DeepSeek Harness Plugin: human-writing-skills

Introduction¶

DeepSeek Harness (DSH) adopts a plugin-based architecture and emphasizes introducing specific quality guidance and review tools during the generation process. In long-form writing tasks, models can easily suffer from style drift, lack of character consistency, or factual errors. The human-writing-skills plugin aims to address these issues by using a set of specific rules and audit tools to transform general generation behavior into output that conforms to a particular style and maintains coherence.

Plugin Overview¶

This plugin is an open-source, advanced multilingual AI humanization toolkit focused on natural prose, voice preservation, long-form continuity, and focused audit. It is not simply a collection of prompts, but includes two layers of capability:

• Executable Python tools: Provide deterministic checks and correction previews.
• Model-executed editing modules: Provide genre and review instructions through Markdown files.

This plugin does not require the installation of additional Python core packages and features zero dependencies.

Installation and Prerequisites¶

Installing this plugin requires the DeepSeek Harness Web profile.

1. Installation command:
   Run the following command to add the plugin to the DSH configuration:

    dsh plugin --profile web add https://github.com/whh110112/human-writing-skills/releases/download/dsh-v0.2.0/dsh-advanced-human-writing.tgz

2. Prerequisites:
   • Python 3.9 or later is required.
   • Node.js is required, with a version matching ^22.19.0 || >=24.0.0.

Core Features¶

The plugin provides the following specific capabilities:

1. Writing and editing: Supports natural rewriting, AI text cleanup, fiction editing, and novel continuation.
2. Style and consistency audit: Provides chunked long-form auditing, writing style unification, and character consistency review.
3. Fact and coherence management: Includes automatic extraction of a continuity ledger, used to organize facts, inferences, and conflicts in existing chapters.
4. Quality control: Builds quality protocols into generation to prevent repetitive rhetoric, ornamental details, and unnecessary explanations.

Typical Usage¶

The plugin provides human-writing-skills command-line tools and MCP service support.

Command-Line Operations¶

1. Fact extraction: Extracts continuity information from existing chapters and generates a prompt for human review.

    human-writing-skills extract-ledger --draft chapters-01-10.md --context novel-ledger.md --output ledger-extraction-prompt.md

2. Style check: Lints a draft and outputs comments in GitHub Actions format.

    human-writing-skills lint --draft - --style general --format github --source-name docs-change.md

3. Build generation prompts: Builds a prompt based on a specific style and task.

    human-writing-skills build --style self-media --task "Explain the supplied test results for buyers." --source test-notes.md

4. Blind review comparison: Compares the original draft with the revised draft and generates a blind review directory.

    human-writing-skills compare-blind --original original.md --candidate revised.md --style self-media --brief writing-brief.md --output-dir blind-review

MCP Service¶

For long-form projects, you can use the human-writing-mcp service for local coordination, supporting orchestration across agents such as Codex, Claude Code, OpenCode, DSH, Manus, and Hermes.

human-writing-mcp --root C:\writing-project

Configuration and Notes¶

1. Project configuration: Create a .humanwriting.json file in the project root directory to persist default settings, such as style, document type, context paths, and the audit allowlist. This configuration file cannot silently activate deep configuration or expensive reference checks.

    {
      "style": "fiction",
      "document_type": "fiction",
      "context": "novel-ledger.md",
      "allow": ["END001"]
    }

2. Runtime permissions: The DSH plugin runs with the permissions of the current DSH process and has no separate system-level process management.
3. Quality limitations: Actual writing quality still requires human evaluation by an independent fluency reviewer; unit tests cannot prove the naturalness of the text.
4. Applicable scenarios: The plugin explicitly supports fact-based self-media articles and argumentative essays, and preserves the original voice during rewriting.

Summary¶

By combining Python tools and Markdown modules, the human-writing-skills plugin provides DeepSeek Harness with a complete workflow from style unification to fact auditing. It does not promise to trick the model through algorithms, but instead improves the coherence and readability of generated text through explicit instructions and constraints.

• Repository: https://github.com/whh110112/human-writing-skills
• License: MIT