cn-source-research

Readme

# cn-source-research

> **EN:** Generate *verifiable evidence packages* for China market, policy, industry, competitor and technology research — not polished-looking but unverifiable reports.
> **中文：** 为中国市场、政策、行业、竞品与技术调研生成"可核查证据包"，而不是看似专业却不可验证的报告。

---

## Demo / 演示

![Demo animation](docs/demo.svg)

> **中文：** 上图为动画占位符。请替换为真实约 30 秒 GIF，建议录制完整流程：界定研究边界 → 建立 source ledger → 结论分级（事实/推论/假设）→ 冲突并列 → 生成四件交付物。
>
> **EN:** The SVG above is an animated placeholder. Replace it with a real ~30s GIF showing the full workflow: define scope → build source ledger → classify claims (fact/inference/hypothesis) → present conflicts in parallel → emit the four deliverables.

---

## What it is / 这是什么

`cn-source-research` 是一个开源、厂商中立的 Agent Skill，遵循开放 Agent Skills 格式，可被 Codex、Claude Code、Cursor、Gemini CLI 等宿主加载。它强制 AI 在中文调研中：

- **优先官方与一手来源**（政府 / 监管 / 公司公告 / 原始论文数据集）
- **区分事实、推论、未知**
- **每条重要结论可回链验证**
- **冲突信息并列呈现，不强行合并**
- **不捏造**网页、文献、数据或引文

**EN:** `cn-source-research` is an open, vendor-neutral Agent Skill in the open Agent Skills format, loadable by Codex, Claude Code, Cursor, Gemini CLI and similar hosts. It forces AI doing Chinese research to: prioritize official/primary sources; separate facts from inferences from unknowns; make every key conclusion back-link verifiable; present conflicts in parallel; and never fabricate web pages, literature, data or citations.

**目标用户 / Target users：** 创业者、咨询顾问、产品经理、投资研究员、跨境团队、学生 / entrepreneurs, consultants, PMs, investment researchers, cross-border teams, students.

---

## Install / 安装

> **重要：** 三个宿主各自独立安装，目录不共享。重复文件无害，但请分别执行命令。
>
> **Important:** Install separately into each host's own skills directory — they do not share folders.

### Claude Code

```bash
mkdir -p ~/.claude/skills
cp -r cn-source-research ~/.claude/skills/
# Windows (PowerShell):
# New-Item -ItemType Directory -Force -Path $HOME\.claude\skills
# Copy-Item -Recurse cn-source-research $HOME\.claude\skills\
```

### Cursor

```bash
mkdir -p ~/.cursor/skills
cp -r cn-source-research ~/.cursor/skills/
# Windows (PowerShell):
# New-Item -ItemType Directory -Force -Path $HOME\.cursor\skills
# Copy-Item -Recurse cn-source-research $HOME\.cursor\skills\
```

### Codex

```bash
mkdir -p ~/.codex/skills
cp -r cn-source-research ~/.codex/skills/
# Windows (PowerShell):
# New-Item -ItemType Directory -Force -Path $HOME\.codex\skills
# Copy-Item -Recurse cn-source-research $HOME\.codex\skills\
```

> 安装后在任一宿主中提出调研需求（如"梳理 2025 年中国新能源汽车充电桩监管政策"），Skill 会自动触发。
>
> After installing, raise a research need in any host (e.g. "summarize 2025 China NEV charging-pile regulation") and the Skill triggers automatically.

---

## Research before / after · 研究前后对比

| 维度 / Dimension | Before（普通 AI 调研） | After（本 Skill） |
| --- | --- | --- |
| **来源 / Sources** | 大段叙述，无台账 / Narrated prose, no ledger | `source-ledger.csv`，逐条登记含级别与失效风险 |
| **结论可信度 / Claim trust** | 事实与推测混为一谈 / Facts and guesses mixed | 三级分级：已证实事实 / 合理推论 / 待验证假设 |
| **冲突 / Conflicts** | 静默选一边 / Silently picks one side | 并列呈现，列入 open-questions |
| **可验证性 / Verifiability** | 结论无法回链 / Claims not back-linkable | 每条结论附出处 id、适用范围、失效风险 |
| **诚信 / Integrity** | 可能补全无法访问的内容 / May fill in inaccessible content | 明确标注限制，不捏造 |
| **高风险领域 / High-risk** | 给出似是而非的结论 / Plausible-but-unlicensed conclusions | 仅做证据准备，标注"不构成专业意见" |
| **交付物 / Deliverables** | 一篇长文 / One long essay | 四件套：brief + ledger + claim-map + open-questions |

---

## Sample / 样例

真实但不含敏感数据的样例见 `examples/`，主题为"中国新能源汽车充电基础设施政策调研"，展示四件交付物的完整格式。

> **注意：** 样例中具体数值与链接为占位符（标记 `[需核实]`），仅演示格式；实际使用时须替换为经核实的真实来源。本 Skill 不在样例中捏造任何具体数据或引文。
>
> **Note:** Figures and links in the sample are placeholders (marked `[需核实]`) demonstrating format only; replace with verified real sources in actual use. The Skill never fabricates specific data or citations in the sample.

文件清单 / Files:

- `examples/research-brief.md`
- `examples/source-ledger.csv`
- `examples/claim-evidence-map.md`
- `examples/open-questions.md`

---

## Privacy / 隐私说明

- **本地 Markdown 工作流优先**：核心流程不调用任何外部服务，不需要 API key，不上传内容。
- 浏览器抓取与数据库连接为**可选扩展**，需用户显式开启并自行承担合规责任。
- 本 Skill 不收集、不外传任何用户内容。

**EN:**

- **Local Markdown workflow first:** the core flow calls no external service, needs no API key, uploads nothing.
- Browser fetching and database connections are **optional extensions** the user must explicitly enable and own compliance for.
- This Skill collects and transmits no user content.

---

## High-risk domains / 高风险领域

医疗、法律、金融、监管合规：仅做信息整理与证据准备，不替代执业医师 / 律师 / 持牌分析师 / 合规专业人士。简报顶部会加注"不构成专业意见"。

**EN:** Medical, legal, financial, and regulatory-compliance domains: information organization and evidence preparation only — never a substitute for licensed professionals. Briefs carry a "not professional advice" notice.

---

## Evaluation / 评测

见 `EVALUATION.md`，含 20+ 用例，重点测试：来源过期、互相冲突、结论超出证据、找不到可靠来源、付费墙来源。

See `EVALUATION.md` for 20+ cases targeting: expired sources, conflicting sources, conclusion-exceeds-evidence, no-reliable-source, paywall sources.

---

## References / 参考模板

- `references/中国政策与监管来源目录模板.md`
- `references/行业研究问题模板.md`
- `references/竞争分析框架.md`
- `references/引用格式规范.md`

---

## License

MIT — see [LICENSE](LICENSE). `Copyright (c) 2026 whaojie797-design`.