HughYau/qiushi-skill

What is Qiushi‑Skill?

Qiushi‑Skill is a library of “skills” (plug‑in modules) for AI agents. It packages nine classic methodological tools drawn from Marxist‑Leninist philosophy (contradiction analysis, practice‑cognition, investigation‑first, mass‑line, criticism & self‑criticism, protracted‑strategy, concentrate‑forces, spark‑prairie‑fire, overall‑planning) and provides them as ready‑to‑use, self‑contained skill definitions that can be injected into a conversation‑agent runtime (Claude Code, Cursor, Codex, etc.).

Each skill contains:

• A short trigger rule (when the skill should be used)
• An operation guide (step‑by‑step actions the agent should perform)
• An output template (structured result the agent returns)
• Hard constraints (discipline rules that must be obeyed)
• A handoff spec (what to do next)

The library also ships with reference texts (original quotations from the source works), sub‑agents (e.g., an “investigator” that produces a fact/guess/unknown report), and workflow definitions that let multiple skills be chained.

---

Why it exists

The author argues that many current AI agents can think but don’t think about thinking: they jump to answers, ignore evidence, and give up when a problem is hard. By giving agents a disciplined, evidence‑first methodology, Qiushi‑Skill aims to make them:

1. Fact‑driven – conclusions must follow verified evidence.
2. Self‑critical – regular checks and criticism loops.
3. Strategic – choose the right “weapon” (skill) for the task and focus resources.

---

How to get it

1️⃣ One‑line install (recommended)

npx qiushi-skill

The CLI detects the host agent (Claude Code, Cursor, etc.), writes the appropriate skill files into the host’s plugin directory, and registers the entry‑point skill (arming‑thought).

You can also target specific hosts or scopes, e.g.:

npx qiushi-skill install --target claude-code --scope user
npx qiushi-skill install --target cursor,codex,opencode,openclaw,hermes,nanobot --scope user

2️⃣ Claude Code Marketplace

Add the plugin via the built‑in marketplace commands:

/plugin marketplace add HughYau/qiushi-skill
/plugin install qiushi-skill@qiushi-skill

If you don’t have Node.js, simply copy the skills/ folder into the host’s skill directory.

---

Using the skills

After installation, the “武装思想” (Arming Thought) kernel is automatically loaded at the start of every session (≈ 50 lines of persistent code). It enforces the top‑level principle 实事求是 (seek truth from facts) and decides, based on the current context, whether to invoke any of the nine methodological skills.

Manual slash commands (if the host supports Markdown slash commands)

Command	Skill
/contradiction-analysis	矛盾分析法
/practice-cognition	实践认识论
/investigation-first	调查研究
/mass-line	群众路线
/criticism-self-criticism	批评与自我批评
/protracted-strategy	持久战略
/concentrate-forces	集中兵力
/spark-prairie-fire	星火燎原
/overall-planning	统筹兼顾
/workflows	工作流组合

Each command opens the corresponding SKILL.md (or runs the sub‑agent) and returns a structured result that the main agent can consume.

---

Project layout (high‑level)

qiushi-skill/
├─ .claude-plugin/   # Claude Code marketplace metadata
├─ .cursor-plugin/   # Cursor plugin metadata
├─ bin/qiushi-skill.mjs   # CLI entry point
├─ commands/          # Markdown slash‑command files
├─ hooks/             # Session‑start injection scripts (POSIX, PowerShell, CMD)
├─ agents/            # Sub‑agents (investigator, self‑critic)
├─ skills/            # One folder per methodological skill + entry kernel
│   ├─ arming-thought/   # entry kernel (实事求是 + routing)
│   ├─ contradiction-analysis/
│   ├─ practice-cognition/
│   └─ … (other 7 skills) …
│   └─ workflows/        # cross‑skill workflow definitions
├─ tests/             # validation scripts for all platforms
├─ docs/              # static site and platform‑specific install guide
└─ package.json

---

Who might use it?

• Developers building custom AI assistants (Claude Code, Cursor, Codex, etc.) who want a built‑in, philosophically‑grounded decision‑making layer.
• Researchers experimenting with agentic prompting who need reproducible, structured reasoning patterns.
• Teams that want to enforce evidence‑first, self‑critical workflows in any LLM‑driven automation.

---

License & Community

• MIT License – free to use, modify, and redistribute.
• Discussions, examples, and third‑party integrations are listed in the README (e.g., a “Zero‑basis AI” book, a “Ralph Qiushi” workflow, user‑submitted case studies).
• The author also maintains AcademicForge, a meta‑tool for installing multiple agent skills across platforms.

---

TL;DR

Qiushi‑Skill is a plug‑and‑play collection of nine disciplined reasoning “skills” for AI agents, delivered via a tiny Node‑based CLI or marketplace entry. It equips agents with a fact‑first, self‑critical workflow, making them better at tackling complex, multi‑step problems.