Rationale

AI Research Skills is built on a simple idea:

In an AI-assisted world, the smallest reusable unit of “software” is often a skill — a precise, teachable, composable way to do a task.

This project turns that idea into an open, community-curated library.

What We Mean by “Skill”

In ordinary life, a skill is a capability you gain through practice and experience. You transfer it by:

1. Demonstrating and explaining,
2. Letting someone practice,
3. Refining the process until it becomes reliable.

With AI, a large part of knowledge work becomes AI-assisted. That changes what “transfer” looks like:

• Instead of teaching only humans, we also teach AI assistants.
• The transfer medium becomes documentation (often Markdown).
• The unit becomes small and reusable: a skill.

In other words: we don’t just write prompts — we write portable practice.

Skills vs. Prompts vs. Agents

Prompts

Prompts are the raw material. They are usually:

• One-off,
• Hard to maintain,
• Hard to reuse across people and projects.

Agents

An agent is “a capable worker”: it can reason, plan, and act. But a single agent still needs specialized training.

Skills

Skills are the scalable layer:

• Specific: one task or workflow, not a whole job role.
• Composable: skills can link to and build on other skills.
• Discoverable: metadata makes them searchable and filterable.
• Progressively loaded: you don’t load everything every time; you load what you need when you need it.

Think of an agent as a person, and skills as the training manuals and playbooks that make that person excellent at a particular job.

Why This Isn’t Just a “Prompt Library”

Prompt libraries existed early, and many were role-based (“act as an editor…”, “act as a historian…”). Skills keep the best part — reuse — but change the emphasis:

• From role-play to procedures: how to do a task reliably, step-by-step.
• From monolith prompts to a graph: skills reference other skills and shared conventions.
• From copy/paste to workflows: skills can be validated, indexed, and integrated with tools.

Skills Don’t Replace Software — They Orchestrate It

A skill can look like “software” because it produces outcomes. But the execution is still grounded in real tools:

• OS utilities (image conversion, PDF processing, file transforms),
• Libraries (Python/R), CLIs (pandoc, imagemagick),
• Web APIs.

That’s why AI Research Skills includes an MCP server: an AI-native interface that exposes tool capabilities in a structured way when plain text instructions aren’t enough.

What This Project Provides

1. A curated skills library (skills/**/SKILL.md)
2. A fast docs site (browse + search + clean reading)
3. A validator CLI (quality gates for contributors)
4. An MCP server (AI-native “software surface”)
5. Optional semantic search (a backup to keyword + native docs search)

These parts can be used together or independently.

Where We’re Going (Roadmap)

AI Research Skills is a project in itself, not a one-off repo:

• Client integrations and plugins (Claude, CLI workflows, editor integrations)
• A stable contribution pipeline for low-skill contributors (web-first, with optional Git/CLI)
• “Skills as benchmarks”: each skill can become an evaluation task with:
  • standardized inputs,
  • expected outputs,
  • scoring rules,
  • model-by-model comparisons.

中文版本（简要）

AI Research Skills 的核心观点是：在 AI 辅助研究成为常态之后，**skill（技能）**会成为最小的可复用单元——它既是“教人怎么做事”的手册，也可以直接用来“教 AI 怎么做事”。

与传统 prompt collection 相比，skills 更强调：

• 更小的 scope（面向具体任务与流程）
• 可组合与可扩展（相互引用、按需加载）
• 可检索与可维护（结构化元数据 + 校验）
• 与工具链结合（当文字说明不够时，用 MCP 把“软件能力”暴露给 AI）

最终效果是：你写 skills，其实就是在写你未来的“软件能力栈”。