Guide · Agent skill
How to use the Deep Company Series Writer agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Deep Company Series Writer is an agent skill from the ai-berkshire repository that guides an AI agent through planning and drafting a series of 3 to 8 long articles explaining one listed company from scratch. It is for independent investors and bloggers who publish value-investing style explainers, and for writers who need consistent numbers across parts. After this guide you will know what a skill is, how to install this one, how to trigger it, and what it needs from your environment.
Deep Company Series Writerxbtlin/ai-berkshire · MIT Download the skill Quick start
- 1Download and unzip. Download the skill ZIP from this page and unzip it so that SKILL.md sits in a skill folder.
- 2Place the folder. Move the folder into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder for one project. Alternatively, upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Open the repository checkout. Work from a checkout of the ai-berkshire repository so the tools/ folder and AGENTS.md rules are available.
- 4Make a natural request. Name the company and ask for a series of long articles explaining it, in plain language.
- 5Review the plan. Check the proposed number of parts and core arguments before any drafting begins.
- 6Review drafts before publishing. Read each saved part and request revisions. Nothing is pushed until you approve.
What is an agent skill and what does this one do?
An agent skill is a folder containing a SKILL.md file of instructions that an AI agent loads when your request matches the skill's purpose. This skill is generated from the file skills/deep-company-series.md in the ai-berkshire repository, so Claude Code and Codex users share one workflow.
Once installed, the agent plans and drafts a series for one company covering its business, moat, profit engines, hidden assets, AI-era risks, financials, management and valuation. It follows a fixed chapter template, a style guide that avoids hype words, and a fact-checking checklist that flags false precision such as weighted probability estimates. It also tells the agent to check today's date, cross-check numbers across chapters, and state uncertainty and data gaps. The workflow asks you to review before anything is published.
- Chooses the number of parts from the company's complexity, between 3 and 8.
- Uses a chapter template with a hook opening, a summary and a closing disclaimer in each part.
- Flags precision claims it cannot support and asks for source labels on non-financial data.
- Does not predict share prices or give investment recommendations.
Install the skill from the ZIP download
Download the ZIP for this skill from our page and unzip it. The skill is a single file, SKILL.md. To make it available across your projects, place the unzipped folder in ~/.claude/skills/. To limit it to one project, place it in that project's .claude/skills/ folder.
If you use the Claude apps instead of the command line, open Settings, go to Capabilities, and upload the skill under Skills.
- Personal install: ~/.claude/skills/ (available everywhere you use it).
- Project install: .claude/skills/ inside the project folder (available only there).
- Claude apps: Settings → Capabilities → Skills, then upload.
~/.claude/skills/Check what the skill needs before you run it
This skill is not a standalone writing tool. It expects you to work inside a checkout of the ai-berkshire repository, because it refers to the repository's tools/ folder and its AGENTS.md rules. It also needs shell access to run the Python tools and the date command, and it uses web search to reach public filings and analyst reports. Git is needed for the review and push steps.
The skill runs scripts and uses the network. Its workflow calls python3 tools/financial_rigor.py for exact arithmetic and runs the date command before research. Review those steps before granting permission, and make sure you are using a repository checkout you trust.
- Shell access to run python3 tools/financial_rigor.py and the date command.
- A checkout of the ai-berkshire repository, including tools/ and AGENTS.md.
- Web search and access to public filings and analyst reports.
- Git for the review and push steps.
- Scripts: yes. Network: yes.
python3 tools/financial_rigor.pyTrigger the skill with a natural request
You do not need special syntax. Name the company and say you want a series of long articles explaining it. The skill's own examples use the phrase 看懂 ("understand") with the company name, so requests in Chinese match its intended use. The skill is designed for Chinese-language public channels such as WeChat-style articles.
Be specific about the reader and the goal, for example whether you want a simple company with few parts or a complex one with many. The agent will propose the number of parts and the core arguments before writing, so you can correct the direction early.
- Example: "Use the deep-company-series skill to plan a 看懂 series on [company name] for my investment blog."
- Example: "Draft the first part of a 3-part series on [company], focused on its business and moat."
- Example: "Check the numbers across all parts of my series on [company] for consistency."
Follow the workflow and review before publishing
The skill works in phases: research first, then writing part by part in order, then a cross-part consistency check, then a final privacy check before pushing. Each part is saved as a markdown file in a folder under reports/ in the repository. If an older series folder already exists, the skill creates a new folder with a date suffix rather than overwriting it.
Nothing is pushed to GitHub until you have reviewed the drafts and asked for changes. When you give revision feedback, the skill checks the facts against the raw data first, then rates each change as a hard error, a subjective wording problem, a presentation detail, or an unreliable third-party figure.
- Research: read recent annual reports and quarterly results, and at least three independent analyst reports.
- Write: save each part in order, starting from part 01.
- Consistency scan: check shared numbers, terms and cross-references across parts.
- Privacy check: grep the repository for local usernames and personal identifiers before pushing.
- Push only after your review.
grep -r "<本机用户名>\|/Users/\|<个人身份信息>" reports/ | headKnow the limits and disclaimers
The series is framed as research. Each part ends with a note that it is not investment advice, and the skill explicitly does not make investment decisions for readers. It does not predict share prices, and it avoids calculating weighted expected returns from subjective probabilities. Scenarios are presented with triggers and direction instead.
The skill depends on public information. It is not suitable for companies without public financial disclosures, because the workflow relies on annual reports and analyst reports. For a single research note, an earnings review or an industry study, the file points you to other skills instead.
- Not for single research reports, earnings reviews or industry research.
- Not for companies with no public financial disclosures.
- Not a source of investment recommendations.
What people use it for
- Planning a 3-part explainer on a company with one clear business and a few key questions.
- Drafting an 8-part series on a diversified company with a large investment portfolio.
- Checking that market value, profit figures and shareholding percentages match across all parts.
- Producing a repeatable template for company deep-dives rather than one-off research notes.
Frequently asked questions
How do I install the deep-company-series skill?
Download the ZIP from this page and unzip it. Place the folder in ~/.claude/skills/ for all projects or in a project's .claude/skills/ folder for one project. In the Claude apps, you can instead upload it under Settings → Capabilities → Skills.
Is the deep-company-series skill free?
The source file is released under the MIT licence, as listed on this page. Check the repository's licence file for the full terms before you reuse or redistribute it.
What do I need to run this skill?
You need shell access to run python3 tools/financial_rigor.py and the date command, a checkout of the ai-berkshire repository including tools/ and AGENTS.md, web search for public filings and analyst reports, and Git for the review and push steps.
Does the skill run scripts or use the internet?
Yes on both counts. It runs Python tools and the date command, and it uses network access for web search and public filings. Review those steps and use a repository checkout you trust.
Does it work for companies without public financial data?
No. The workflow relies on annual reports and analyst reports, so it is not suitable for companies with no public financial disclosures.
What should I use instead for a single research note or earnings review?
This skill is for multi-part company series only. The skill file points users to other skills for single research reports, earnings reviews and industry research, namely investment-research, earnings-review and industry-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
Use this skill if you publish a structured, value-investing style explainer series on a listed company and want consistent numbers and strict fact-checking across parts. Skip it if you need a single research note, an earnings review, investment advice, or if the company does not publish financial disclosures.
Download the skill