flicy/enterprise-company-research-agentGit
Company research and due diligence
Research businesses and verify evidence for procurement and investment decisions.
Before you start
Inputs: Research question, source scope and output requirements
- Check the original documentation for supported AI tools, models and runtime dependencies.
- Open the repository and follow the author’s installation instructions. Model APIs or external services may require separate accounts.
- Try your own material, then review the facts, layout and output files.
License: MIT
Source snapshot: 2026-09-15. README ↗
Original license & copyright
MIT License Copyright (c) 2026 Enterprise Company Research Agent contributors Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
Try a task like this
Use Company research and due diligence with my own source material. Confirm the required inputs and available tools, then produce an editable result.
Enterprise Company Research Agent
简体中文 | English below
给企业客户做供应商、公司与产品调研的 Agent Skill：先核验对象是谁和现在怎么样，再输出能支持采购、合作、竞品、客户研究和投融资初筛的证据型报告。
Enterprise Company Research Agent 是一个轻量但结构化的 企业调研 skill。用户只需要先说清楚想研究谁、为了什么决策，它会帮助 Agent 把公开资料整理成：
- 身份与范围核验
- 从成立/发布到今天的发展脉络
- 产品能力和商业化状态
- 竞品、替代供应商和上下游依赖对比
- 风险、合规、数据安全和交付能力判断
- 可复核的证据账本
- 采购或合作建议
它的重点不是写一篇公司介绍，而是把“能不能采购、能不能合作、能不能依赖、还缺什么尽调”变成可检查的结论。
适合什么场景
如果你希望：
- 评估一个新供应商能不能进入采购名单
- 判断一个产品是否真实商用、是否有可信客户
- 对比多个竞品、替代供应商或合作伙伴
- 快速弄清楚某家公司、子品牌、产品线和母公司的关系
- 为采购、法务、风控、安全、销售、战略或投委会准备材料
- 获取最新、准确、带来源的企业和产品研究
那么这个 Skill 就很适合你。
如果你只是需要三句话百科解释、纯营销文案或不需要来源的灵感整理，它通常不是最佳入口。
一句话需求进来，它怎么接住
用户不需要先填完整表格，也不需要知道公司法律主体。直接用业务语言说：
帮我调研这家公司能不能作为供应商。
研究 Foodpal 现在是什么情况、产品是什么、团队背景如何。
比较 A、B、C 三家公司，看看哪家更适合进入 shortlist。
Agent 会先补齐研究边界，再按企业决策场景选择合适的深度。
| 输入清晰度 | Agent 怎么处理 | 输出重点 |
|---|---|---|
| 只有公司或产品名 | 先确认官网、法律主体、品牌/产品边界和同名实体 | 身份核验、当前状态、未核验项 |
| 已有官网和决策场景 | 直接进入证据检索、产品能力、客户证明和风险分析 | 标准企业调研报告 |
| 已给竞品或备选供应商 | 增加横向对比、替代性和采购建议 | shortlist / vendor comparison |
| 已给风控或合规关注点 | 优先检查诉讼、处罚、认证、数据安全、制裁和出口管制 | 风险 memo 和补充尽调问题 |
用户需要输入什么
最小输入：
| 字段 | 示例 |
|---|---|
| 研究对象 | 公司、供应商、产品、品牌、集团、子公司、人物或行业概念 |
| 决策场景 | 采购准入、续约、竞品分析、客户研究、合作评估、投融资初筛或风险预警 |
更好的输入：
| 字段 | 示例 |
|---|---|
| 已知链接 | 官网、产品页、招聘页、新闻稿、工商或监管链接 |
| 地域范围 | 中国、美国、欧盟、东南亚、全球 |
| 关注问题 | 数据安全、交付能力、产品成熟度、价格、诉讼处罚、财务稳定性 |
| 对比对象 | 已知竞品、备选供应商、内部替代方案 |
| 输出要求 | 快速 memo、标准报告、完整尽调；Markdown、PDF、PPT、表格、双语 |
信息不完整也可以启动。Agent 会把合理推断和未核验项标出来，不会把猜测写成事实。
用户会得到什么
标准报告通常包含：
- 决策摘要：一句话结论、推荐动作、最高风险、最强证据。
- 身份与范围核验：法律主体、品牌、产品、母子公司、同名实体排除。
- 发展脉络：成立/发布背景、关键节点、战略转向、融资、客户和组织变化。
- 当前状态：经营状态、组织信号、融资/收入/客户线索、招聘与扩张迹象。
- 产品与能力：核心功能、技术路线、部署方式、价格/合同线索、成熟度。
- 客户与市场证明：客户案例、行业分布、采购人群、续费/替换/用户反馈线索。
- 横向对比：竞品、替代供应商、上下游依赖、能力/价格/交付/风险矩阵。
- 风险与合规：诉讼、处罚、舆情、制裁、出口管制、资质、隐私和安全认证。
- 情景判断：最可能、最危险、最乐观三种情景。
- 证据账本：已访问来源、关键事实表、冲突事实、不可达来源、下一步尽调问题。
和普通搜索 / 公司简介有什么不一样
| 方式 | 重心 | 用户主要得到什么 | 适合什么 |
|---|---|---|---|
| 普通搜索 | 找资料 | 链接、摘要、零散信息 | 快速了解一个名字 |
| 公司简介生成 | 写介绍 | 公司故事、产品描述 | 市场、销售或内部背景材料 |
| Enterprise Company Research Agent | 企业决策 | 证据账本、风险判断、横向对比、采购/合作建议 | 供应商准入、竞品研判、合作评估、投融资初筛 |
这个 Skill 最有价值的地方，是把“资料看起来很多”变成“哪些事实可信、哪些结论可用、哪些地方不能下判断”。
典型真实场景
| 场景 | 为什么适合这个 Skill | 主要产物 |
|---|---|---|
| 新供应商准入 | 企业需要知道法律主体、交付能力、客户证明和风险 | 供应商准入报告、风险清单、补充尽调问题 |
| 竞品或替代方案 shortlist | 不只是看功能，还要看成熟度、客户、价格和替代风险 | 横向对比表、采购建议、替代性判断 |
| 客户或合作伙伴研究 | 销售/BD 需要理解公司现状、产品和组织信号 | 客户画像、业务线索、合作切入点 |
| 投融资初筛 | 需要快速判断公司是否真实活跃、产品是否成熟 | 初筛 memo、证据等级、风险提示 |
| 品牌/子公司关系混乱 | 同名公司、产品线、海外主体容易混淆 | 身份核验、范围说明、冲突来源 |
| 负面舆情或合规预警 | 单一数据库容易误导，需要多源核验 | 风险 memo、冲突事实表、下一步核查 |
默认带上的能力
- 实时核验：涉及公司现状、融资、法务、产品能力、客户案例、价格、招聘和资质时，默认联网核验。
- 证据优先级：监管/工商/SEC/交易所文件优先，其次是官网、公告、文档、高管原话、客户案例、招聘、专利、开源仓库、权威媒体和第三方数据库。
- 事实与判断分离：每个关键判断都要能追溯到事实；推断必须明确标注。
- 对象边界控制：不把父公司数据误写成子产品数据，不把同名实体混为一谈。
- 采购视角：报告必须回答“能不能采购、合作或依赖它”，不只讲公司故事。
- 冲突处理：资料冲突时，把冲突写进证据账本，而不是静默选择一个说法。
仓库结构
enterprise-company-research-agent/
├── SKILL.md
├── README.md
├── LICENSE
├── skill.yml
├── agents/
│ └── openai.yaml
└── references/
└── enterprise-research-playbook.md
SKILL.md 是薄入口：触发条件、流程、深度控制和硬规则。references/enterprise-research-playbook.md 是深度参考：证据分级、检索入口与查询手法、研究模块、低证据协议、报告模板和质检门槛，agent 按需读取。
安装
Codex
git clone https://github.com/flicy/enterprise-company-research-agent.git
cp -R enterprise-company-research-agent ~/.codex/skills/enterprise-company-research-agent
Claude Code
git clone https://github.com/flicy/enterprise-company-research-agent.git
mkdir -p ~/.claude/skills
cp -R enterprise-company-research-agent ~/.claude/skills/enterprise-company-research-agent
QClaw / OpenClaw
git clone https://github.com/flicy/enterprise-company-research-agent.git
# 按你的 QClaw / OpenClaw skills 搜索路径复制本目录，入口保持 SKILL.md。
CodeBuddy / WorkBuddy 相关 coding agent
git clone https://github.com/flicy/enterprise-company-research-agent.git
mkdir -p .codebuddy/skills
cp -R enterprise-company-research-agent .codebuddy/skills/enterprise-company-research-agent
skill.yml 是 WorkBuddy 风格的元数据桥接文件；核心入口仍然是 SKILL.md。
快速开始
在支持 Agent Skills 的客户端里，新开会话后直接说：
使用 enterprise-company-research-agent，调研 [公司/产品名]，场景是 [采购/竞品/合作/投资/风险]。
更完整的提示词：
使用 enterprise-company-research-agent 调研 [公司名]。
决策场景：供应商准入。
关注点：法律主体、产品成熟度、客户案例、数据安全、诉讼处罚、交付能力。
对比对象：[竞品 A]、[竞品 B]。
输出：中文标准报告，附证据账本和下一步尽调问题。
质量与安全
- 这个 Skill 需要联网能力才能回答“最新情况”。没有网络时，必须明确标注当前事实无法核验。
- 不要把第三方数据库当成一手事实；关键数字和结论尽量回到原始来源。
- 不要提交客户资料、内部评审材料、访问 token、私有路径或账号信息到公开仓库。
- README 保持中英双语；
SKILL.md和references/是 agent 指令层，保持中文单语以控制加载成本。报告输出语言跟随用户，双语报告由报告层实现。
相关 Skill
| Skill | 用途 |
|---|---|
| wechat-publish-from-transcript | 把访谈/直播逐字稿变成公众号文章和跨平台内容包 |
| prepare-livestream-event | 筹备线上直播活动，生成海报、文案和发布清单 |
| publish-skill-to-github | 把本地 Skill 包发布到 GitHub |
许可证
MIT — 见 LICENSE
English Version
An Agent Skill for enterprise supplier, company, and product research: verify who the subject is and what its current state is, then produce evidence-backed reports for procurement, partnership, competitive analysis, customer research, and investment screening.
Enterprise Company Research Agent is a lightweight but structured enterprise research skill. The user only needs to say who they want to research and what decision they need to make. The Skill helps an agent turn public evidence into:
- identity and scope verification
- lifecycle timeline from founding or launch to today
- product capability and commercialization status
- competitor, substitute vendor, and dependency comparison
- risk, compliance, data security, and delivery-capacity judgment
- traceable evidence ledger
- procurement or partnership recommendation
Its job is not to write a company profile. Its job is to make “can we buy from it, partner with it, depend on it, or diligence it further?” inspectable.
When To Use
Use this Skill when you want to:
- evaluate whether a new supplier should enter a procurement shortlist
- check whether a product is commercially real and has credible customers
- compare competitors, substitute vendors, or potential partners
- clarify relationships between a company, subsidiary, brand, product line, and parent company
- prepare material for procurement, legal, risk, security, sales, strategy, or investment teams
- get current, accurate, source-backed company and product research
It is usually not the best entry point for three-sentence definitions, pure marketing copy, or source-free brainstorming.
How It Handles A One-Line Request
The user does not need to fill out a full form or know the legal entity first. They can start with business language:
Research whether this company is safe to onboard as a supplier.
Research Foodpal's current state, product, and team background.
Compare company A, B, and C and decide which one should enter the shortlist.
The agent first completes the research boundary, then chooses the right depth for the enterprise decision context.
| Input clarity | How the agent handles it | Output focus |
|---|---|---|
| Only company or product name | Verifies website, legal entity, brand/product boundary, and similarly named entities | Identity, current state, unverified items |
| Website and decision context provided | Runs evidence retrieval, product review, customer proof, and risk analysis | Standard enterprise research report |
| Competitors or alternative vendors provided | Adds horizontal comparison, substitutability, and procurement recommendation | Shortlist / vendor comparison |
| Risk or compliance concerns provided | Prioritizes litigation, penalties, certifications, data security, sanctions, and export controls | Risk memo and follow-up diligence questions |
What The User Should Provide
Minimum input:
| Field | Example |
|---|---|
| Research subject | Company, supplier, product, brand, group, subsidiary, person, or category concept |
| Decision context | Procurement onboarding, renewal, competitive analysis, customer research, partnership review, investment screening, or risk alert |
Recommended additions:
| Field | Example |
|---|---|
| Known links | Official website, product page, hiring page, press release, registry or regulatory link |
| Geography | China, U.S., EU, Southeast Asia, global |
| Focus questions | Data security, delivery capacity, product maturity, pricing, litigation, penalties, financial stability |
| Comparison targets | Known competitors, backup vendors, internal alternatives |
| Output requirements | Quick memo, standard report, full diligence; Markdown, PDF, slides, spreadsheet, bilingual |
Incomplete input is fine. The agent labels reasonable assumptions and unverified items instead of turning guesses into facts.
What The User Gets
A standard report usually includes:
- Decision summary: one-line conclusion, recommended action, top risks, strongest evidence.
- Identity and scope verification: legal entity, brand, product, parent/subsidiary relationships, and exclusion of similarly named entities.
- Lifecycle narrative: founding/launch background, milestones, strategy shifts, financing, customers, and organizational changes.
- Current state: operating status, organization signals, funding/revenue/customer signals, hiring and expansion signals.
- Product and capability: core functions, technology, deployment, pricing/contract signals, maturity.
- Customer and market proof: customer cases, industries, buyers, renewal/replacement/user-feedback signals.
- Horizontal comparison: competitors, substitute vendors, upstream/downstream dependencies, capability/price/delivery/risk matrix.
- Risk and compliance: litigation, penalties, reputation issues, sanctions, export controls, certifications, privacy, and security posture.
- Scenario judgment: most likely, most dangerous, and most optimistic scenarios.
- Evidence ledger: visited sources, key facts, conflicting claims, unreachable sources, next diligence questions.
How It Differs From Search Or A Company Profile
| Method | Focus | Main output | Best for |
|---|---|---|---|
| Web search | Finding information | Links, snippets, fragmented facts | Quickly learning a name |
| Company profile generation | Writing a description | Company story and product description | Marketing, sales, or background material |
| Enterprise Company Research Agent | Enterprise decision-making | Evidence ledger, risk judgment, peer comparison, procurement/partnership recommendation | Supplier onboarding, competitive analysis, partnership review, investment screening |
The Skill is most useful when “there is a lot of information” needs to become “which facts are reliable, which conclusions are usable, and where we cannot judge yet.”
Common Scenarios
| Scenario | Why this Skill fits | Main artifacts |
|---|---|---|
| New supplier onboarding | Enterprises need legal entity, delivery capacity, customer proof, and risk checks | Supplier report, risk list, follow-up diligence questions |
| Competitor or substitute shortlist | The decision depends on maturity, customers, pricing, and replacement risk, not only features | Comparison table, procurement recommendation, substitutability judgment |
| Customer or partner research | Sales and BD teams need company state, product context, and organization signals | Customer profile, business signals, partnership entry points |
| Investment screening | Teams need to know whether the company is active and whether the product is mature | Screening memo, evidence grades, risk notes |
| Confusing brand/subsidiary structure | Similar names, product lines, and overseas entities are easy to mix up | Identity verification, scope notes, conflicting sources |
| Negative news or compliance alert | Single databases can mislead; multiple sources need reconciliation | Risk memo, conflict table, next checks |
Default Capabilities
- Current verification: For company status, funding, legal, product capability, customer cases, pricing, hiring, and certifications, browse and verify.
- Evidence priority: Prefer regulatory, registry, SEC, and exchange filings; then official pages, announcements, docs, executive statements, customer cases, hiring, patents, repos, reputable media, and third-party databases.
- Fact/judgment separation: Every key judgment must trace back to facts. Inference must be labeled.
- Scope control: Do not substitute parent-company data for product data or confuse similarly named entities.
- Procurement lens: The report must answer whether the subject is safe to buy from, partner with, or depend on.
- Conflict handling: When sources conflict, document the conflict in the evidence ledger instead of silently choosing one version.
Repository Structure
enterprise-company-research-agent/
├── SKILL.md
├── README.md
├── LICENSE
├── skill.yml
├── agents/
│ └── openai.yaml
└── references/
└── enterprise-research-playbook.md
SKILL.md is a thin entry point: triggers, flow, depth control, and hard rules. references/enterprise-research-playbook.md is the deep reference: evidence grades, search entry points and query craft, research modules, low-evidence protocol, report template, and quality gates, loaded on demand.
Installation
Codex
git clone https://github.com/flicy/enterprise-company-research-agent.git
cp -R enterprise-company-research-agent ~/.codex/skills/enterprise-company-research-agent
Claude Code
git clone https://github.com/flicy/enterprise-company-research-agent.git
mkdir -p ~/.claude/skills
cp -R enterprise-company-research-agent ~/.claude/skills/enterprise-company-research-agent
QClaw / OpenClaw
git clone https://github.com/flicy/enterprise-company-research-agent.git
# Copy this directory into your QClaw / OpenClaw skills search path. Keep SKILL.md as the entry point.
CodeBuddy / WorkBuddy-related coding agents
git clone https://github.com/flicy/enterprise-company-research-agent.git
mkdir -p .codebuddy/skills
cp -R enterprise-company-research-agent .codebuddy/skills/enterprise-company-research-agent
skill.yml is a WorkBuddy-style metadata bridge. The core entry point remains SKILL.md.
Quick Start
In an Agent Skills compatible client, open a new session and say:
Use enterprise-company-research-agent to research [company/product name] for [procurement/competition/partnership/investment/risk].
A fuller prompt:
Use enterprise-company-research-agent to research [company name].
Decision context: supplier onboarding.
Focus: legal entity, product maturity, customer cases, data security, litigation, penalties, delivery capacity.
Compare against: [competitor A], [competitor B].
Output: standard English report with evidence ledger and follow-up diligence questions.
Quality And Safety
- This Skill needs web access for “latest status” research. Without network access, current facts must be labeled unverified.
- Do not treat third-party databases as primary facts. Critical numbers and conclusions should be traced to original sources when possible.
- Do not commit customer data, internal review material, access tokens, private paths, or account information to a public repository.
- The README stays bilingual.
SKILL.mdandreferences/are the agent instruction layer and stay Chinese-only to control context cost; report output language follows the user, and bilingual reports are produced at the report layer.
Related Skills
| Skill | Purpose |
|---|---|
| wechat-publish-from-transcript | Turn interview/livestream transcripts into WeChat articles and cross-platform content packs |
| prepare-livestream-event | Prepare livestream event launches with posters, copy, and publishing checklists |
| publish-skill-to-github | Publish local Skill packages to GitHub |
License
MIT — see LICENSE