The DeepSeek Harness (DSH) ecosystem contains a large number of plugins, but there is currently no quality assessment tool based on npm metadata that can directly provide “a specific score” and “fix suggestions.” dsh-quality-score addresses this need by providing a standardized quality scoring system for DSH plugins.

Plugin Overview

  • Name: dsh-quality-score
  • Maintainer: zoahdev
  • Category: Model inference
  • License: MIT

This is a quality scorecard for DeepSeek Harness (dsh) plugins. Based on the same methodology as ecosystem supply chain health scanning, it generates a 0-100 composite score, grade, and per-component breakdown for each plugin, and lists specific issues with fix suggestions.

Installation

Install the plugin using the official command:

dsh plugin add dsh-quality-score

The plugin has zero runtime dependencies, defaults to read-only operations, and does not require an API Key.

Core Features

  1. Scoring and grading: Outputs a 0-100 composite score and automatically calculates the grade (A-F).
  2. Component breakdown: Decomposes the score into multiple concrete components, showing each component’s score and maximum possible score.
  3. Issues and suggestions: Identifies potential issues in the plugin configuration and provides fix suggestions.
  4. Batch leaderboard: Supports batch scanning and generates a plugin quality ranking list.
  5. Agent invocation: Can be integrated as a tool into DSH Agent and invoked through the quality_score tool.
  6. Technical characteristics: No runtime dependencies and graceful degradation in offline environments.

Usage

CLI Command Line

Score a single plugin:

dsh-quality-score <npm-name>

Batch-score multiple plugins in a list and output a leaderboard (one package name per line in the file; lines starting with # are treated as comments):

dsh-quality-score --batch <names-file>

Agent Invocation

In a DSH Agent session, you can directly invoke the quality_score tool. The Agent will return a structured report containing schema, score, grade, components, issues, and suggestions.

Scoring Components

The score is composed of the following six components, with a total of 100 points:

Component Max Score Evaluation Criteria
manifest 20 Whether the latest version includes name, version, description, and license
peer-resolvable 20 Whether each peer dependency range has a published matching version
dist-tag 15 Health of the latest tag (for example, #2763-type issues)
dead-ranges 15 Whether peer dependency ranges reference package names that do not exist
freshness 10 Whether the latest version was published within 365 days
dsh-tools-peer 10 Whether the declared dsh-tools peer compatibility satisfies the current latest

Grade ranges:
- A ≥ 95
- B ≥ 85
- C ≥ 70
- D ≥ 55
- F < 55