克劳德代码/法典的单一独立研究技能。

一个围绕单一研究脊梁构建的自成一体的技能档案。

英文 |中文

这项技能的作用

researcher 用于以下研究任务：

• 开放式
• 多源
• 判断力强
• 容易被炒作、薄弱的总结或片面叙述所影响

它设计用于以下工作：

• 绘制新的行业或领域地图
• 调查公司或产品
• 验证主张或叙述
• 比较路径并做出决策
• 寻找真实的机会与喧嚣叙事
• 理解从业者的现实与隐性摩擦
• 回顾学术文献和最新技术

建筑

一项整合的研究技能：

skills.md 包含完整方法：准备、信息收集， 垂直历史、水平比较、横轴洞察、对象类型 适应、证据和引用规则、输出结构、写作标准， 以及最终的质量检查清单。

技能主体以中文编写以便阅读。YAML元数据保持 英语触发词汇，所以英语研究提示仍能引导到它。

核心设计

1.单一核心方法

Researcher

A single standalone research skill for Claude Code / Codex.

One self-contained skill file built around a single research spine.

VERTICAL history + HORIZONTAL comparison -> CROSS-AXIS judgment

English | 中文

What This Skill Does

researcher is for research tasks that are:

• open-ended
• multi-source
• judgment-heavy
• vulnerable to hype, weak summaries, or one-sided narratives

It is designed for work like:

• mapping a new industry or field
• investigating a company or product
• validating a claim or narrative
• comparing paths and making decisions
• finding real opportunities vs loud narratives
• understanding practitioner reality and hidden friction
• reviewing academic literature and state of the art

Architecture

One consolidated research skill:

researcher/
└── skills.md                         ← horizontal-vertical research method

skills.md contains the full method: preparation, information collection, vertical history, horizontal comparison, cross-axis insight, object-type adaptation, evidence and citation rules, output structure, writing standards, and a final quality checklist.

The skill body is written in Chinese for readability. The YAML metadata keeps English trigger terms so English research prompts can still route to it.

Core Design

1. One Core Method

The skill is not a pile of research tactics. Every investigation follows one structure:

• Vertical axis: how the object became what it is.
• Horizontal axis: where it stands now among peers, substitutes, users, and incentives.
• Cross-axis judgment: what the combination reveals.

2. Evidence Supports the Axes

Evidence work is organized around the two axes instead of accumulated for its own sake. The skill still distinguishes official, behavioral, operator, lived experience, adversarial, market proxy, and artifact sources.

3. Challenge Before Judgment

Every strong thesis must be attacked before synthesis: criticism, failed cases, methodology issues, adversarial sources, incentives, and behavioral evidence all matter.

4. Adaptive Output

The skill defaults to a concise chat answer. It creates a durable research artifact only when the user asks for one, provides an output path, or the work is too large to remain useful inline.

Installation

Clone:

git clone https://github.com/recomby-ai/researcher-skill.git

Install to Claude Code:

mkdir -p ~/.claude/skills/researcher
cp researcher-skill/researcher/skills.md ~/.claude/skills/researcher/SKILL.md

Install to Codex:

mkdir -p ~/.codex/skills/researcher
cp researcher-skill/researcher/skills.md ~/.codex/skills/researcher/SKILL.md

Web Upload

If your runtime requires the entry file to be named SKILL.md, rename researcher/skills.md to SKILL.md before uploading. The file is otherwise self-contained.

Example Prompts

Use $researcher to map the AI agent engineering job market.