Install
$ agentstack add skill-haochenhust-ch-skills-company-analyzer ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Company Analyzer
Produces a company analysis organized as Story × Logic × Judgment — a deliberate departure from SWOT-style templated reports. The three layers map to how humans actually understand a business: the narrative arc makes it memorable, the economic logic makes it rigorous, and the forced judgment makes it useful.
The benchmark for style and depth is 《晚点 LatePost》-class industry writing (see references/style-benchmark.md for an annotated example). If the output reads like a translated McKinsey slide with English jargon sprinkled in, it failed — even if all the facts are correct.
When to use
- "分析一下 XX 公司" / "深度研究 XX" / "我想搞懂 XX 这家公司"
- Weighing an investment, job, or partnership with a specific company
- Wants to understand the company as a thesis, not a data dump
When NOT to use
- Quick factual lookup (founding year, headcount) — answer directly
- Real-time prices, intraday earnings reactions — route to a market data tool
- Comparing ≥5 companies at once — depth-first, not breadth-first
- Pure concepts/technologies (Transformer, RAG) — not a company
Writing style (非协商)
This is the hardest part to get right and the easiest to fail. Before writing any section, hold the following rules in memory.
- 母语化中文,不要自造中英夹杂术语。"护城河"说护城河,不要说 "moat";"资本配置"说资本配置,不要说 "capital allocation";"工艺 know-how" 最多保留 "know-how" 这样已被中文业界吸收的词。行业通用英文缩写(CAGR、PE、TIER 1、ROE、FCF)可保留;自造英译概念(process power、moat type、unit economics 当小标题用)不要。
- 小标题必须是带动词的一句完整论断,不是名词短语。
- 好:「过往的路径依赖,让公司错失了新能源车的增长红利」
- 差:「历史沿革」「业务模式」「护城河分析」
- 每段话都要自我解释因果。不写"它有规模优势",写"它在 X 细分市场占 Y% 份额,这让它能摊薄 Z 这条固定成本——但同一份额在 W 市场并不成立"。
- 因果链至少 3 层深。事件 → 为什么发生 → 业务影响 → 投资含义。停在任何中间层都是懒惰。
- 段间用中文连接词过渡,不要靠 markdown 视觉结构硬拼(无 emoji、无表格作为论证主体——表格只用于展示分段数字)。
- 数字要带解释。不要"营收 147 亿元,同比 +4.83%",要"营收 147 亿元,同比只 +4.83%,而扣非反而 -4.39%——意味着这 4.83% 增速里一多半是卖资产,不是主业。"
Workflow
Phase 0 — Scope alignment (可跳过)
Default: use AskUserQuestion to confirm three things before any web search:
- Target disambiguation — "Cursor" = 编辑器 (Anysphere) 还是概念?
- Purpose — 投资 / 求职 / 竞争情报 / 好奇. 决定哪一层权重最重.
- Depth — quick / standard / deep (见下方 length quota).
跳过条件: 用户在请求里已经明确指定了公司 + 目的 + 深度(或者深度从上下文清楚),直接进入 Phase 1,不要为了"按流程"打断用户. 但如果任一维度歧义(公司名有两个候选 / 没说目的 / 没说深度),必须问.
Phase 1 — Scenario branching
按公司生命周期选 ONE. 下列权重是叙事重心,不是字数机械切分.
- A — Early-stage (<3y): Story 60% / Logic 25% / Judgment 15%. 创始人 DNA 主导;护城河是假设.
- B — Growth (3–10y): Story 40% / Logic 40% / Judgment 20%. 商业模式走完第一个周期.
- C — Mature (10y+, private): Story 30% / Logic 45% / Judgment 25%. 数字和结构位置主导.
- D — Public company: 同 C,额外叠加财务层 — 见
references/financial-analysis.md.
Phase 2 — Three-layer parallel execution
派 3 个子 agent 并行. 每个 agent 负责一层,收到 prompt 后要做的第一件事是独立核实父 agent 传过来的初始事实——创始人履历、年份、关键数字这些,如果和公开资料冲突,以独立核实为准,不要照搬父 agent 的叙述。父 agent 也会出错。
- Sub-agent 1 — Story layer. Read
references/story-patterns.md. - 产出:创始人的来路 + 关键决策(不是流水账). 为 Mature / Public 公司,Story 可以压缩为"三到四个决定公司命运的分叉";不需要给每家公司做电影式开场.
- 对早期公司 (A/B) 可以有场景化细节;对 Mature / Public 公司以战略事件为主(并购、合资股权比例、高管变动、重大监管事件、关键客户签入或丢失).
- Sub-agent 2 — Logic layer. Read
references/business-logic.md、references/moat-types.md、和references/industry-structure.md. - 必须先做"行业结构分析"再讲公司. 行业层次是什么(细分应用市场各占多少、集中度如何、技术壁垒高低). 公司在这张地图上的坐标是什么. 如果上来就写"公司业务模式如何……",是失败.
- 产出:行业结构 → 公司定位 → 收入/成本机制 → 护城河 → 配对比较(强制).
- 配对比较是必做项,不是可选. 选 1 个最能说明问题的对手做深度对比(该对手做对了什么、该公司做错了什么,或反之). 如果找不到合适对手,要命名理由(为什么这是一个孤立市场),不能跳过.
- Sub-agent 3 — Judgment layer. Read
references/anti-patterns.mdfirst. 在 Story + Logic 落地后运行. - 产出:公司现在在生命曲线的哪里、未来 3–5 年的赌注(2–3 个可证伪的杠杆)、命名立场("我会"或"我不会"或"我会等+触发条件"——骑墙算失败)、5 个监控指标.
Phase 3 — Synthesis
主 agent 把三层编织成一篇文章,不是三段拼接. 语言风格参照 references/style-benchmark.md 的《晚点》示例:平白中文、整句论断作标题、段间用"但问题在于"/"对比而言"/"所以"做过渡. 不保留英文术语标签作为 section header(不要出现 "### Moat Analysis" 这种东西).
Length quota
量化配额(中文字符数),低于下限视为研究深度不够,高于上限视为灌水.
- Quick: 2000–3500 字
- Standard: 7000–10000 字
- Deep: 15000–22000 字
(注:这是 synthesis 后最终产出字数,不是三层草稿相加;单层草稿会在合成时被压缩和穿插.)
Anti-patterns (必避)
详见 references/anti-patterns.md. 短清单:
- 流水账年表 — "2015 年 A,2016 年 B……"
- 咨询八股 — 赋能 / 抓手 / 闭环 / 打造 / 生态位;英文 synergize / leverage / holistic / empower.
- 自造中英夹杂术语 — "moat"、"process power"、"capital allocation"、"unit economics"、"go-to-market"、"stickiness" 作为中文段落里的概念词出现.
- 模板骨架 — SWOT、波特五力、PEST 作为 section header.
- 财报文字版 — 把 10-K 数字粘贴成段落而不解读.
- 骑墙立场 — "既有机遇也有挑战".
- 综上所述 / In conclusion 总结.
- 名词性标题 — 章节标题是"业务模式"而不是"业务模式决定它只能做中等生意"这种带论断的句子.
Progressive disclosure (参考文件)
references/style-benchmark.md— 《晚点 LatePost》风格标注示例(v2 新增)references/industry-structure.md— 如何做行业结构分析(v2 新增)references/story-patterns.md— 五种叙事形态(成熟公司可简化使用)references/business-logic.md— 收入机制 + 商业性质 + 管理层分析references/moat-types.md— 7 类护城河 + 监管/IP/资本(用作思考工具,输出时用中文自然语言)references/financial-analysis.md— 上市公司财务 4 步读法references/anti-patterns.md— 详细负面清单 + 重写对照templates/report-template.md— 建议骨架(标题按一句话论断写)
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: Haochenhust
- Source: Haochenhust/ch-skills
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.