Guide · Agent skill
How to use the earnings-team agent skill: four-investor earnings analysis
Updated · written with AI from the project's own documentation and checked against our catalogue data
earnings-team is an agent skill from the AI Berkshire project. You give the agent a company name and reporting period, and it runs a structured earnings-report analysis split across four investor-style research passes, then drafts a public-friendly article in Chinese. This guide explains what the skill does, how to install it, how to trigger it, and what to check before you rely on the output. After reading, you should be able to set it up in Claude or Codex and run it on a filing.
Earnings Report Team (Four Investors)xbtlin/ai-berkshire · MIT Download the skill Quick start
- 1Download the ZIP. Download the earnings-team ZIP from this page.
- 2Install the skill. Unzip into ~/.claude/skills/ or a project's .claude/skills/, or upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
- 3Get the repository tools. Make sure you have the AI Berkshire repository with tools/ and AGENTS.md, and Python 3 installed.
- 4Make a request. Ask for an analysis by company name and period, such as 腾讯 2025Q4.
- 5Let the four passes run. The agent runs four parallel research passes and merges them into a report.
- 6Check the figures. Verify key numbers against the official filings and note any source gaps.
- 7Review the article. Read the edited article and the reader-review notes before any publication.
What the earnings-team skill makes the agent do
An agent skill is a folder with a SKILL.md file that gives an AI agent a specific workflow to follow. When you ask for a matching task, the agent reads the skill and follows its steps instead of improvising.
earnings-team works in three phases. First, four research passes run in parallel, each framed around a different investor's questions: business quality (段永平), financial quality (巴菲特), competitive change (芒格), and management risk signals (李录). Second, the Team Lead agent merges their findings into a research report, looking for points of agreement and contradiction. Third, an editor agent rewrites the report as a WeChat-style article and a reader-reviewer agent critiques it from a general investor's point of view.
- Input format in the source: 公司名 + 季度, for example 腾讯 2025Q4, PDD 2025年报, or 美团 最新.
- The skill is a workflow for reading financial filings. It is not a source of market data on its own; the agent must fetch filings from the web.
- The skill says analysis depth depends on source quality: A (full original documents), B (partial or third-party summaries), C (news and data-site summaries only).
Install the skill: ZIP download or Claude Settings
Option 1 is a manual install. Download the ZIP from this page and unzip it into your user skills folder ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill limited to one project.
Option 2 is to upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
- Confirm the unzipped folder contains SKILL.md (the only file listed for this skill).
- The skill's description refers to skills/earnings-team.md in the source repository; the packaged SKILL.md includes a Codex adapter note for running it in Codex.
Set up the repository tools the skill expects
The skill calls shared tools from the AI Berkshire repository, not from the skill folder alone. It expects you to run commands from the repository root, using paths such as tools/financial_rigor.py. It also refers to AGENTS.md for research-quality rules, so the agent needs access to that file.
The skill also says that if the current session starts outside the repository, the agent should locate the actual checkout path rather than assume a fixed home-directory path.
- Python 3 is required to run tools/financial_rigor.py.
- Access to the repository's tools/ folder and AGENTS.md file is needed.
- Subagent or background agent capability is needed to run the four research passes in parallel. Without it, the workflow as written cannot run its parallel step.
Trigger the skill with natural requests
You do not need a special command. Name the company and the period in a request that asks for an earnings analysis or a public article. The source describes its input as a company name followed by a reporting period.
Give the agent the filing or the access it needs. The skill's first step is to fetch primary material, so a request that names the company and period is the starting point.
- Example: 用 earnings-team 分析腾讯 2025Q4 财报
- Example: 对 PDD 2025年报 做四大师财报精读，并写成公众号文章
- Example: 美团 最新 财报精读，标注资料可得性等级
- Note: the skill's own examples are written in Chinese; Chinese requests match its wording most closely.
What the output looks like and how to check it
The research report follows a fixed structure: a one-line conclusion, the three most important changes, a four-lens scoring table, a key-metrics table, deeper analysis per lens, a management tone and promise-tracking section, a section on what each investor would do, and a conclusion. The article draft is then produced from that report.
The skill's cash-flow checks are explicit. It treats operating cash flow above 100% of net profit as favourable and below 80% as a warning sign, and it asks the agent to cross-validate key figures across at least two sources. Verify these yourself against the filing before you use any number.
- Check every figure against the original filing and note which source it came from.
- Look for the uncertainty labels the skill asks for and treat unlabelled claims with caution.
- Read the reader-review section; it lists remaining questions the article did not answer.
Safety notes: scripts and network access
This skill runs scripts and uses the network. The source says the agent runs python3 tools/financial_rigor.py for calculations such as market cap checks, valuation and three-scenario estimates, and fetches filings and transcripts from the web. Review the repository's tools/ folder before running anything, and only run it on a checkout you trust.
The skill's sources include investor relations pages, SEC EDGAR, HKEX, CNINFO, Seeking Alpha and Xueqiu. Network access is needed for these, so run it only where that is acceptable to you.
- Read tools/financial_rigor.py before running it.
- Network access is required for fetching filings and transcripts.
- Do not treat the output as investment advice; the skill itself says the output is AI-generated analysis with no accuracy guarantee.
What people use it for
- Drafting a Chinese-language earnings analysis for a WeChat-style public account.
- Running a checklist of cash flow, margin and management-commitment checks against official filings.
- Using the AI Berkshire workflow inside Codex or Claude Code for repeated earnings reviews.
- Comparing a company's results with what management promised in the previous period.
Frequently asked questions
Is the earnings-team skill free?
The source gives the licence as MIT. The skill itself carries no price in the source, but running it needs a Claude or Codex setup with subagent capability and web access.
How do I install earnings-team?
Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder. Alternatively, upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
What do I need to run earnings-team?
Python 3 for the shared tools/financial_rigor.py script, access to the repository's tools/ folder and AGENTS.md, web access to fetch filings, and subagent or background agent capability for the four parallel passes.
Can earnings-team tell me whether to buy or sell a stock?
No. The skill is an AI-generated analysis with no guarantee of accuracy, and it says uncertainty must be labelled. It is not designed to make buy or sell decisions, so verify figures before acting on any conclusion.
What happens if the agent cannot find the company's official filings?
The skill lowers analysis depth. If only news summaries are available (grade C), it focuses on core data changes, skips footnote analysis and labels the result as insufficient primary sources. Results are weaker in that case.
How does earnings-team compare with doing the analysis myself?
The skill gives you a structured checklist and a consistent report format, covering cash flow, margins, competition and management commitments. It does not replace reading the filing, and its output needs the same checks you would apply to any AI-generated research.
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
earnings-team suits Chinese-speaking investors and writers who want a structured earnings draft, and analysts who want a checklist of cash flow, margin and management-commitment checks against official filings. It is a poor fit for anyone wanting a buy or sell answer, or for publishing unchecked output as financial advice. Verify every figure against the original filing before use.
Download the skill