How to use the Industry Investment Research agent skill (industry-research)

What is an agent skill, and what does this one make the agent do?

An agent skill is a folder with a SKILL.md file that gives an AI agent a reusable set of instructions for one kind of task. When a request matches the skill, the agent follows its workflow instead of improvising.

Once installed, this skill works when you name an industry or investment theme. The agent builds a logic chain from an underlying trend to beneficiary companies, sketches the industry's upstream-to-downstream supply chain, and scans listed and some unlisted companies in A-shares, Hong Kong, the US and other markets. It then analyses leading companies on business quality, moat, risks, management, valuation and a star rating, and ends with a sample portfolio allocation and a risk review.

• The skill is written in Chinese and follows a workflow modelled on well-known investors' thinking.
• It asks the agent to run the date command first and state a data cutoff date in the report.
• It asks for a data-audit step before the report is published.

Install the industry-research skill

Download the ZIP for this skill from this page. Then choose one of two routes.

For Claude Code, unzip the file into ~/.claude/skills/ to make it available to you everywhere, or into a project's .claude/skills/ folder to limit it to that project. For the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

The skill's own instructions also refer to a Codex version generated from the same source, so Codex users can follow the same workflow with the adapter notes in the file.

• Claude Code, all projects: unzip into ~/.claude/skills/
• Claude Code, one project: unzip into the project's .claude/skills/
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

What you need before you run it

The skill does more than write text. It expects several things to be in place, and the file we reviewed does not include some of them.

The skill calls helper scripts in the ai-berkshire repository, such as tools/financial_rigor.py and tools/report_audit.py. These are not included in the file reviewed, so you need the full repository checkout for the audit and calculation steps to work. The skill also needs shell access to run the date command, web search and reliable financial data sources, and subagent or task capability to scan companies in the background.

• Shell access (for the date command and local tools)
• The ai-berkshire repository checkout, including tools/ scripts such as financial_rigor.py and report_audit.py
• Web search access and reliable financial data sources
• Subagent or task capability for background company scanning

How to trigger the skill with a natural request

You do not need a special command. Name an industry or investment theme in plain language, and say you want a research report or industry map. The skill's workflow starts when the agent sees a request that matches its description.

Be specific about the theme. A narrow theme such as a single supply-chain segment usually gives a more useful map than a broad label such as 'technology'. Remember that the skill's instructions are in Chinese, so Chinese-language requests fit the workflow best.

• Example: research the nuclear power equipment supply chain and list the listed companies by tier.
• Example: build a logic chain from data-centre power demand to beneficiary companies.
• Example: produce an industry map and a sample portfolio allocation for the humanoid robot theme.

What the output contains

The report is written to a file in your home directory, named after the industry. Its sections follow the source workflow: logic chain validation, supply-chain map, global company scan sorted into four tiers, four-investor analysis of leading companies, industry risk review, long-term trend judgement, portfolio suggestions and a summary memo.

Companies are given an information-sufficiency grade (A, B or C) so you can see how much to rely on each analysis. Treat these grades and every figure as a starting point for your own checking.

• Logic chain and supply-chain map
• Company scan split into Tier 1 to Tier 4
• Analysis per leading company, with a 1 to 5 star recommendation
• Portfolio layers and a summary verdict table

Safety notes: scripts and network access

This skill runs scripts and uses the network. The scripts are local helper tools in the repository, and the web searches and data lookups go to external sources. Review the repository's tools/ folder before you run it, and be aware that the agent may run shell commands on your machine.

The skill asks for a data-audit step before publishing. It samples about 15% of the figures and asks the agent to check them against reliable sources. Passing that audit does not make the investment advice correct.

• Review the scripts in tools/ before running them.
• Run the agent in a workspace where shell commands are acceptable to you.
• Check the audit output and verify key figures against primary sources yourself.

Tips for better results

Ask for one segment at a time when the theme is large. The skill's scan is broad, and a focused request keeps the company tables manageable. Ask the agent to label uncertainty and source gaps, since the skill already asks for this, and check that it did.

Read the logic chain critically first. If an arrow in the chain is weak, the rest of the report inherits that weakness.

• Narrow the theme to one or two supply-chain segments.
• Ask for English-language summaries of a Chinese report only if you accept that the skill itself is written in Chinese.
• Check that each price target or position size has a cited source before you use it.