How to use the UZI Stock Analysis Suite agent skill

What is an agent skill and what does UZI make the agent do?

An agent skill is a folder of instructions and supporting files that an AI agent such as Claude reads when a request matches the skill's purpose. The instructions tell the agent what to do and which scripts to run.

With the UZI skill, the agent can analyse a ticker with bundled Python scripts that score the stock and generate a Bloomberg-style HTML report. It can also simulate an investor jury of famous-investor personas, check hot-money seat activity on the Dragon-Tiger list (LHB), screen for pump-and-dump trap signals, and produce valuation and IC memo style output.

• Deep research and quick scans of a single stock
• Investor panel review (which famous-investor personas would buy)
• Hot-money and Dragon-Tiger list (LHB) analysis for A-shares
• Trap detection for pump-and-dump warning signs
• Valuation, IC memos and HTML reports

How do you install the UZI skill?

Download the ZIP of the skill from this page and unzip it. You have two options depending on how you use Claude.

For Claude Code, unzip the folder into your user skills directory ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill available only in that project.

In the Claude apps, you can upload the ZIP instead. Go to Settings, then Capabilities, then Skills, and add the skill there.

• Option A: unzip into ~/.claude/skills/ for all projects
• Option B: unzip into .claude/skills/ inside one project
• Option C: upload the ZIP in the Claude apps under Settings → Capabilities → Skills

What does the skill need to run?

The skill runs Python scripts, so you need Python 3 installed. It also needs the packages listed in the requirements.txt file that comes with the skill. Install those packages before the first run.

The skill fetches market data, so it needs internet access. Cloudflare Tunnel (cloudflared) is optional and is only needed for the --remote option, which creates public links for viewing reports on other devices.

• Python 3
• Packages listed in requirements.txt
• Internet access to fetch market data
• Optional: cloudflared, only for --remote public links

How do you trigger the skill with natural requests?

You do not need to type commands. Ask in plain language that names a stock and the kind of review you want, and the agent should pick the matching workflow. The skill's own guidance says a request such as 'analyse' or 'valuation' for a stock should trigger the full deep-analysis workflow, while a request about the Dragon-Tiger list or hot money should trigger the LHB workflow.

Useful phrasings from the skill's routing guidance include requests about investor panels ('which investors would buy'), hot-money or seat activity, and checks on a tip from a teacher, group or friend, which triggers the trap detector.

• 'Analyse <ticker>' or 'Do a valuation for <ticker>' for a full deep analysis
• 'Which investors would buy <ticker>?' for an investor panel view only
• 'Check the Dragon-Tiger list and hot-money activity for <ticker>' for LHB analysis
• 'A friend recommended <ticker>, is it safe?' for trap detection

How do you run the scripts yourself?

You can also run the entry point directly. Run run.py from the repository root (not from a scripts/ subfolder). Replace <ticker> with the stock code.

The --no-browser flag stops the report from opening automatically. The --remote flag produces a link for mobile or remote viewing and uses the optional Cloudflare Tunnel. The --school F flag runs a single investor school, such as the A-share hot-money school. The --depth option accepts lite, medium or deep, with lite the fastest and deep the most complete.

• --no-browser: generate the report without opening a browser
• --remote: create a remote or mobile link (needs cloudflared)
• --school F: run a single investor school, such as A-share hot-money
• --depth lite | medium | deep: choose how thorough the run is

What are the safety notes for scripts and network access?

This skill runs third-party Python code that fetches data from the internet, and the --remote option may open a tunnel. Read the scripts and the requirements file before you run them, especially if you are not comfortable reviewing code. Do not run the skill in an environment where outbound internet access or executing third-party code is not allowed.

The skill's own rules tell the agent to treat script output as scoring data rather than a final verdict, and to avoid inventing numbers. Check any figure in a report against the underlying data before you rely on it.

• Review the Python code before running it
• Scripts fetch market data over the network
• --remote may open a Cloudflare Tunnel and create a public link
• Do not use it where executing third-party code is restricted

What tips help you get useful results?

Run the lite or medium depth first to see whether the stock is worth a closer look. Reserve the deep depth for when you need a DCF, an IC memo or initiation-style output, because the deep workflow asks the agent to review the panel data itself before the final report is assembled.

If you work on the repository itself, note that the skill keeps several entry files. The root SKILL.md points to the specialised workflows in skills/deep-analysis/, skills/investor-panel/, skills/lhb-analyzer/ and skills/trap-detector/. Start from the file that matches your task.

• Start with --depth lite or medium for a first pass
• Use deep only when you need a DCF or IC memo
• Read the specialised SKILL.md for the workflow you need
• Keep report output as one input among several sources