Deep Stock Analysis Workflow (UZI-Skill deep-analysis)
Input a stock ticker, automatically collect 22 data dimensions, run 65 AI investor evaluations, and produce a Bloomberg-style HTML analysis report.
The skill explicitly requires data from public sources and prohibits fabrication, and includes data gap handling. However, it lacks explicit user confirmation mechanisms (except for name correction and update prompts), and automatic Playwright browser automation and remote tunnel features exist without clear permission scoping. Data flow transparency is moderate, dependencies on external services (akshare, yfinance, etc.) are not fully assessed. Score 10: main risks visible but permissions/isolation incomplete.
The skill has detailed workflows and hard gates, but heavily relies on external data sources and browser automation, posing network and dependency risks. Test files exist but are limited in number and don't cover critical paths. Static review cannot execute, so reliability score is capped at 10, but evidence is insufficient, hence 8.
The skill targets A-shares, HK, and US stocks, with clear scenarios and good Chinese support. However, core functions depend on overseas services (e.g., yfinance, DuckDuckGo) and domestic data sources, which may be unreachable in network-restricted environments. Trigger conditions are clear but boundary descriptions are limited, hence 12.
Documentation is thorough, including version, author, license, update log, FAQ, but lacks a clear changelog and installation dependency notes (e.g., Playwright browser installation). Maintenance responsibility is clear, but governance is incomplete, hence 10.
The skill claims to generate complete reports, but static review cannot verify output quality. There are many rules against vague language, but no actual output samples. Marginal value may be high, but evidence is lacking, hence 6.
There are claimed tests (e.g., test_all_renderers.py) but no test results provided, and key paths are not covered. Independent reproduction is not possible, hence 4.
- Core functionality relies on multiple free data sources (akshare, yfinance, etc.) and DuckDuckGo search, which may be unreachable from mainland China networks, causing analysis failure or incomplete data.
- The skill includes automatic Playwright browser automation (may launch Chromium) and Cloudflare Tunnel remote mode, requiring explicit user consent and permission control.
- Simulated comments from 66 investors are AI-generated and do not represent real opinions, posing a risk of misleading users; disclosure is necessary.
- Data quality depends on crawler stability; fields may be outdated or wrongly scraped. Users should verify key data independently.
What does this skill do, and when should you use it?
This is an Agent Skill for individual stock research covering A-shares, Hong Kong and US stocks. It orchestrates parallel data collection across 22 dimensions (quotes, financials, K-line, LHB, sentiment, etc.), runs 17 institutional valuation models (DCF, Comps, LBO, 3-statement), and has 65 virtual investors from different schools (Buffett, Zhao Laoge, etc.) score the stock. The agent must intervene for qualitative analysis and role-play, then generate a self-contained HTML report with bull/bear debates, sensitivity heatmaps, and buy zones. The skill emphasizes data integrity and fact-checking, with 13 mechanical self-review gates.
Runs Python scripts (e.g., stage1 and stage2) to complete the pipeline; concurrently fetches 22 data dimensions (market data, financials, technical indicators, industry chain, macro, policy, etc.); executes DCF, Comps, LBO, three-statement projection via lib/fin_models; scores 65 virtual investors using rule engine and writes panel.json; requires agent to read panel.json, spawn sub-agents for role-play, and write agent_analysis.json between stage1 and stage2; generates HTML report, share card and war report images (share-card.png, war-report.png).
- An investor wants a comprehensive fundamental and technical analysis before deciding to buy a specific A-share stock.
- An analyst needs to quickly produce an institutional-style initiation of coverage report for an upcoming investment committee meeting.
- A user suspects a stock may be a scam (杀猪盘) and uses the `scan-trap` command for targeted screening.
- A user wants equivalent deep analysis for HK or US stocks, e.g., inputting `00700.HK` or `AAPL`.
- A user is away from the computer and needs to view the report on mobile, using the `--remote` mode to get a public link.
What are this skill's strengths and limitations?
- Completely free data sources, no API key required.
- Multi-market (A-shares, HK, US) and multi-platform (Claude Code, Codex, Gemini CLI, etc.) support.
- Built-in strict quality controls (13 self-review gates) that prevent generating erroneous reports.
- Offers multiple depth modes (quick, standard, deep) to fit different time constraints.
- Deep analysis takes 5-20 minutes, which may be too slow for quick decisions.
- Heavily relies on LLM reasoning and role-play capabilities; performance may vary by model.
- Data collection is dependent on network stability; some sources may have anti-scraping measures (e.g., Eastmoney push2).
- Accessing Chinese data sources from overseas may be slow; setting MX_APIKEY is recommended.
How do you install this skill?
For Claude Code: add the plugin marketplace and install: /plugin marketplace add wbh604/UZI-Skill then /plugin install stock-deep-analyzer@uzi-skill. For Codex: instruct it to follow the .codex/INSTALL.md in the repo. A more general method is to git clone the repo and then pip install -r requirements.txt. URL: https://github.com/wbh604/UZI-Skill.
How do you use this skill?
In Claude Code, send a command like /stock-deep-analyzer:analyze-stock 贵州茅台 to trigger full analysis. Specialized commands are also available, e.g., /stock-deep-analyzer:dcf 600519 for DCF valuation, /stock-deep-analyzer:quick-scan 002217 for a quick scan. For direct script use: python run.py 600519 --depth medium.