agent-skill-industry-research

Agent Skill: Industry Research（行业研究 Agent）

An autonomous agent skill that conducts comprehensive industry research and produces professional Word reports (.docx). Covers market sizing, competitive landscape, company analysis, and investment recommendations — with built-in cognitive safeguards and Red Team self-challenge.

一个自主完成行业研究并输出专业 Word 报告的 Agent Skill。适用于投资研究、创业赛道分析、竞争格局扫描等场景。

---

Table of Contents / 目录

• Features
• Installation (15+ platforms)
• Usage
• Research Depth Levels
• Analysis Frameworks
• Best Practices
• Customization
• Project Structure
• Dependencies
• License

---

Features

• Autonomous workflow — 4-phase process: Intake → Research → Report Generation → Human Review
• Three depth levels — L1 Quick Memo (2-3 pages), L2 Sector Scan (15-40 pages), L3 Deep Dive (40+ pages)
• Multilingual — output language follows input language; bilingual terminology annotation (e.g., "护城河（Moat）")
• 9 analysis frameworks — 5W1H, PEST, Top-Down/Bottom-Up market sizing, Lanchester's Laws, Porter's Five Forces, Moat Theory, Gartner Hype Cycle, Crossing the Chasm, Growth Flywheel
• Cognitive safeguards — 12 cognitive traps identification + 5-level cross-validation methodology
• Red Team self-challenge — automated bear-case stress testing of all conclusions
• Professional .docx output — styled Word reports with data tables, bilingual terms, and source citations

功能亮点： 自主四阶段工作流、三级研究深度、中英双语支持、9 大分析框架、12 种认知陷阱防护、Red Team 自我挑战、专业 Word 报告输出。

---

Installation

First, clone the repository:

git clone https://github.com/147356/agent-skill-industry-research.git

Then follow the instructions for your agent platform below.

---

Category A: Coding Agents (with file system access)

These agents can read local files, execute code, and generate .docx reports autonomously.

User-level skill (available in all conversations):

cp -r agent-skill-industry-research ~/.claude/skills/industry-research

Project-level skill (available only in that project):

cp -r agent-skill-industry-research /path/to/your/project/.claude/skills/industry-research

The skill will be automatically discovered by Claude Code.

cp -r agent-skill-industry-research /mnt/.claude/skills/industry-research

Copy SKILL.md and references/ into your project, then reference them in your AGENTS.md:

cp agent-skill-industry-research/SKILL.md /path/to/your/project/
cp -r agent-skill-industry-research/references /path/to/your/project/

Add to your AGENTS.md:

## Industry Research
See [SKILL.md](SKILL.md) for the full agent instructions.
Reference files are in the `references/` directory.

Add the skill as a project rule:

mkdir -p /path/to/your/project/.cursor/rules
cp agent-skill-industry-research/SKILL.md /path/to/your/project/.cursor/rules/industry-research.md
cp -r agent-skill-industry-research/references /path/to/your/project/references

cat agent-skill-industry-research/SKILL.md >> /path/to/your/project/.windsurfrules
cp -r agent-skill-industry-research/references /path/to/your/project/references

mkdir -p /path/to/your/project/.github
cp agent-skill-industry-research/SKILL.md /path/to/your/project/.github/copilot-instructions.md
cp -r agent-skill-industry-research/references /path/to/your/project/references

---

Category B: Autonomous AI Agents

These agents can browse the web and execute multi-step tasks. Paste SKILL.md as the task prompt, and upload references/frameworks.md + references/traps-and-validation.md as attachments.

Platform	How to Use
Manus	Create a new task → paste SKILL.md content as the task description → upload reference files as attachments
OpenClaw	Add as a custom skill/prompt in your agent configuration
Devin	Paste SKILL.md as the task prompt in a new session

---

Category C: AI Assistants (chat-based, with web search)

These platforms support long-context input and web search. Best suited for L1 Quick Memos. For L2/L3 reports, use a Category A or B agent for better file generation.

Platform	How to Use
Kimi (Moonshot AI)	Open a new conversation → paste SKILL.md as the first message → then describe your research need. Kimi's long-context window (200K+) can handle the full prompt. Upload reference files directly in chat.
MiniMax / Hailuo AI	Paste SKILL.md into system prompt or first message → start research conversation
ChatGPT (with web browsing)	Use as Custom Instructions or paste into conversation → enable web browsing
Gemini	Paste SKILL.md as context in Google AI Studio or Gemini chat
DeepSeek	Paste SKILL.md into conversation → supports web search for research phase
通义千问 (Qwen)	Paste into conversation or use as agent system prompt in Alibaba Cloud
文心一言 (ERNIE Bot)	Paste into conversation → supports web search
豆包 (Doubao)	Create a custom bot → paste SKILL.md as bot instructions

Tip: When using chat-based assistants, paste SKILL.md first, then paste references/frameworks.md and references/traps-and-validation.md as follow-up messages before starting your research request. This gives the agent the full analytical toolkit.

---

Category D: Agent Building Platforms

Build a reusable industry research bot using these platforms.

Platform	How to Use
Coze / 扣子 (ByteDance)	Create a new bot → paste SKILL.md as system prompt → upload reference files to the knowledge base → enable web search plugin
Dify	Create a new Agent app → paste SKILL.md as the system prompt → add reference files as context → enable web search tool
FastGPT	Create a new app → use SKILL.md as prompt template → import reference files as knowledge base documents
Langflow / Flowise	Use SKILL.md as the system message in a conversational agent chain → add web search and file tools

---

Compatibility Summary

Capability Required	For L1	For L2/L3
Long context (50K+ tokens)	Required	Required
Web search	Required	Required
File generation (.md, .docx)	Optional	Required
Code execution (Node.js)	Not needed	Required for .docx

安装方式： 支持 Claude Code、Codex、Cursor、Windsurf、Copilot 等编程 Agent，Manus、Devin 等自主 Agent，Kimi、ChatGPT、DeepSeek、通义千问等 AI 助手，以及 Coze、Dify 等 Agent 搭建平台。核心文件 SKILL.md 是通用结构化 prompt，适配任何支持联网搜索的 AI 平台。

---

Usage

Trigger the skill by describing your research need in natural language:

# English
"Analyze the AI legal tech vertical for me"
"Is the rescue robotics sector worth entering?"
"Produce an L2 report on cross-border e-commerce SaaS"

# 中文
"帮我研究一下 XX 行业"
"XX 赛道值不值得进"
"做一份 XX 的行研报告"
"XX 市场有多大"

The agent will begin with a brief intake (2-3 questions to confirm scope), then autonomously conduct research and deliver a .docx report.

---

Research Depth Levels

Level	Description	Output	Use Case
L1 Quick Memo	Rapid scan	2-3 pages	Quick screening, "should I look deeper?"
L2 Sector Scan	Thorough analysis	15-40 pages	Investment decisions, sector evaluation
L3 Deep Dive	Exhaustive research	40+ pages	Major strategic decisions

Production-validated example: Rescue Robotics L2 report — 36 pages, 40,600 Chinese characters, 18 data tables, 5 chapters + 5 appendices. See examples/sample-L2-toc.md for the full table of contents.

---

Analysis Frameworks

Framework	Purpose
5W1H	Industry definition and scoping
PEST	Macro driving forces & constraints
Top-Down + Bottom-Up	Dual market sizing with convergence check
Lanchester's Laws	Market structure classification (5 types)
Porter's Five Forces	Industry structural analysis
Moat Theory	Competitive advantage assessment
Gartner Hype Cycle	Technology maturity positioning
Crossing the Chasm	Adoption lifecycle judgment
Growth Flywheel	Business model feedback loop identification

Full framework details: references/frameworks.md Cognitive traps & validation: references/traps-and-validation.md

---

Best Practices

Reference materials dramatically improve quality

With 3+ detailed reference files (broker reports, company analyses, investment memos), the agent reliably produces investment-grade output. Web search alone yields noticeably thinner reports.

Recommended reference types:

• Broker / consulting firm industry reports
• Target company prospectuses or financial reports
• Existing internal analysis documents
• Competitor materials

提供参考资料能大幅提升质量： 当用户提供 3+ 份详细参考资料，Agent 能稳定输出投资级别的专业报告。推荐提供券商研报、招股书、内部分析文档等。

---

Customization

What to customize	File to edit
Report structure & templates	SKILL.md — "Report Templates by Level" section
Analysis frameworks	references/frameworks.md
Cognitive traps & validation	references/traps-and-validation.md

---

Project Structure

agent-skill-industry-research/
├── SKILL.md                     # Core skill file (agent instructions)
├── README.md                    # This file
├── LICENSE                      # MIT License
├── CHANGELOG.md                 # Version history
├── .gitignore                   # Git ignore rules
├── evals/
│   └── evals.json               # Evaluation test cases (4 scenarios)
├── examples/
│   └── sample-L2-toc.md         # Real L2 report sample (TOC + table list)
└── references/
    ├── frameworks.md             # Analysis frameworks (PEST, Porter, Lanchester, etc.)
    └── traps-and-validation.md   # 12 cognitive traps + cross-validation methodology

---

Dependencies

• For L1 reports: Any AI assistant with web search (Kimi, ChatGPT, DeepSeek, etc.)
• For L2/L3 reports: An AI agent with web search + file generation + code execution (Claude Code, Codex, Cursor, Manus, etc.)
• For .docx generation: Node.js + docx npm package
• Optional: Python + python-docx package (for docx validation)

---

Contributing

Contributions are welcome! Feel free to:

• Open an issue for bugs or feature requests
• Submit a PR for improvements
• Share your generated reports as examples (anonymized)

---

License

MIT License. See LICENSE for details.