zephyrpersonal / dsh-plugin-development-skill

dsh-plugin-development-skill

English | 简体中文

A reusable skill that turns a user idea into a DeepSeek Harness plugin.

This repository is the standalone source for the dsh-plugin-development skill. The skill teaches an agent how to:

• decide whether an idea belongs in a plugin,
• map the idea to the correct Harness extension point,
• choose between a scratch plugin, workspace package, installable bundle, or dynamic runtime plugin,
• implement a Cordis plugin with correct lifecycle, effects, configuration, and tool contracts,
• compose and run it,
• verify and finish with the repository's required checks.

Dependencies

The skill itself has no runtime dependencies. The @deepseek-ai/* imports in the blueprints are provided by the DeepSeek Harness runtime and repository workspace; do not install them separately when developing inside a deepseek-harness checkout or an installed dsh runtime.

Installation

npx skills

Install directly from this GitHub repository with the skills CLI:

npx skills add zephyrpersonal/dsh-plugin-development-skill

To install only the dsh-plugin-development skill without prompting:

npx skills add zephyrpersonal/dsh-plugin-development-skill --skill dsh-plugin-development --yes

Claude Code / Codex

Clone or copy this repository into your project's skills directory:

# Clone into a project
git clone https://github.com/zephyrpersonal/dsh-plugin-development-skill.git .claude/skills/dsh-plugin-development

# Or copy the skill files directly
cp -R dsh-plugin-development-skill /path/to/your/project/.claude/skills/

The skill is also compatible with .agents/skills/ because DeepSeek Harness and many agent runtimes discover skills from that directory.

Repository layout

.
├── SKILL.md                         # Main skill instructions
├── agents/
│   └── openai.yaml                  # Cross-product invocation metadata
├── references/
│   └── plugin-blueprints.md         # Copyable plugin templates
├── tests/
│   └── test_skill.py                # Standard-library test suite
├── README.md
└── README.zh-CN.md

Usage

Once installed, ask the agent to:

• "Turn this idea into a DeepSeek Harness plugin"
• "Add a tool plugin to deepseek-harness"
• "Create a service plugin with configuration"
• "Package this plugin as a bundle"

The skill will route the request through the decision workflow in SKILL.md.

Development

Run the test suite with the Python standard library:

python -m unittest discover -s tests -v

The tests validate:

• required skill metadata and sections,
• existence of references and README,
• balanced code fences,
• all relative Markdown links and anchors resolve,
• Claude/Codex invocation-policy alignment.

License

MIT