Preface

DeepSeek Harness (DSH) uses a plugin-based architecture. For developers dealing with ambiguous requirements that are difficult to convert into concrete, verifiable capabilities, the community contributed the dsh-strength-refine plugin. It provides a capability point decomposition engine called Highest (Peak), aimed at transforming requirements of any domain and any granularity into a hierarchical capability point tree, and attaching clear acceptance criteria to each leaf node.

Core Value

The plugin is essentially a capability point decomposition engine. Given a requirement, it collaborates with the user to clarify it and decomposes it into a hierarchical capability point tree. Each leaf node includes clear, actionable, and determinable acceptance criteria, along with sufficient argumentation proving that the set of capability points fully covers the original requirement.

Installation and Enablement

Install the plugin using the official command:

dsh plugin --profile web add github:ShadowBruceMeaningLau/dsh-strength-refine

After installation, restart dsh web. When creating a new session, select Highest in the preset selector to use this capability decomposition engine.

Working Principles

The plugin itself is a minimal host-side component. On startup, it idempotently writes the built-in preset directory (preset/strength-refine/) to the DSH user preset root directory ~/.dsh/.agent-presets/, and triggers the preset selector to discover it.

Regarding preset management, the plugin follows the following rules:
* Do not overwrite local modifications: If the target location already exists and the preset has no .dsh-plugin marker (for example, a version edited by the user), the plugin will not modify it; only copies installed by the plugin itself will be updated during version upgrades.
* Fully portable: The preset bundle (agent.cordis.yml) references the official @deepseek-ai/dsh-* packages, the skill root is resolved via baseUrl relative paths, and the entire preset can run on any machine without any network installation steps.

Typical Usage

In the DSH interface, create a new session and find and select Highest in the preset selector. If a preset already exists at the target location without the .dsh-plugin marker, the plugin will not overwrite it, keeping your local configuration unchanged.

Methodology

The plugin includes complete methodology documentation in SKILL.md of the bundled skill strength-refine. Core concepts include:
* Hierarchical gate system: requirement clarification → layer-by-layer decompose-self-check-delivery approval cycle → aggregated final check → archive.
* Coverage persuasiveness: mathematical proof-style argumentation across the dimensions of definition, necessity, coverage, and boundaries.
* Four elements of validation criteria: action + object + judgment criterion + evidence.
* Logical relationships: prerequisite dependency DAG among leaves, collaborative combination argument clusters, and alternative options.
* Mermaid visualization: supports native rendering of four diagrams in Obsidian.

Uninstallation

If you need to uninstall the plugin, run the following command:

dsh plugin --profile web remove dsh-strength-refine

After the plugin is uninstalled, the preset directory (~/.dsh/.agent-presets/strength-refine) remains because it is self-contained. If the preset directory is a version you edited, the plugin will not overwrite it; if you want to remove it as well, manually delete the directory.

  • GitHub: https://github.com/ShadowBruceMeaningLau/dsh-strength-refine
  • Directory: https://www.skillhub.cn/plugins/ShadowBruceMeaningLau/dsh-strength-refine