install):

dsh plugin --profile web add dsh-client-ui-trajectory-categories
# or install straight from source (no build step, lib/ ships with the package):
dsh plugin --profile web add github:QWQ-nn/dsh-client-ui-trajectory-categories

Restart the dsh web service and refresh the page — a 「分类」 tab appears in the conversation header.

Manual install (when you want to control the patch layer yourself):

1. Put this package in the target profile's node_modules (e.g. ~/.dsh/profiles/node_modules/dsh-client-ui-trajectory-categories). 2. Append to the profile's cordis.patch.yml:

``yaml - insert: - id: ui-trajectory-categories name: 'dsh-client-ui-trajectory-categories' ``

3. Restart dsh web and refresh the page.

Files

  • lib/index.js — host-side no-op (browser-only plugin).
  • lib/client.js — the browser bundle (window.__ModuleLoader__.load format, hand-built, React-only, Chinese UI copy).
  • test-smoke.mjs — smoke tests: bundle load, apply(), server-render with data / empty / expanded detail / Chinese tool names.
  • test-interact.mjs — jsdom interaction tests: click-to-detail, drag-pan, ctrl+wheel zoom, hover, repulsion, node drag (parent moves children), connector alignment, problem switcher, tap-to-toggle, tap-guard after drag.

Verification

npm install        # installs react/react-dom/jsdom devDependencies
npm test           # SSR smoke tests
npm run test:interact   # jsdom interaction tests

Contributors

  • FlapPearLabs
  • (https://github.com/FlapPearLabs) — contributed the heatmap + category ranking feature (PR #1) and the eventLocations fallback hardening (PR #2).

License

MIT © QWQ-nn