dsh-context-explorer：在 Web UI 里看清 Session 发给模型的上下文
dsh-context-explorer 是 LionNatsu 维护的 DSH 客户端插件，在 Web UI 右侧「上下文 / Context」标签用 token 方格网格与分区列表展示系统 prompt、工具 schema 与消息的占比与顺序。安装命令为 dsh plugin --profile <name> add dsh-context-explorer。面板数字来自提供商最后一次请求计数或 Harness 同款启发式估算；正文为 serving 实现重建而非抓包。MIT 许可，非 DeepSeek 官方插件。
Read Moredsh-context-length：自定义模型上下文窗口与思考强度
DeepSeek Harness 插件 `dsh-context-length` 旨在解决现有配置难以精细控制上下文窗口和推理强度的问题。该插件在设置面板新增“上下文长度”页面，支持为指定渠道的模型自定义 32K 至 1M 等窗口大小及 7 档思考强度。通过渠道过滤、预设快捷填入及实时预览功能，用户可便捷管理。配置变更需点击保存才写入生效，并提供恢复默认及聚合管理功能。安装需通过 Git 仓库，并注意桌面版与 Web 版需分别安装。
Read MoreDSH Too Long Warning
`dsh-toolong-warning` is a plugin for DeepSeek Harness (DSH) designed to address pain points in context window management. It tracks conversation compaction events and issues a warning only when compactions are frequent and involve significant additional token consumption, preventing unnecessary interruptions to users. The plugin provides a draggable, bilingual floating window and supports customizable trigger thresholds, helping developers identify cost-inefficient conversations while balancing context management with cost. Installation requires DSH 0.1.7-rc.1 and Node.js 22.19 or later.
Read Moredsh-ballast: DSH Web Context Window Attribution Plugin
`dsh-ballast` is a client plugin for DeepSeek Harness (DSH) Web, designed to address the challenges of context window management. It leverages the host's token meter to display, on a per-item basis, the current token usage of a surface, along with body summaries and pricing information. Core features include attributing token usage by message type, aggregated statistics, window pressure estimation, session list viewing, and cross-session top ranking. The plugin retrieves data through read-only interfaces, supports truncating body content after collapsing whitespace, and includes environment dependencies and applicable use cases, helping developers monitor context usage in real time and troubleshoot window overflow issues.
Read Moredsh-model-context-catalog: Maintaining the Context Window for llm-pi-ai
The dsh-model-context-catalog plugin, maintained by HOWILLMAKEIT, addresses false-positive overflow issues in DeepSeek Harness (DSH) for llm-pi-ai models caused by an excessively small default context window of 262,144 tokens. By maintaining accurate route-level contextWindow values and correcting metadata, it prevents tool-calling failures. It provides a web management interface that allows users to configure the actual context windows for different provider/model routes. After installation, users need to add and save the model in settings, ensuring that the plugin runs the model according to its true capacity.
Read MoreContext-Vista: Visualizing Context Token Usage and Cost in DSH
context-vista is a client plugin developed by GooodWei for DeepSeek Harness, offering the /context slash command and a floating ring chart on the right side. It displays real-time token usage for system, tools, messages, and remaining space, while estimating session costs. The floating card supports dragging and collapsing, with a button at the bottom to trigger /compact for context compression. Built-in DeepSeek pricing allows customization of peak and off-peak unit prices per route in ~/.dsh/settings.yaml. Install with: dsh plugin --profile web add github:GooodWei/context-vista, then start dsh web. Licensed under MIT.
Read More