Buffett Investment Thinking System
A Claude Code skill that turns 70 years of Warren Buffett's investing wisdom into structured buy/hold/sell analysis, from an 8-question quick screen to deep-dive company evaluations.
Pure prompt-based knowledge skill: no permissions requested, no scripts executed, no external side effects or data exfiltration; data flow is transparent and least-privilege holds. Deducted for: Buffett quotes and figures (e.g., 1965–2022 returns) carry no verifiable sources, so attribution is incomplete.
Instructions are self-consistent; task-type-to-reading-path mapping is clear; on-demand reading reduces context-overload risk; explicit stop-and-explain requirement outside circle of competence. Deducted for: static review cannot execute/validate, no tests or failure-feedback design for abnormal input, and the read-references protocol depends on model compliance.
Scenarios (stock analysis, report interpretation, valuation, sell decisions) are concrete; fixed output format includes explicit circle-of-competence boundaries. Deducted for: aggressive trigger description ('proactively trigger even if Buffett is not mentioned') creating false-trigger risk; no evidence of Chinese-language fit; non-fit boundaries (short-term trading, derivatives pricing) undeclared.
Good layered architecture (SKILL.md main flow plus 8 topically organized reference files) with an index table and reading protocol. Deducted for: unknown license metadata, no versioning or changelog, no disclosed maintainer/update path, publisher unverified.
Structured framework with fixed output template offers clear marginal value over generic chat (8-question filter, owner-earnings formula, four sell criteria are directly usable). Deducted for: static review cannot verify actual output quality; valuation numbers (multiple ranges, discount tiers) are rule-of-thumb values requiring user verification.
Key claims cite Buffett shareholder-letter years (1983, 1991, 1993, etc.), giving some traceable form. Deducted for: self-references within files only, no third-party corroboration, no tests or independent reproduction material, and quote accuracy cannot be verified in a static read; fact/inference boundaries partly rest on the author's curation.
- Trigger scope is overly broad: any investment-related topic forcibly activates the skill, which may misfit non-value-investing queries (short-term trading, options pricing).
- Concrete figures (valuation multiples, margin-of-safety discounts) are rule-of-thumb values without cited sources and should not be used directly as investment bases.
- Buffett quotes and performance figures are not independently verified; paraphrasing distortion is possible.
- License and version metadata are missing; confirm compliance and update responsibility before enterprise adoption.
- Output is an analytical framework, not investment advice; financial data must be supplied and verified by the user.
What does this skill do, and when should you use it?
This is a Claude Code skill that packages Warren Buffett's investing framework — margin of safety, moats, circle of competence, Mr. Market, capital allocation — into a reusable analysis system. It consists of one SKILL.md plus eight on-demand reference files covering thinking frameworks, investment philosophy, business moats, management assessment, financial metrics, valuation and capital allocation, risk behavior, and industry playbooks. It offers three dispatch paths: a 2-minute 8-question quick screen, a full company deep analysis, and direct lookup for specific questions. All output follows a mandatory template including conclusion, circle-of-competence verdict, key assumptions, valuation range, and item-by-item checks of four sell criteria. According to the repo's own eval, it lifts pass rate from 66.7% to 100% across three test cases, at a clear token and time cost.
Once installed, the skill auto-triggers on stock analysis, financial report interpretation, moat assessment, buy/hold/sell decisions, or value-investing concepts. Path A runs an 8-question filter (circle of competence, durability, moat, pricing power, earnings quality, debt safety, management integrity, price); two "No" answers need strong justification, four mean pass. Path B reads references/03-business-moat.md, 04-management-governance.md, 05-financial-metrics.md, and 06-valuation-capital.md in order, adding 07-risk-behavior.md and the relevant 08-industry-playbooks.md chapter as needed. Analysis requires computing owner earnings (net income + D&A − maintenance capex − working capital increase), checking 10-year average ROIC (target >15%) and cash conversion (target >90%), and assigning margin-of-safety discounts of 20–50% by certainty tier. Management integrity is an automatic veto. Output is a fixed structured template ending with a verdict delivered in Buffett's voice.
- A retail investor wondering whether a stock deserves deeper research uses the 8-question screen to filter out weak candidates fast
- A shareholder deciding whether to sell checks the four sell criteria (severe overvaluation / moat destruction / management integrity / better opportunity) one by one
- After reading an annual report or shareholder letter, an analyst assesses management integrity, capital allocation record, and institutional imperative
- An analyst doing Buffett-style work in a specific sector — insurance, banking, consumer, media, energy, railroads, tech — jumps straight to the relevant industry playbook
- A value-investing learner applies concepts like intrinsic value, compounding, and concentration to real cases
- Anyone discussing buybacks, dividends, or capital allocation who wants the analysis framed in Buffett's terms
What are this skill's strengths and limitations?
- Methodology is explicit and traceable: the 8-question screen, five moat types, owner-earnings formula, margin-of-safety tiers, and four sell criteria are all concretely defined, not vague prompting
- Sensible progressive disclosure: SKILL.md always loads, eight reference files load on demand by task type, controlling context spend
- Mandatory output template forces sections that models tend to skip, like circle-of-competence verdict and key assumptions for later verification
- Has quantitative eval: three benchmark cases show pass rate rising from 66.7% to 100%, attributed mainly to systematic reference-file reading
- Management integrity is an automatic veto, reflecting an actual Buffett decision rule
- No license information; commercial-use compliance is unclear
- The eval covers only three self-reported test cases — a tiny sample with no third-party validation
- Material cost: roughly 30,000 extra tokens and 59 extra seconds per analysis on average; overkill for lightweight questions
- No live market or financial-data integration; the user must supply all valuation inputs
- The Buffett-voice verdict is stylized interpretation and should not be treated as real investment advice
- Only verified in Claude Code; other Agent Skills clients are untested
How do you install this skill?
Clone the repo and copy the skill folder into your project's .claude/skills/ directory:
git clone https://github.com/agi-now/buffett-skills /tmp/buffett-skills
cp -r /tmp/buffett-skills/skills/buffett your-project/.claude/skills/buffett
The final layout must be .claude/skills/buffett/SKILL.md plus a references/ directory containing the eight reference files (01–08). Claude Code auto-discovers skills at that path; no registration needed. The repo documents no other installation method.
How do you use this skill?
Just ask in Claude Code — you don't need to say "Buffett"; any investment-analysis or business-quality question triggers the skill per its description. "Analyze Apple — is it a buy at current prices?" invokes the deep-analysis path; "Should I sell this position?" forces a read of 07-risk-behavior.md and an item-by-item check of the four sell criteria. For quick filtering, ask for the 8-question checklist. Note the skill insists on actually reading reference files rather than relying on the model's built-in knowledge, so expect noticeably higher token usage and latency than a plain chat.
How does this skill compare with similar options?
Versus simply asking Claude to "analyze this stock like Buffett," the skill's difference is enforced structured reference reading (11+ tool calls vs 2 in the eval) and a strict output template, reducing improvised answers from built-in knowledge. The repo names no competing skills or products.