How to use the Earnings Review Deep-Read agent skill (earnings-review)

What the earnings-review skill makes the agent do

Once installed, the skill gives the agent a fixed workflow for analysing one company's results. It starts by checking how easily the primary materials can be obtained, then pulls the original filings and call transcripts from company investor relations pages and filing sites. The source text names SEC EDGAR, HKEX, CNINFO, Seeking Alpha, Xueqiu and others as places to look.

The workflow then covers revenue and profit, cash flow, balance sheet health, management tone and promise tracking, and footnote red flags. It compares the current period with earlier quarters and with management's prior guidance. The skill is written in Chinese and is a Codex adapter of a longer canonical workflow.

• Rates source quality from A (full original text) to C (news and summaries only) and adjusts the analysis to match.
• Checks cash flow against net profit, free cash flow, buybacks and dividends.
• Looks at management tone, such as whether answers are specific or evasive, and tracks promises made in earlier periods.
• Scans footnotes for related-party deals, dilution, contingent liabilities and concentration of customers.
• Writes a report to a reports folder with a clear verdict on whether results beat, met or missed expectations, and how they affect the investment thesis.

Install the skill from a ZIP file or the Claude apps

There are two routes. For Claude Code or Codex, download the ZIP from the directory page and unzip it into ~/.claude/skills/ for all your projects, or into a project's .claude/skills/ folder to limit it to that project. The skill's folder contains a single file, SKILL.md.

In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills. The source does not describe any other installation method, so do not assume one.

• Personal install: unzip into ~/.claude/skills/.
• Project install: unzip into the project's .claude/skills/ folder.
• Claude apps: upload the ZIP under Settings → Capabilities → Skills.

Get the full repository before you rely on the workflow

The SKILL.md file alone is not enough for the full workflow. It refers to tools/financial_rigor.py, tools/report_audit.py, skills/financial-data.md and AGENTS.md, and those files live in the full xbtlin/ai-berkshire repository. Without them, the agent may fail or skip steps.

The Codex adapter also says to run commands from the repository root and to locate the actual checkout path rather than assuming a fixed home-directory path. Python 3 is needed for the repo tools.

• Clone or check out the full xbtlin/ai-berkshire repository so that tools/ and skills/ are present.
• Install Python 3, which the referenced tools require.
• Run the agent from inside the repository, or point it to the checkout location.

How to trigger the skill with natural requests

Ask for the skill by naming a company and a period. The source gives these input patterns: a company name followed by a quarter, a fiscal year, or the word for latest. If you omit the period, the workflow defaults to the most recent report.

The skill runs its date check first, so the report header states the data cutoff date. Results depend on what web sources are reachable at run time.

• Tencent 2025Q4
• PDD 2025 annual report
• Meituan latest (reads the most recent period by default)

What the output looks like and how to check it

The report is saved as reports/{company}-earnings-{period}.md in the repository. It opens with a one-page table of core figures, then covers the three biggest changes, management tone, footnote findings, key Q&A questions, the link to your investment thesis if you hold the stock, and a conclusion.

Before a report is published, the source describes a data check using tools/report_audit.py. The report passes or is sent back for correction. Treat the output as a first pass and verify important numbers yourself.

• Expect tables for income statement, cash flow and balance sheet trends.
• Check the source-quality grade the report states; a C-grade read skips footnote analysis.
• Review the audit verdict before relying on any figure.

Safety notes: scripts and network access

This skill runs scripts and uses the network. The workflow calls Python tools in the repository and fetches data from filing sites and third-party sources. The source names no API keys, but the data sources it uses may be inaccurate. The skill itself flags cross-source differences above 1% for key figures.

The skill outputs position suggestions. It is not a substitute for independent checking before any buy or sell decision.

• Read the scripts in tools/ before running them on your machine.
• Be aware that outputs depend on live web sources and may change between runs.
• Keep your own verification step before acting on a suggested position.

Tips for getting better reports

Give the agent the full company name or ticker and the exact period, so it picks the right filing. Ask for the source grade explicitly if you want to know how much primary material was used. If you already hold the stock, say so, because the report includes a section on how the results affect your thesis.

• Use the company name in the language you are most comfortable with; the skill accepts Chinese or English inputs.
• Start with a single company and period before comparing several.
• If the report lists a C-grade source, treat it as a summary rather than a primary-source read.