How to use the Investment Thesis Tracker agent skill

What the thesis-tracker skill makes the agent do

When you ask the agent about a company, the skill first builds an investment thesis: a short statement of why you bought, a list of testable assumptions, and red-line events that should trigger a sale. The thesis is saved as a file so the reasons stay visible.

On later check-ins, the agent gathers fresh financial news, tests each assumption against the new information, scores the thesis's health out of 10, and appends a dated entry to the log. The source describes it as a discipline tool that keeps the reasons for an investment visible, not a source of buy or sell advice.

• Creates a thesis: five one-sentence answers covering the business, moat, management, margin of safety and downside.
• Lists 3 to 7 testable assumptions, each with a verification method and frequency.
• Lists red-line conditions, each with a severity and a required action.
• Re-checks assumptions and red lines on later runs, then scores thesis health.
• Saves and appends records to reports/{company}-thesis.md.

Install the thesis-tracker skill with a ZIP file

Download the ZIP for this skill from this page and unzip it. The skill is a folder containing SKILL.md. Place that folder in one of these locations, depending on how you use the agent.

For Claude Code, use your personal skills folder ~/.claude/skills/ so the skill is available across projects, or a project's .claude/skills/ folder so it applies only to that project.

In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

• Personal (all projects): ~/.claude/skills/thesis-tracker/
• Single project: .claude/skills/thesis-tracker/ inside your project folder
• Claude apps: Settings → Capabilities → Skills, then upload

Set up the repository and required tools

The skill refers to shared tools and a rules file in the AI Berkshire repository, so you need a checkout of that repository. The source says the Codex adaptation prefers running commands from the repository root and locating the actual checkout path rather than assuming a fixed folder.

The workflow calls tools/financial_rigor.py for valuation checks, which requires Python 3. It also needs local write access to a reports/ folder, where thesis files are saved. The agent needs web search to collect share prices, valuation and earnings data.

• AI Berkshire repository checkout (the skill refers to tools/ and AGENTS.md)
• Python 3 and tools/financial_rigor.py
• Web search available to the agent
• Write access to a reports/ folder

Trigger the skill with natural requests

You do not need special syntax. Name the company and say what you want. The skill's input forms are a company name for first use or follow-up checks, a company name followed by 'build thesis' to force a new thesis, and a company name followed by 'quarterly check' to review against the latest earnings report.

The skill checks whether reports/{company}-thesis.md already exists. If it does not, the agent builds a new thesis. If it does, the agent runs a tracking check. If you say you already have a thesis but the agent cannot find the file, it will ask for the file path.

• "Use thesis-tracker to build an investment thesis for [company] — I bought at [price]."
• "Run a quarterly thesis check on [company] using its latest earnings."
• "Rebuild the thesis for [company] from scratch."
• "Check whether my [company] thesis is still intact."

What a tracking check looks like

A tracking check loads the saved thesis, collects the latest reports, news, share price and insider trading records, then tests each assumption. Each assumption is marked as holding, weakening at the margin, damaged, or broken. Each red line is marked as triggered or not triggered.

The report contains a health score from 1 to 10, computed from these statuses and red-line triggers. The source gives a formula that subtracts points for broken, damaged and weakening assumptions and for each triggered red line. The score is then mapped to a suggested action such as continue holding, reduce, or sell.

The skill also asks the report to state whether the thesis is still intact, what action to take, and when to check next. Each check is appended to the log table in the thesis file.

• Health score 9-10: all assumptions hold; consider adding.
• Health score 7-8: core assumptions hold; keep holding.
• Health score 5-6: one or two assumptions damaged; hold with more caution.
• Health score 3-4: several assumptions damaged; consider reducing.
• Health score 1-2: red line triggered or core assumption broken; strongly consider selling.

Safety notes: scripts, network and what to verify

This skill runs scripts and uses the network. The workflow runs tools/financial_rigor.py and uses web search to fetch prices and financials. Review the scripts in the repository before running them, and make sure you are comfortable with the agent reading and writing files in your reports/ folder.

The data is web-sourced and can be wrong or stale. The source says scores depend on web search results the user should verify, and that the skill does not replace professional advice. Cross-check prices, valuation and earnings figures against primary sources before you act on them.

• Check the price and earnings figures against the company's filings or your broker before acting.
• The skill tells the agent to state the data cutoff date and label uncertainty and source gaps.
• Keep the thesis file as your record; the score is a checklist result, not a trade signal.

Tips for getting useful theses

Write the thesis before you buy, when you are calm. The source's key principle is to set sell conditions before the purchase, because decisions made under panic are usually worse. Make each assumption specific enough to check. 'The company is good' is not testable; a stated threshold for a measurable figure is.

When a red line triggers, the source says to act rather than wait. It also distinguishes a thesis break from a price drop: a 30% fall does not automatically mean sell, but a broken thesis does.

• Keep assumptions to roughly 3 to 7; too many means the thesis is unfocused.
• Run a quarterly check after each earnings report.
• Admit mistakes in the thesis file rather than defending it.