Market Research Skills

中文版本见本文件下半部分 / Chinese version below

Turn Claude — or any LLM terminal — into a disciplined research analyst.

A collection of Claude Skills covering the full market / equity / industry research workflow: verify a fact against primary sources, spin up a topic briefing, or draft a flagship report — each step is its own skill. Every skill enforces the discipline real analysts live by: cite the original source, flag what can't be verified, never fabricate a number. Use any skill standalone, or chain them into an end-to-end pipeline. Runs in Claude Code / Desktop and ports to other LLM terminals (Codex / Gemini / Copilot).

Skills currently included

Skill	Purpose	Typical triggers
verifying	Trace a statement back to whitelisted primary sources	"verify X" / "is this true" / "find the original source"
topic-brief	Generate a thematic observation briefing (HTML, paste-into-WeChat-ready) for any subject (region / industry / issue / institution)	"做一份 XX 观察" / "generate a briefing on Y" / "/topic-brief"
analyst-research	Three-mode end-to-end research workflow. User picks scope at trigger: light (4-5 page memo, 0 charts, ~15 min), medium (12-15p topic analysis, 6-10 charts, ~1 h), or heavy (flagship 30-40p / 15k+ word report, 25-35+ charts, ~2-3 h, full multi-stage workflow: framing → sourcing → analysis → drafting → review, multi-LLM optional). Reports default to English; the AI replies in the user's chat language. Battle-tested on the Saudi Vision 2030 deep-dive.	"research report" / "投研报告" / "深度分析" / "做研报" / "5-page memo" / "/analyst-research"
local-vault	Build & query a local Markdown knowledge base: convert PDF / Office / images / code into retrieval-friendly Markdown (local-first, cloud OCR fallback), then answer questions over the vault with retrieval discipline (coverage self-checks, lossy-content flags, MOC proposals)	"build/sync my local knowledge base" / "convert these files to markdown for AI" / "整理我的资料库" / "本地知识库"

The list grows with each release. Full change history in CHANGELOG.md.

Installation

Option 1 — Claude Code one-click `/plugin install` (recommended for Claude Code users)

This repo ships with .claude-plugin/marketplace.json — it is itself a marketplace. In Claude Code:

/plugin marketplace add https://github.com/genli-ai/market-research-skills.git
/plugin install market-research-skills@market-research-skills

Future updates in one command:

/plugin update market-research-skills@market-research-skills

Option 2 — Clone the entire repo to `~/.claude/skills/` (works for Claude Desktop / other LLM terminals)

git clone https://github.com/genli-ai/market-research-skills.git
cp -r market-research-skills/skills/* ~/.claude/skills/

Subsequent updates:

cd market-research-skills && git pull
cp -r skills/* ~/.claude/skills/

Option 3 — Install only one skill

Use sparse-checkout to pull a single subdirectory:

git clone --filter=blob:none --no-checkout https://github.com/genli-ai/market-research-skills.git
cd market-research-skills
git sparse-checkout init --cone
git sparse-checkout set skills/verifying
git checkout
cp -r skills/verifying ~/.claude/skills/

Option 4 — Download a packaged `.zip` file

Each Release attaches a .zip file for every skill. After download:

unzip verifying.zip -d ~/.claude/skills/verifying

Option 5 — Cross-terminal install (Codex / Gemini / Copilot / etc.)

A .zip file simply contains SKILL.md and any resources — the format is portable across every LLM terminal. To install on non-Claude terminals:

Unzip the .zip file:
```
unzip verifying.zip -d ./verifying
```
Move the unzipped folder into your terminal's skills directory (check your terminal's docs):
- Codex CLI: typically ~/.codex/skills/ (verify in current Codex docs)
- Gemini CLI: ~/.gemini/skills/ (verify in Gemini docs)
- Other terminals: see the terminal's plugin / skill loader documentation

Tool-call action verbs inside each SKILL.md (read full text / fetch web body / search-engine query / image recognition / database query) are described in generic semantic terms, so each terminal's LLM can map them to its own local tool set.

Option 6 — `npx skills` / skills.sh (one command, 60+ terminals)

This repo is indexed by skills.sh (the open npx skills tool by Vercel Labs). One command installs straight from this repo — no clone, no manual copy — and always resolves the latest from main, so there is no version to pin:

# Install all skills into Claude Code
npx skills add genli-ai/market-research-skills -a claude-code -s '*'

# Pick one skill, and/or a different terminal (cursor / github-copilot / gemini-cli / codex / ...)
npx skills add genli-ai/market-research-skills -s verifying -a cursor

# Preview what's available without installing
npx skills add genli-ai/market-research-skills --list

Update later with npx skills update. Omit -a to auto-detect installed terminals and choose interactively.

Directory layout

market-research-skills/
├── README.md                       # This file
├── LICENSE                         # MIT
├── CHANGELOG.md                    # Version history
├── .claude-plugin/
│   ├── plugin.json                 # Plugin manifest
│   └── marketplace.json            # Marketplace manifest (self-registering)
├── skills/
│   └── <skill-name>/
│       ├── SKILL.md                # Canonical English version (loaded by LLM)
│       ├── SKILL.zh.md             # Chinese reference (kept in sync; not loaded)
│       ├── README.md               # Skill-specific docs (optional)
│       ├── references/             # Reference materials (optional)
│       ├── scripts/                # Skill-specific scripts (optional)
│       └── assets/                 # Static assets (optional)
├── scripts/
│   └── pack.sh                     # Package a skill as a .zip file
└── releases/                       # Locally generated .zip artifacts (gitignored)
    └── <skill-name>.zip            # Easy to copy and share directly

Every skill must have a SKILL.md (English, canonical, loaded by LLMs) whose frontmatter contains at minimum:

---
name: <skill-name>           # Must match the folder name
description: <When to trigger and what the skill does — written in English>
---

A SKILL.zh.md may co-exist as a Chinese reference. The two files must be kept in sync when edited.

Packaging a skill as `.zip`

./scripts/pack.sh verifying
# Output: releases/verifying.zip

The packaged .zip only includes the canonical SKILL.md plus other resources — the Chinese reference SKILL.zh.md is excluded.

Distributing the `.zip`

There are two distribution paths:

Manual sharing: copy releases/<skill-name>.zip and send to anyone directly (chat, email, Slack, etc.).
Public download via GitHub Release: when you publish a Release on GitHub, attach the .zip file. Users download it from the Releases page.

The releases/ folder is .gitignored — the repo does not store binary artifacts. Each Release on GitHub carries the .zip as an attached asset.

Contributing / adding a new skill

Create a new folder under skills/. Folder name = skill name (lowercase + hyphens).
Create a SKILL.md (English, canonical) with proper frontmatter.
Optionally maintain a SKILL.zh.md Chinese reference — keep both in sync when modified.
Add a row to the "Skills currently included" table in this README.
Record the addition in CHANGELOG.md.

License

MIT

市场研究 Skills（中文版）

把 Claude（或任意 LLM 终端）变成一个讲纪律的研究分析师。

一组覆盖市场研究 / 投研 / 行业分析完整工作流的 Claude Skills：核实一个事实、产出一份主题简报、撰写一篇旗舰研报——每个环节都是一个独立 skill。每个 skill 都贯彻真实分析师的纪律：引用一手来源、标注无法核实的内容、绝不编造数字。可单独用任一 skill，也可串成端到端流水线。支持 Claude Code / Desktop，并可移植到其他 LLM 终端（Codex / Gemini / Copilot）。

当前包含的 Skills

Skill	用途	典型触发语
verifying	信息核实：把陈述追溯到白名单内的一手来源	「帮我核实 X」「这条信息是真的吗」「找一下原始出处」
topic-brief	主题观察简报生成器：为任意主题（区域 / 行业 / 议题 / 机构）产出可粘贴公众号的 HTML 简报	「做一份 XX 观察 / 简报」「`/topic-brief`」
analyst-research	三档端到端研究工作流。用户触发时选档：light（4-5 页备忘、0 图、约 15 分钟）/ medium（12-15 页主题分析、6-10 图、约 1 小时）/ heavy（30-40 页 / 1.5 万字+ 旗舰报告、25-35+ 图、约 2-3 小时、完整多阶段工作流：框定 → 取数 → 分析 → 起草 → 复盘、可选多 LLM 协作）。报告默认英文，AI 按用户聊天语言回复。已在沙特 Vision 2030 深度报告项目跑通。	「写研报」「投研报告」「深度分析」「主题分析」「5 页 memo」「`/analyst-research`」
local-vault	本地 Markdown 知识库：把 PDF / Office / 图片 / 代码转成带检索 frontmatter 的 Markdown（本地优先、云端 OCR 兜底），再基于 vault 负责任地回答问题（覆盖度自检、有损内容标注、MOC 沉淀）	「整理我的资料库」「把文件转成 md 给 AI 读」「本地知识库」「读我的本地 vault 回答」

列表会随版本更新。完整变更见 CHANGELOG.md。

安装

方式一：Claude Code 一键 `/plugin install`（推荐 Claude Code 用户）

本 repo 自带 .claude-plugin/marketplace.json，本身就是一个 marketplace。在 Claude Code 里：

/plugin marketplace add https://github.com/genli-ai/market-research-skills.git
/plugin install market-research-skills@market-research-skills

未来更新一键完成：

/plugin update market-research-skills@market-research-skills

方式二：整体克隆到 ~/.claude/skills/（适合 Claude Desktop / 其他 LLM 终端）

git clone https://github.com/genli-ai/market-research-skills.git
cp -r market-research-skills/skills/* ~/.claude/skills/

后续更新：

cd market-research-skills && git pull
cp -r skills/* ~/.claude/skills/

方式三：只装某一个 skill

不想要全套时，用 sparse-checkout 只拉某个子目录：

git clone --filter=blob:none --no-checkout https://github.com/genli-ai/market-research-skills.git
cd market-research-skills
git sparse-checkout init --cone
git sparse-checkout set skills/verifying
git checkout
cp -r skills/verifying ~/.claude/skills/

方式四：下载打包好的 `.zip` 文件

每个 Release 会附上每个 skill 单独的 .zip 文件。下载后：

unzip verifying.zip -d ~/.claude/skills/verifying

方式五：跨终端安装（Codex / Gemini / Copilot 等）

.zip 文件里装着 SKILL.md 与资源文件——zip 是通用格式，任意 LLM 终端都能直接解压使用。安装到非 Claude 终端的步骤：

解压 .zip 文件：
```
unzip verifying.zip -d ./verifying
```
把解压后的文件夹放进你终端的 skills 目录（具体路径查阅各自文档）：
- Codex CLI：一般在 ~/.codex/skills/（以最新 Codex 文档为准）
- Gemini CLI：一般在 ~/.gemini/skills/（以 Gemini 文档为准）
- 其他终端：见各自的 plugin / skill 加载器文档

每个 SKILL.md 里的工具调用动作（通读全文 / 抓取网页正文 / 搜索引擎检索 / 图像识别 / 数据库查询）都用通用语义动词描述，让各终端的 LLM 映射到自家工具。

方式六：`npx skills` / skills.sh（一条命令，支持 60+ 终端）

本 repo 已被 skills.sh（Vercel Labs 的开源 npx skills 工具）收录。一条命令直接从本 repo 安装——无需克隆、无需手动拷贝——且每次都拉取 main 的最新版，没有版本号要锁定：

# 全部 skill 装进 Claude Code
npx skills add genli-ai/market-research-skills -a claude-code -s '*'

# 只装某一个，或换一个终端（cursor / github-copilot / gemini-cli / codex / ...）
npx skills add genli-ai/market-research-skills -s verifying -a cursor

# 不安装，仅预览有哪些 skill
npx skills add genli-ai/market-research-skills --list

之后用 npx skills update 更新。省略 -a 会自动检测已装终端并让你交互选择。

目录约定

market-research-skills/
├── README.md                       # 本文件
├── LICENSE                         # MIT
├── CHANGELOG.md                    # 版本变更
├── .claude-plugin/
│   ├── plugin.json                 # 本 repo 作为 Claude Code plugin 的清单
│   └── marketplace.json            # 本 repo 同时作为 marketplace（自我注册）
├── skills/
│   └── <skill-name>/
│       ├── SKILL.md                # 英文权威版（被 LLM 加载）
│       ├── SKILL.zh.md             # 中文参考版（与 SKILL.md 同步；不被 LLM 加载）
│       ├── README.md               # 该 skill 的独立说明（可选）
│       ├── references/             # 引用的资料（可选）
│       ├── scripts/                # 该 skill 自带的脚本（可选）
│       └── assets/                 # 静态资源（可选）
├── scripts/
│   └── pack.sh                     # 打包某个 skill 为 .zip 文件
└── releases/                       # 本地生成的 .zip 产物（.gitignored）
    └── <skill-name>.zip            # 方便你直接拷贝发给别人

每个 skill 必须有一个 SKILL.md（英文权威版，被 LLM 加载），文件顶部 frontmatter 至少包含：

---
name: <skill-name>           # 与文件夹名一致
description: <一句话英文：什么时候该触发、做什么事>
---

可同时维护一个 SKILL.zh.md 中文参考版。两份在修改时必须保持同步。

打包 skill 成 `.zip`

./scripts/pack.sh verifying
# 输出 releases/verifying.zip

打包好的 .zip 只包含权威的 SKILL.md 及其他资源——中文参考版 SKILL.zh.md 不会被打进 zip。

分发 `.zip`

两条分发路径：

手动分发：把 releases/<skill-name>.zip 拷出来，微信 / 邮件 / Slack 直接发给任何人
GitHub Release 公开下载：发 Release 时把 .zip 作为附件挂上去，用户在 Releases 页一键下载

releases/ 文件夹本身在 .gitignore 里——repo 历史不存二进制产物。每次发版的 .zip 走 GitHub Release 的 attachment 路径。

贡献 / 新增 skill

在 skills/ 下新建一个文件夹，名字 = skill name（小写 + 短横线）
创建 SKILL.md（英文权威版），写好 frontmatter
可选维护 SKILL.zh.md 中文参考——修改时两份同步
在本 README 上半部分（英文版）+ 下半部分（中文版）的 Skills 表格里各加一行
在 CHANGELOG.md 记录本次新增

许可

MIT

Market Research Skills

Skills currently included

Installation

Option 1 — Claude Code one-click /plugin install (recommended for Claude Code users)

Option 2 — Clone the entire repo to ~/.claude/skills/ (works for Claude Desktop / other LLM terminals)

Option 3 — Install only one skill

Option 4 — Download a packaged .zip file