AI Berkshire is a collection of Claude Code and Codex skills that structures investment research around the methods of four value-investing thinkers and uses multiple agents for adversarial analysis. It is intended for investors who want a disciplined process for researching companies and making valuation-based decisions.
Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/xbtlin/ai-berkshirenpx agentmods add skills/xbtlin/ai-berkshire/thesis-driftWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/xbtlin/ai-berkshire/thesis-drift)<a href="https://agentmods.dev/skills/xbtlin/ai-berkshire/thesis-drift"><img src="https://agentmods.dev/badge/skills/xbtlin/ai-berkshire/thesis-drift/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/xbtlin/ai-berkshire/thesis-drift"><img src="https://agentmods.dev/badge/skills/xbtlin/ai-berkshire/thesis-drift.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- Snyk pass
- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00034 | $0.03007 |
| Opus 5 | $0.00017 | $0.01503 |
| Sonnet 5 | $0.00007 | $0.00601 |
| Haiku 4.5 | $0.00003 | $0.00301 |
Grade A, and why
thesis-drift scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 226 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Codex adapter note
This skill is generated from skills/thesis-drift.md so Claude Code and Codex users share one canonical workflow.
- Treat
$ARGUMENTSas the user's request in the current Codex thread. - When the source mentions Claude-only surfaces such as Task, Agent, WebSearch, Bash, Read, or Write, use the closest Codex capability available in this session: subagents when available, web search when needed, shell commands for local tools, and normal file edits for workspace files.
- Use shared project tools from
tools/in this repository. Prefer running commands from the repository root with paths likepython3 tools/financial_rigor.py ...; if the current thread starts outside the repo, locate the actual checkout path first instead of assuming a fixed home-directory path. - Before starting research, run the
datecommand to confirm today's date; treat it as the baseline for "latest" data and state the data cutoff date in the report header. Never assume the current date from training data. - Preserve the research quality rules from
AGENTS.md: cross-check financial data, use exact arithmetic tools for valuation/math, and clearly label uncertainty and source gaps.
投资论文漂移检测：分清事实变化与措辞变化
对 $ARGUMENTS 执行投资论文漂移检测。
支持输入格式：
公司名 旧报告路径 新报告路径— 指定两份研究报告或论文快照进行对比公司名 reports/{公司名}-thesis-旧日期.md reports/{公司名}-thesis-新日期.md— 对比两份带日期的论文快照公司名— 自动查找reports/{公司名}-thesis.md及同目录历史快照；如果没有基线则转入缺失基线处理
"当事实改变时，我就改变想法。你呢？" —— 凯恩斯
"股价波动不是论文漂移，事实变了才是。" —— AI Berkshire
设计理念
长期持仓最难的不是每天读新闻，而是区分三件事：
- 事实改变：收入、利润率、竞争格局、管理层行为、资本配置发生可验证变化
- 价格改变：市场情绪或估值倍数变化，但生意本身未变
- 措辞改变：两份报告表达不同，但底层证据和判断没有变化
投资论文漂移检测的目标是：只在证据变化时承认论文变化。不能因为报告换了写法就制造漂移，也不能因为股价涨跌就误判基本面。
本 Skill 依赖 /thesis-tracker 输出的结构化维度：核心假设清单、红线清单、估值锚点、追踪记录表。没有这些结构时，先补齐基线，再做漂移检测。
执行流程
第一步：判断操作模式
解析 $ARGUMENTS：
- 如果提供两份报告路径 → 进入指定报告对比模式
- 如果只提供公司名 → 查找
reports/{公司名}-thesis.md及历史快照，进入自动快照对比模式 - 如果只找到一份报告或没有历史基线 → 进入缺失基线处理模式
- 如果两份报告不是同一家公司 → 停止并要求用户确认，不做跨公司漂移判断
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 12d ago First seen · 226 lines · 34 tokens per session scan A 28363374d4b0
thesis-drift is a skill published in the GitHub repository xbtlin/ai-berkshire (16,273 stars, last pushed 3d ago), licensed MIT. It adds 34 tokens to every session and 3,007 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
dcf-model
Build discounted cash flow valuation workbooks in Excel.
comps-analysis
Build comparable-company valuation workbooks in Excel.
hyperliquid
Hyperliquid market data, account history, trade review.
stocks
Stock quotes, history, search, compare, crypto via Yahoo.
omni-budget
Configure spending limits, token quotas, and rate-limit policies per API key or globally. Inspect current consumption and enforce cost controls across providers.
polymarket
Query Polymarket: markets, prices, orderbooks, history.