xingtu1996/dsh-xingtu-skills ↗★ 0
dsh-xingtu-skills
星图 26 个生产级 AI Agent 技能插件，包含 Token 压缩、代码审查、安全重构、迁移等技能。
AI 分析
核心用途是为 DSH 注入 26 个开箱即用的专业级 Agent 技能。适合需要进行代码重构、审查、迁移等复杂开发任务的用户。一键安装后由 Agent 按需触发。
安装
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:xingtu1996/dsh-xingtu-skills说明文档
阅读完整 README ↗dsh-xingtu-skills · XingTu Skills Plugin for DeepSeek Harness (DSH)
26 production-grade AI agent skills, packaged as a standard DeepSeek Harness (DSH) plugin — one command to install.
What is this
This plugin packages the 26 production-grade SKILL.md skills from xingtu-skills into a standard DeepSeek Harness plugin bundle. After install, DSH's session skill catalog loads all skills automatically and triggers them on demand (progressive disclosure) — no manual copying.
Each skill follows the cross-tool de-facto standard (name + description + when_to_use) and is also compatible with Claude Code / CodeBuddy / Codex / Cursor / Gemini CLI.
Install
# From npm (recommended)
dsh plugin add dsh-xingtu-skills
# From the GitHub release tarball
dsh plugin add -w https://github.com/xingtu1996/dsh-xingtu-skills/releases/latest/download/dsh-xingtu-skills-0.1.0.tgz
Skills (26)
| Series | Skills |
|---|---|
| Caveman · token compression | caveman caveman-commit caveman-compress caveman-discover caveman-evidence-review caveman-explore caveman-help caveman-learn caveman-manage caveman-optimize caveman-review caveman-setup caveman-stats cavecrew |
| Ponytail · minimalism & debt | ponytail ponytail-audit ponytail-debt ponytail-gain ponytail-help ponytail-review |
| Engineering practice | investigate-first lean-build migration safe-refactor surgical-patch verify-and-stop |
Structure
dsh-xingtu-skills/
├── package.json # dsh.bundle.patch → ./cordis.patch.yml (standard plugin declaration)
├── cordis.patch.yml # inserts this plugin row (bundle layer)
├── lib/index.js # entry: apply() registers every SKILL.md under skills/
└── skills// # 26 skill directories (SKILL.md + their resources)
- Plugin contract: exports
name+apply(ctx)(inject: ['skills']); every contribution goes throughctx.effect()/ctx.skills.register()and is auto-disposed on unload. - Zero dependencies: no Cordis copy is bundled;
@deepseek-ai/dshis only an optional peerDependency. - Validated: passes
dsh-plugin-dev check(static checks) +dsh-plugin-dev verify(clean-DSH_HOME install / boot / uninstall).
Related
- xingtu-skills — source repo (cross-tool)
- xingtu-harness — one-command AI engineering scaffold
- xingtu-ai-engineering — flagship methodology repo
License
MIT License. Skills content derived from real production engineering practice; all company-sensitive information removed.