How to use the Private Company Research agent skill

What an agent skill is and what this one does

An agent skill is a folder with a SKILL.md file that tells an AI coding agent how to handle a certain kind of task. When the skill is installed and matches your request, the agent follows its instructions instead of improvising.

This skill sets out a structured research workflow for unlisted companies. It proposes a team of specialist roles: business model, financials, competition, risk and governance, technology, and alternative signals such as hiring, patents, lawsuits, app data and supply chain. It then gathers scattered public information, cross-checks it, and labels each data point by confidence (high, medium or low).

• Separates verified facts from the agent's own inferences
• Looks for financing history and valuation changes across funding rounds
• Falls back to a first-principles mode when information is very thin

Install the skill from a ZIP download

Download the ZIP for private-company-research from this page and unzip it. The skill is a folder containing SKILL.md. Place that folder in one of these locations:

For all of your projects, put it in ~/.claude/skills/. For a single project, put it in the project's .claude/skills/ folder.

You can also upload the skill in the Claude apps under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/private-company-research/
• Project only: .claude/skills/private-company-research/ inside your repository
• Claude apps: Settings → Capabilities → Skills, then upload

Check the requirements before you start

The skill is built for Claude Code or Codex, and it expects the agent to have subagent or team capability, because it references TeamCreate and TaskCreate. It also needs web search to pull public sources, and shell access to the repository's tools/ directory for exact arithmetic.

The skill runs scripts and uses the network. Your repository should contain an AGENTS.md file with research quality rules, and the agent needs the date command to confirm today's date so that data cutoffs are stated correctly.

• Claude Code or Codex with subagent/team capability
• Web search access
• Shell access and the repository's tools/ directory (for example python3 tools/financial_rigor.py)
• The date command
• An AGENTS.md file in the repository

How to trigger the skill with a natural request

You do not need special syntax. Name the company and say what you want, and the agent will match the request to the skill. The skill itself asks you to confirm the team structure before it starts, so you can review the plan first.

Good requests name a specific unlisted company and a clear goal, such as due diligence before an investment or a structured valuation for a study. Ask for the confidence labels and source gaps if you want them spelled out.

• Do a private company research report on Stripe, focusing on business model and valuation, with confidence labels on every number.
• Research Xiaohongshu as an unlisted company: map the competition and list the risks, and tell me which facts are verified and which are your inferences.
• I'm doing early due diligence on a private company. Use the private company research framework and state the data cutoff date.

What a research run produces

The skill asks the agent to build a team and create tasks. Each specialist covers one area, for example the business model, the financial reconstruction with several valuation methods, the competitive landscape, and the risk and governance picture. The team lead then combines the results and cross-checks them.

Expect many tables and confidence markers. The skill asks for funding history, multiple valuation methods, and a list of data points that could not be verified. Treat the output as a structured draft to check, not a finished answer.

• Business model and unit economics estimates
• Funding timeline and valuation changes
• Valuation ranges from several methods, with stated assumptions
• Competitive and risk assessment with confidence ratings
• Explicit notes on gaps and unverified single-source data

Safety notes and how to verify output

The skill warns that AI tends to fill information gaps with plausible guesses, and that public sources can lean toward good news because companies publicise their wins. Do not make investment decisions based on the report without independent verification.

Because the skill uses web search and secondary sources, figures may be outdated or incomplete. It also runs scripts and makes network requests, so review what the agent executes in your shell and only run it in a repository you trust.

• Check every high-stakes number against the primary source it cites
• Ask the agent to say so when it has no data rather than estimate
• Review shell commands before approving them
• Do not use the skill for listed-company stock picking; standard filings and market data are more reliable there

Tips for better results

Give the agent the company's full name and any listing or ticker context, so it searches the right entities. Ask for the data cutoff date in the report header, which the skill is designed to state.

If a company has very little public data, ask for the first-principles view: what problem it solves, why this team, the realistic ceiling and failure modes, and the key validation milestones.

• Use the full legal or common name of the company
• Ask for the first-principles mode when data is scarce
• Ask the agent to list single-source claims separately