Introduction

If you run agent tasks with DSH (DeepSeek Harness) on a daily basis, you might have encountered issues where token consumption is scattered across various sessions. You might want to know which model is used most, what the cache hit rate is, and which tools are called most frequently. Currently, you have to manually check session records to perform statistics.

Harness Insights solves this problem. It adds a local-first Usage insights page under Harness settings, aggregates structured metadata of local sessions and displays it, without modifying upstream Harness source code. Below introduces its positioning, functions, installation method, and data boundaries.

What is it

Harness Insights (mapan0424/deepseek-harness-insights) is an independent community plugin maintained by mapan0424, not released by DeepSeek. One-sentence positioning: Unofficial local-first usage insights plugin for DeepSeek Harness.

Several basic facts:

  • Release package name: @anarkhgatsby/deepseek-harness-insights, current version 0.1.4
  • License: MIT (explicitly declared in package.json as license: MIT)
  • Platform: web-only, targeting Harness projection and client APIs for version 0.1.1-rc.2

Core Features

After the plugin is installed, a Usage insights page is added under Harness settings, displaying the following:

  • token activity
  • model usage
  • active sessions
  • cache hit rate
  • activity breakdown
  • most used tools

The interface supports dual themes: Light features a classic Emerald/Forest Green gradient; Dark features a Codex-style deep black canvas with neon pink/magenta saturation curves (#ff3388). The core of the page is the Golden Ratio Heatmap Matrix, using a 2.7px grid spacing and enlarged cells.

Regarding icons, the plugin includes 26 pairs of Codex-style tool SVG icons in both light and dark versions, embedded into the client bundle during build. It switches via Harness’s body[data-ds-dark-theme] contract, and the plugin itself does not maintain separate theme preferences.

How Data is Processed

As a usage plugin, the data boundaries are worth explaining separately. Harness Insights only aggregates structured session metadata:

  • assistant/message.data.usage
  • assistant/message.data.message.source.provider/model
  • tool/call.data.name
  • event timestamps and session projection identity

Historical aggregation is performed via Harness’s official sessionProjectionCache.coldSnapshot() path. The plugin does not request session history in the browser, does not store message content, does not read API keys, does not upload usage data, and only persists the projection checkpoint owned by Harness.

Installation and Activation

The plugin follows the official Harness bundle manifest and is installed into a profile via dsh plugin. The following command installs the plugin to the demo profile:

dsh plugin --profile demo add @anarkhgatsby/deepseek-harness-insights

Subsequently, start DSH normally:

dsh --profile demo

After startup, enter Harness settings to see the Usage insights page. If you want to remove the plugin later, execute:

dsh plugin --profile demo remove @anarkhgatsby/deepseek-harness-insights

Local Development

If you are developing on a checkout of this repository, the same bundle can be installed directly from a local path:

dsh plugin --profile demo add ./packages/harness-insights

The plugin preserves dsh.client metadata, and the browser entry point will load along with the required Harness client services.

Packaging and Loading

Two implementation details worth knowing:

  1. The plugin does not contain architecture-specific native modules; its build artifacts are shared among macOS arm64, macOS x86_64, and Windows x86_64.

  2. Harness Desktop deploys its pure JavaScript runtime copy to the standard plugin root directory under $DSH_HOME/node_modules and loads it via a --patch overlay. As a web-only plugin, it injects client services such as dsh-client-runtime, dsh-client-connection, dsh-client-locale, and dsh-client-ui-settings.

Use Cases and Notes

Suitable scenarios: developers who often run sessions in DSH and want to intuitively understand token and model usage locally, but are unwilling to modify Harness source code for this.

A few points to note before installation:

  • This is an independent, unofficial community plugin. The plugin runs with the permissions of the current dsh process. It is recommended to check the source code and license (MIT) before installation.
  • The plugin targets only the web platform and Harness projection and client APIs for version 0.1.1-rc.2. After upgrading the Harness version, pay attention to interface compatibility.
  • The community plugin directory is an independent site with no official affiliation to DeepSeek or Firmode. Do not treat it as an official app store.

Conclusion

The value of Harness Insights lies in: adding a local-first usage view to Harness without modifying upstream source code, with clear data boundaries. If you use DSH and care about where tokens go, it is worth a try.

  • Directory: https://www.skillhub.cn/plugins/mapan0424/deepseek-harness-insights
  • GitHub: https://github.com/mapan0424/deepseek-harness-insights