Guide · Agent skill
How to use the Portfolio Review agent skill (portfolio-review)
Updated · written with AI from the project's own documentation and checked against our catalogue data
Portfolio Review is an agent skill from the AI Berkshire project that reviews a whole portfolio rather than one stock. You give the agent your holdings and weights, or your share counts and buy prices, and it returns a structured report in Chinese on concentration, hidden links between holdings, opportunity cost and rough stress-test outcomes. This guide explains what an agent skill is, how to install this one, how to ask for a review, and where its limits are.
Portfolio Reviewxbtlin/ai-berkshire · MIT Download the skill Quick start
- 1Check prerequisites. Confirm Python 3 is installed and that you have a local checkout of the ai-berkshire repository with tools/ and AGENTS.md.
- 2Download and unzip. Download the skill ZIP from the directory page and unzip it to get the portfolio-review folder with SKILL.md.
- 3Install the skill. Copy the folder to ~/.claude/skills/ for all projects or to .claude/skills/ in one project, or upload it under Settings → Capabilities → Skills in the Claude apps.
- 4Start a new session. Open a fresh agent session so the newly installed skill is loaded.
- 5Send your holdings. Describe your positions with weights or with share counts and cost prices, and ask for a portfolio review.
- 6Read the report. Check the concentration, overlap, opportunity cost and stress-test sections, then verify key figures yourself.
- 7Save and revisit. Keep the saved reports/portfolio-latest.md file and run the review again at your next scheduled check.
What is an agent skill and what does this one do?
An agent skill is a folder of instructions (here, a SKILL.md file) that an AI agent such as Claude or Codex loads when your request matches the skill's purpose. The skill tells the agent which workflow to follow, which tools to run and what output to produce.
Portfolio Review takes a holdings list and runs a fixed workflow: it parses the positions, looks up current prices and valuation figures through web search, checks each position's health, analyses the portfolio as a whole, and writes a report with specific buy, trim or hold suggestions. It is written in Chinese and follows an investing style based on the ideas of Buffett and Li Lu. It is analysis support, not a personal financial adviser.
- Concentration: largest position, top-three share, number of holdings and cash share compared with suggested ranges
- Hidden links: pairs of holdings exposed to the same theme, industry or country
- Opportunity cost: whether the lowest-ranked position is expected to beat holding cash
- Stress tests: rough impact of scenarios such as a global recession or a US-China conflict
- Actions: add, trim, exit, open a new position or hold, each with a reason
What do you need before you install it?
The skill needs three things: Python 3 to run the shared tools/financial_rigor.py script, web search access for current prices, valuation and analyst estimates, and a checkout of the ai-berkshire repository so that the tools/ folder and the AGENTS.md rules are present. Without the repository, the script and research rules are missing.
The skill runs scripts and uses the network, so review what it does before you run it on a machine with sensitive files.
- Python 3 installed on your computer
- Web search available in your agent session
- A local checkout of the ai-berkshire repository, so tools/ and AGENTS.md exist
python3 tools/financial_rigor.py verify-valuationHow do you install the portfolio-review skill?
Download the ZIP for the skill from the directory page and unzip it. Then place the skill folder in one of two locations. For a personal skill available in all your projects, use ~/.claude/skills/. For a skill available only in one project, put it in that project's .claude/skills/ folder.
In the Claude apps you can instead upload the skill under Settings, then Capabilities, then Skills. Start a new session after installing so the agent picks it up.
- Personal install: ~/.claude/skills/ (available in every project)
- Project install: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills, then upload
How do you trigger the skill with a natural request?
You do not need a special command. Describe your holdings and ask for a review, and the agent should match the request to the skill. The skill accepts a list with weights, a list with share counts and cost prices, or a request to use your saved portfolio if a file already exists at reports/portfolio-latest.md.
If you use Codex, the skill's adapter note asks the agent to run the date command first and state the data cutoff date in the report header, so the figures are read against today's date.
- Weights: "Review my portfolio: Tencent 30%, Meituan 20%, Moutai 20%, Nvidia 15%, cash 15%"
- Shares and cost: "Review: Tencent 500 shares at HKD 480, Meituan 1000 shares at HKD 130"
- Saved file: "Review my portfolio" when reports/portfolio-latest.md already exists
What does a review look like step by step?
The agent first parses your holdings into a table with symbol, quantity, cost, price, market value, weight and profit or loss. It then gathers current data for each holding in parallel and tags each one with a data-confidence grade (A, B or C). Positions graded C have their conclusions marked as low confidence.
Next come a health check for each position, the portfolio-level analysis and the adjustment plan. The report ends with the overall verdict, the single most important action, the biggest current risk, and a reminder of when to review again. The agent also saves the summary to reports/portfolio-latest.md, appending adjustment records over time.
- Parse holdings into a standard table
- Collect price, valuation (PE, PB, dividend yield), recent financial changes, major events and analyst expectations
- Check each position with a short set of questions, such as whether you would still buy it at today's price
- Analyse concentration, correlation, opportunity cost and stress scenarios
- Write the report and save it to the portfolio file
python3 tools/financial_rigor.py three-scenarioWhat should you check and what are the limits?
The skill itself says that some data may be low confidence, and its results depend on web search, which can be wrong or out of date. Check the prices and valuation figures yourself before acting. The skill also states that it avoids recommending individual stocks and will point you to other research skills instead.
It is not a regulated adviser and cannot guarantee a valuation. Treat the output as a checklist that makes you ask the right questions, not as an instruction to trade.
- Verify prices and PE figures against a second source before any trade
- Expect no individual stock picks; the skill suggests other skills such as industry-research or investment-checklist
- Re-run the review on a regular schedule, the skill suggests quarterly rather than daily
What people use it for
- Checking whether three or four holdings are secretly exposed to the same Chinese internet or consumer theme
- Reviewing a Hong Kong, US or China-listed mix once a quarter with a fixed checklist
- Deciding whether a weak position ranks below cash on expected return and should be trimmed
- Running rough scenario checks on how a portfolio might fare in a recession or a US-China escalation
Frequently asked questions
Is the portfolio-review skill free to use?
The source for this listing does not state a price or licence terms beyond the project's MIT licence. Check the repository's licence file for the terms that apply to the code. Web search access and your agent subscription may have their own costs.
What do I need to install portfolio-review?
You need Python 3, web search access in your agent, and a checkout of the ai-berkshire repository so the tools/ folder and AGENTS.md are present. Then copy the skill folder into ~/.claude/skills/ or a project's .claude/skills/ folder.
Where do I put the skill folder in Claude Code?
Put it in ~/.claude/skills/ to make it available in every project, or in .claude/skills/ inside one project to limit it to that project. In the Claude apps, you can upload it under Settings, then Capabilities, then Skills.
Does portfolio-review recommend stocks to buy?
No. The skill states that it avoids recommending individual stocks and redirects you to other research skills such as industry-research or investment-checklist for screening.
Is the output accurate enough to trade on?
Not on its own. The skill relies on web search results and its own estimates, and it notes that some data may be low confidence. Verify prices and valuation figures yourself before making any decision.
How is this different from a normal stock research prompt?
It looks at the portfolio as a whole, covering concentration, hidden links between holdings, opportunity cost against cash and stress scenarios, rather than analysing one company at a time.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Portfolio Review suits individual investors with a handful of stocks who want a repeatable, checklist-style look at their mix, especially Chinese-speaking investors with Hong Kong, US or China-listed holdings. It is a poor fit for anyone who needs regulated advice, guaranteed valuations or stock picks, and its output should always be checked against your own figures.
Download the skill