Introduction

DeepSeek Harness (DSH) is built on the philosophy that “everything is a plugin.” In a general conversational session, handling study plans, new vocabulary, algorithm derivations, reviews, and archival all at once requires a clear entry point. dsh-study-steward provides this entry point, turning an ordinary session into a personal learning and management system with real execution capability.

What Is This

This is a DSH agent preset developed by adsikito. It is a configuration preset, and its core substance is composed of cordis.patch.yml (declaring the persona and tool set) and a common Markdown rule pack under the rules/ directory.

Core Features

  1. Full Tool Coverage: The plugin list declared by the preset is exactly the same as the standard preset built into DSH (including shell, files, search, background jobs, skills, goals, plan mode, compaction, subagents, workflow engine, todo, web, and present). Switching to this preset only changes the persona; no functionality is lost.
  2. Study Steward Persona: It predefines discipline rules and maintains three dimensions: content, tasks, and mastery. Daily plans and actual records are stored separately. Recommendations move from silent suggestions to commitments. Receipts only record events that actually occurred.
  3. Common Rule Pack: Provides daily record conventions, behavior conventions, templates, and archival rules.
  4. Templates and Examples: Includes 26 content templates and 19 examples.

Installation and Enablement

Install it via the command line:

dsh plugin add github:adsikito/dsh-study-steward

After installation, start a new session and choose “Personal Learning and Execution Steward” in the preset selector. Existing sessions retain the preset version they were started with.

Configuring the Rule Pack

This preset’s persona points to the rules/ directory. To make the agent work as expected, copy the rule pack to your local notes directory (for example, an Obsidian vault):

cp -r "<installed package>/rules" ./study-rules

The persona preferentially uses the local copy of the rule pack, meaning the local copy takes priority over the built-in pack.

Notes

  1. Pure Configuration Preset: The repository contains only the preset declaration and the rule pack. It does not include any personal study materials (such as course documents or personal profiles), so users must provide them themselves.
  2. Fictional Content: The content under rules/参考示例/ is entirely fictional. Its demonstration tasks, time spent, completion dates, and scores must not be used as real tasks, progress, or historical records.
  3. Language: The persona is language-neutral, but the rule pack is in Chinese.