A Claude skill that performs rigorous, mathematically consistent value investing analysis on any publicly traded stock.
Built to fix the most common failure modes of AI-generated stock analysis: reverse-engineered conclusions, inflated DCF valuations, and "margin of safety" that's really just a 10% discount.
Given a stock ticker, this skill runs a complete value investing workflow:
- Data Freshness Validation — Enforces use of the latest annual report. If the most recent filing isn't available, it flags the gap explicitly and estimates the valuation error.
- Moat Analysis — Identifies and rates moat sources (network effects, switching costs, intangibles, scale) with specific evidence, not just label-dropping.
- Assumption Disclosure Table — Every DCF input is listed with its source and justification before the calculation begins. No hidden assumptions.
- DCF Valuation — 10-year two-stage model, cross-validated with Gordon Growth Model. If the two methods diverge by more than 30%, the calculation is flagged for review.
- Scenario Analysis — Bear / base / bull cases with probability weighting when material risks (e.g. regulatory, geopolitical) are present.
- Honest Margin of Safety — Buy price = intrinsic value × 0.70. If the current price doesn't qualify, the skill says so directly instead of lowering the threshold to manufacture a buy signal.
- Buy / Hold / Sell Rules — Clear, trigger-based rules tied to fundamentals, not price action.
| Rule | Why It Matters |
|---|---|
| Must use latest annual report | Using stale FCF data can skew intrinsic value by 50%+ |
| DCF must be mathematically self-consistent | Many AI analyses state inputs and conclusions that don't actually compute |
| Discount rate must include regional/regulatory risk premium | Chinese and HK-listed stocks carry risks not captured by standard WACC |
| Margin of safety is a hard threshold (30%), not decorative | A "30% discount" that's actually 8-12% provides no real downside protection |
| No normalized FCF without a specific filed basis | Upward "normalization" is the most common way to inflate a valuation |
- Download
value-investing-evaluation.skill - Double-click to install in Claude desktop app
- Trigger by saying things like:
- "Analyze Tencent using value investing"
- "Is NVIDIA undervalued? DCF analysis"
- "价值投资分析腾讯"
- "帮我算一下药明康德的内在价值"
━━━━━━━━━━━━━━━━━━━━━━━━━━
Data Freshness Check
━━━━━━━━━━━━━━━━━━━━━━━━━━
Analysis date: 2026-03-25
Latest available annual report: FY2025 (published 2026-03-19)
Report used: FY2025 ✅
One-line verdict: Wide moat | Fairly valued | No buy signal at current price
[Assumption & Estimate Disclosure Table]
[Moat Rating Table]
[Key Financials with Sources]
[DCF Calculation Table]
[Scenario Analysis]
[Margin of Safety Conclusion]
[Buy / Hold / Sell Rules]
This skill was built after reviewing how a leading AI model analyzed Tencent's stock and found three critical errors in the same analysis:
- FCF understated by ~50% (used 100B HKD vs actual ~200B HKD)
- DCF inputs and conclusions were mathematically inconsistent (stated inputs implied ~221 HKD/share; conclusion was 550-600 HKD)
- "Graham 30% rule" was cited but the suggested buy price was only 8-13% below the stated intrinsic value
This skill is designed to make those errors impossible by construction.
- Relies on web search for financial data — accuracy depends on search results
- Does not replace professional financial advice
- Qualitative moat judgments remain subjective even with a structured framework
- Works best for large-cap stocks with multiple years of public financials
MIT