dsh-plugin-skillsDeepSeek Harness plugin

dsh-plugin-skills

English | 中文

Agent skills for building and testing DeepSeek Harness plugins — from scaffolding a new plugin package to choosing the right test tiers, entirely inside an agent session.

What's inside

Skill	What it does
dsh-write-plugin	Scaffolds a plugin end to end: pick the right shape (tool / LLM adapter / hook / service / config), walk the package checklist (package.json invariants, tsconfig registration, README + Model Experience, verify gates), with a self-contained reference file per shape.
dsh-test-plugin	Picks the right test tiers for a plugin change: unit, per-file coverage gate, real-API e2e, keyless snapshots, web browser snapshots — and when each is required, including real-entry-path and built-bin smoke coverage.

Both skills are fully self-contained: no external docs or other skills needed at runtime — everything required is written inside the skill.

Install

Copy the skill folders into your project's agent skills directory:

sh
cp -r dsh-write-plugin dsh-test-plugin <your-project>/.agents/skills/

Claude Code projects usually symlink .claude/skills to .agents/skills; if yours does not:

sh
ln -s ../.agents/skills <your-project>/.claude/skills

That's it. The agent picks the skills up automatically (the skill catalog hot-refreshes on disk changes) — just ask it to write a plugin or test this plugin change, or invoke a skill by name.

Requirements

Runs in any mainstream agent product — the skills are plain SKILL.md Markdown directories with no product-specific glue, so they work wherever an agent can read skill instructions: Codex, Claude Code, DeepSeek Harness, Cursor, Windsurf, Gemini CLI, GitHub Copilot, Cline, Roo Code, OpenHands, Aider, Devin, and more.