Deep management assessment — the 'buying a stock is buying a person' layer. For a company or a named executive, evaluate integrity (promise-vs-delivery tracking
复制下面这句话，粘贴给 Claude Code、Codex、Cursor 等 AI 编程工具，它会读取安装说明并在你确认后完成安装。
请阅读 https://ai.atlankj.com/install/asset/gh-management-deep-dive-f51c7353afb9 ，按照其中的说明把「management-deep-dive」安装到你（当前 AI 工具）中。执行前先告诉我将运行的命令和写入的位置，等我确认。
查看 AI 将读取的安装说明正在读取 GitHub 原文…
内容来自 GitHub 原始文件，由原作者维护。在 GitHub 查看
Deep management research on a company (or person company).
"Buying a stock is buying a person. Find someone you trust, then hold long-term." — Duan Yongping
"To judge management, look at what they do when no one is watching." — Buffett
Most analysis stops at the surface: résumé, shareholding, compensation. But Buffett spends time talking with management, Li Lu says investing is fundamentally investing in people, Duan Yongping says buying a stock is buying a person.
AI can't have dinner with management, but it can via public information: ① track whether words match deeds (promises vs delivery); ② analyze the return of every major capital-allocation decision; ③ infer character from decisions made in hard times; ④ cross-check via employee / merchant / customer feedback.
Use when: standard research leaves management score uncertain (★★★ or below), or when management is the core investment logic.
Use web_search / get_stock_profile / get_sec_filings:
| Role | Name | Tenure | Background | Shareholding/options |
|---|---|---|---|---|
| CEO/Chairman | ||||
| CFO | ||||
| Founder (if not in role) | ||||
| Actual controller (if different from CEO) | ||||
| Other key executives |
Distinguish "who decides" from "whose name is on the title" — some founders remain the soul even after stepping down.
Then gather four data categories (sequentially via web_search + get_stock_news + get_financial_statements + get_sec_filings, or run_swarm): ① CEO public statements and prediction record; ② capital-allocation decisions (M&A / buybacks / dividends / new business); ③ governance and compensation; ④ side validation (employee / customer / industry reputation).
Use web_search to collect the CEO's public statements over the past 5 years (shareholder letters, calls, interviews, social media); extract their judgments:
| Time | CEO's judgment/prediction | Actual outcome | Accuracy |
|---|---|---|---|
| "We believe market X will..." | X actually... | ✅/❌ |
Key questions: Has the CEO made correct calls ahead of the market? Stayed calm when everyone was bullish? Is their industry understanding independent or following the crowd?
| Dimension | Assessment | Evidence |
|---|---|---|
| Strategy-to-execution | Did they do what they said? | |
| Organization | Can they attract and retain talent? | |
| Crisis handling | How do they respond to difficulty? | |
| Iteration speed | How fast do they correct mistakes? |
Buffett: "We look for three qualities: integrity, intelligence, and energy. Without the first, the other two will kill you."
Extract specific commitments from the past 3 years (calls, letters, interviews) via get_sec_filings / web_search / get_stock_news:
| # | Time | Commitment | Occasion | Delivery | Rating |
|---|---|---|---|---|---|
| 1 | "X business profitable in 2025" | 2024 annual call | ✅/⚠️/❌ |
| Delivery rate | Rating |
|---|---|
| >80% | Excellent — does what it says |
| 60-80% | Acceptable — right direction, execution gaps |
| 40-60% | Concerning — over-promises, under-delivers |
| <40% | Serious problem — untrustworthy |
Search for major crises (price crash, earnings miss, regulatory shock, competition); analyze management's response: proactive comms or hiding? Internal attribution or blaming external? Used the crisis to do the hard right thing or to pander short-term?
| Stakeholder | Attitude | Evidence | Rating |
|---|---|---|---|
| Shareholders | Respect/ignore/exploit | ||
| Employees | Treat well/exploit/ignore | ||
| Customers/users | Customer-first/short-term extract | ||
| Merchants/suppliers | Fair/extreme price pressure | ||
| Regulator/society | Compliant/corner-cutting |
Buffett's most-valued ability — for every dollar earned, how much does management turn it into?
Search past 5 years of capital decisions (web_search / get_financial_statements); score each:
M&A record: time / target / amount / strategic logic / after-the-fact return / score(1-5)
Buybacks (use financial_rigor command=verify_valuation to check PE at buyback time vs now): time / amount / avg price / PE then / retrospective / score(1-5)
Dividends: year / amount / payout ratio / vs FCF / sustainable
New business investment: time / area / cumulative spend / status / return / score(1-5)
| Dimension | Score(1-5) | Notes |
|---|---|---|
| M&A discipline | Right price? Integration? | |
| Buyback timing | Buy low, stop high? | |
| Dividend rationality | Payout vs FCF? | |
| New business | Success rate? Stop-loss discipline? | |
| Cash management | Reasonable reserves? Hoarding? | |
| Overall |
| Item | Detail | Risk |
|---|---|---|
| Dual-class / super-voting? | ||
| Founder/controller stake? | ||
| VIE structure? | ||
| Independent directors truly independent? | ||
| Recent major-holder changes? |
| Executive | Total annual comp | % of net income | vs peers | Reasonable? |
|---|
Is incentive aligned with long-term shareholders or encouraging short-term behavior?
| Related party | Transaction | Amount | Arm's length? | Risk |
|---|
AI can't meet face-to-face, but cross-check via public channels:
web_search for Glassdoor ratings, discussions (mark platforms requiring login as "user can supplement") — culture, management rating, intensity, compensation, growth.Buffett: "A great company should be one that a fool can run — because eventually one will."
Structure: ① Key people overview; ② Integrity (delivery rate + hard times + stakeholders); ③ Ability (foresight + execution + capital allocation); ④ Governance; ⑤ Side validation; ⑥ Overall score and verdict.
| Dimension | Weight | Score(1-5) | Weighted |
|---|---|---|---|
| Integrity | 35% | ||
| Strategy & execution | 25% | ||
| Capital allocation | 25% | ||
| Governance | 15% | ||
| Total | 100% |
All three "yes" = ★★★★★; first two "yes" = ★★★★; only the first "yes" = ★★★; first is "no" = ★ (don't invest).
Use write_file to reports/{company}-management-{YYYYMMDD}.md. Then report_audit (extract → verify → verdict) as a quality gate on the numbers.