Introduction

When developing agents in DeepSeek Harness, common requirements go beyond just getting Chat working; they also involve node orchestration, model routing comparison, and evaluation result management within the same session. DeepSeek Harness Flow is a community plugin that adds a new Harness Flow view to the existing Harness session while preserving the original Chat, sessions, settings, tools, and existing workflow records.

Below, we introduce its positioning, core capabilities, installation method, and boundaries to be aware of.

What is it

The repository name for DeepSeek Harness Flow is alison-xx/deepseek-harness-flow. It is positioned as a community visualization workflow and multi-model evaluation plugin for DeepSeek Harness. It is maintained by alison-xx and has an MIT license; the built-in benchmark prompts are original project content and are released under the same license.

The current beta version is pinned to DeepSeek Harness commit 47f9438 and matching RC packages. This is a community project. It is not affiliated with or endorsed by DeepSeek.

Core Capabilities

Adding a Harness Flow view within a Harness session

The plugin does not replace the original Harness experience. After installation and enabling, you can select Harness Flow next to the original Chat view. The original Chat, sessions, settings, tools, and existing workflow records remain unchanged.

Studio: Editing and running workflow nodes

Studio is used to edit and run the following nodes:

  • Input
  • Agent
  • Map Agent
  • Condition
  • Merge
  • Output

Bench: Running preset or imported evaluation suites

Bench is used to run preset or imported suites and to evaluate multiple provider routes and independent judges. Supported model routes include:

  • DeepSeek
  • OpenAI
  • Anthropic
  • Google
  • Custom OpenAI-compatible routes

Built-in evaluation suites include:

  • Quick Smoke
  • Comprehensive Agent
  • Coding
  • Safety

The default behavior is to repeat 3 times, with a limit of 2 global jobs and 1 job per provider route.

Runs: Viewing results and exporting

Runs is used to view persisted results, informal target results, formal rankings, and redacted exports. The exporter will redact credential-like fields, absolute workspace paths, and sensitive environment-variable names.

Public API

The plugin exposes flow and benchmark related Typert Remote namespaces as a public API.

Installation and enabling

The installation command provided is:

dsh plugin --profile web add https://github.com/alison-xx/deepseek-harness-flow/releases/download/v0.1.0-beta.2/deepseek-harness-flow-0.1.0-beta.2.tgz

After installation, restart the web profile, open any session, and select Harness Flow next to the original Chat view.

Desktop launcher

The repository also contains a Windows-oriented Tauri desktop launcher. It launches an isolated local Harness profile with a default port of 4181. The Desktop beta uses a local DeepSeek Harness checkout and Node.js runtime, rather than packaged credentials or a second Harness instance.

To start the desktop launcher, you can use:

pnpm install --frozen-lockfile
pnpm desktop:dev

Evaluation and ranking rules

The formal 100-point quality score uses the following weights:

  • objective 50%
  • judge 30%
  • safety 10%
  • stability 10%

Formal ranking requires non-competitive independent judges. When there is no independent judge, results remain objective-only, and no formal rank is assigned.

Development and verification

Development requires Node.js 22+ and pnpm. Common verification commands:

pnpm install --frozen-lockfile
pnpm verify
pnpm test:browser
pnpm pack:release

If you only want to run the standalone visual test shell, you can use:

pnpm preview

The address of this shell is http://127.0.0.1:4178.

Privacy and security

The plugin has no telemetry. It does not upload test prompts, model answers, reports, or API keys.

Regarding key handling, plaintext keys are only sent to credentials.set; the plugin does not persist keys. Import commands remain disabled until explicitly authorized during run.

Applicable scenarios and notes

Suitable for the following scenarios:

  • Wanting to edit and run Agent workflow nodes directly within a DeepSeek Harness session.
  • Needing to compare evaluation results across multiple provider routes.
  • Needing to view persisted results, formal rankings, and redacted exports.
  • Using DSH for community plugin development or local verification.

Notes:

  • This plugin is in beta and pinned to DeepSeek Harness commit 47f9438 and matching RC packages.
  • This is a community project. It is not affiliated with or endorsed by DeepSeek.
  • The plugin runs with the permissions of the current dsh process; you should check the source code and license before installing.
  • When importing third-party evaluation suites, pay attention to the boundary where import commands remain disabled until explicitly authorized during run.

Links

GitHub:

https://github.com/alison-xx/deepseek-harness-flow

Directory page (based on plugin entry):

https://www.skillhub.cn/plugins/alison-xx/deepseek-harness-flow