UZI Stock Deep Analysis Engine
Enter a ticker and get a Bloomberg-style HTML report scored by 66 simulated investors, powered by 22 data dimensions and 17 institutional methods.
Evidence shows clear role boundaries (agent must not fabricate numbers, --remote requires user confirmation, data gaps are disclosed transparently), and heavy use of public data sources with no mandatory API key. However, permission and isolation are incomplete: scripts read many environment variables (potentially sensitive), there are auto-install dependencies and curl script execution (though explained, may exceed least privilege), and no explicit sandbox or rollback mechanism. Additionally, Hermes scanner reported 168 security findings (though project explains as false positives, cannot verify statically). Hence 15, deducted for incomplete permission management, sandbox isolation, and lack of third-party security verification.
Many self-reported tests (claiming 332 or 642 pytest), but static review only confirms existence of some test files, covering internal components like renderers and base fetchers, not full end-to-end critical path. Main run.py depends on akshare, yfinance, etc., and network dependencies (multiple data sources) may cause instability, but project provides multi-source fallback. Error messaging is well-designed (explicit network issue prompts, data gap handling). However, cannot verify actual run success, so only 8.
Use cases and audience are very clear (A-share/HK/US stock analysis), support Chinese data sources and mainland network environment fallbacks (e.g., pip mirrors). Trigger conditions are explicit (via /stock-deep-analyzer commands or natural language), and offers three depth modes and --school parameter. However, capability boundaries (e.g., limitations on financial data accuracy) are not fully declared, and some functionality heavily depends on overseas services (e.g., yfinance for US stocks) which may be restricted in mainland China; though alternatives exist, not fully proven. Hence 12.
Documentation structure is quite complete: AGENTS.md, CLAUDE.md, CODEX.md, README, changelog, BUGS-LOG, version (v3.9.2), MIT license, author and maintainer info (FloatFu-true). Version management (.version-bump.json) and install instructions exist. However, minor inconsistencies (e.g., package.json description says 51 investors, SKILL.md claims 66), and a dedicated changelog file (RELEASE-NOTES.md) is not in provided files. Hence 12.
Value proposition is clear (22-dimension analysis report and HTML), concrete outputs (DCF sensitivity table, IC memo). But cannot verify actual output quality, and workflow heavily depends on agent intervention (e.g., role-play 66 investors), potentially introducing subjectivity. Alternatives not prominently mentioned, marginal value limited. Hence 6.
Test files exist (e.g., test_all_renderers.py) but no test execution results seen, and no CI configuration. Key claims (e.g., '685 tests pass') cannot be independently verified. Third-party evidence (Hermes scan) is mentioned but no specific link provided, mostly self-reported. Hence only 4.
- Core functionality depends on multiple external data sources (Eastmoney, Xueqiu, akshare, etc.) and network; stability is limited by data source availability; performance may differ in mainland China vs overseas containers; thorough testing recommended.
- Workflow requires agent to self-role-play 66 investors and write agent_analysis.json, which introduces subjectivity risk and potential report bias.
- Contains behaviors like auto-installing dependencies, executing remote scripts (e.g., install-hermes.sh), and --remote launching Cloudflare Tunnel; ensure explicit user authorization.
- Project was flagged by Hermes scanner as DANGEROUS (168 findings); author claims all false positives, but careful code review before use or consider more mature alternatives.
- Some features (e.g., scrapping social hot lists from Weibo, Zhihu, etc.) may violate platform terms; use with caution.
What does this skill do, and when should you use it?
UZI-Skill is a deep stock analysis skill for A-shares (China), Hong Kong, and US stocks. It uses Python scripts to automatically collect 22 data dimensions (quotes, financials, LHB, social trends, etc.), apply 17 institutional analysis methods (DCF, Comps, LBO, etc.), and simulate 66 investors of diverse styles (from Buffett to hot-money traders) to generate scores and comments. The final output is a self-contained HTML report, share cards, and a text summary. All operations run locally with free data sources, no API key required. Current version v3.9.2, maintained by FloatFu-true, MIT licensed.
Reads a user-provided ticker (e.g., 600519 or Guizhou Moutai), runs a multi-stage pipeline via Python scripts. First, it concurrently fetches 22 data dimensions (real-time quotes, historical financials, candlestick data, LHB (Dragon-Tiger List), peer comparisons, social trends). Then it executes institutional modeling: DCF, Comps, LBO, three-statement forecasts. A rules engine and agent then simulate 66 investors to produce scores and qualitative commentary. Finally, it generates a self-contained HTML report with overall score, bull-bear debate, jury votes, DCF sensitivity heatmap, IC memo, and more, plus a sharable vertical image and a text summary. CLI supports flags like --school to filter by style, --versus for multi-stock comparison, --portfolio for portfolio analysis, and --remote to generate a public link.
- An individual investor wants a quick fundamental and valuation check on an A-share before buying; they type 'Guizhou Moutai' and wait 5-8 minutes for a full report.
- A trader wants to see a stock from a specific style perspective, e.g., using '--school F' to view only hot-money short-term analysis.
- An analyst needs a professional institutional-style report to share with colleagues, featuring DCF valuation and sensitivity tables in HTML.
- A user is suspicious of a stock recommended by a teacher or group and uses the trap-detector to scan for pump-and-dump signals.
- A portfolio manager compares 2-4 stocks using '--versus' to get a head-to-head report, or uploads a CSV to analyze portfolio health.
What are this skill's strengths and limitations?
- Completely free data sources with zero API keys, ideal for budget-conscious individual investors.
- Covers A-shares, Hong Kong, and US stocks, multi-market support.
- Rich analysis: 22 dimensions, 17 institutional methods, 66 simulated investors, professional and polished reports.
- Offers multiple depth levels (lite/medium/deep) to match time constraints.
- Active community with frequent updates (v3.9.2 latest) and detailed changelog.
- Relies on multiple third-party data sources; some (like Eastmoney) may block scraping, requiring MX_APIKEY or fallback chains.
- Long runtime: full analysis takes 5-8 minutes, deep mode 15-20 minutes.
- Simulated investors are based on rules and agent judgment, not real opinions, and do not constitute investment advice.
- Some data sources (e.g., Xueqiu portfolio contest) require login; skipped by default.
- In restricted networks (e.g., overseas access to Chinese sources), performance may be slow, requiring extra configuration.
How do you install this skill?
Supports multiple agents. For Claude Code: /plugin marketplace add wbh604/UZI-Skill then /plugin install stock-deep-analyzer@uzi-skill. Alternatively, clone and install: git clone https://github.com/wbh604/UZI-Skill.git && cd UZI-Skill && pip install -r requirements.txt. For Hermes, due to upstream Skills Guard false positives, use the one-liner script: curl -fsSL https://raw.githubusercontent.com/wbh604/UZI-Skill/main/install-hermes.sh | bash. Refer to the README for full details.
How do you use this skill?
In an agent, send commands like: /stock-deep-analyzer:analyze-stock 贵州茅台. Common commands include: /stock-deep-analyzer:quick-scan 002217 (30-second scan), /stock-deep-analyzer:scan-trap 002217 (trap check), /stock-deep-analyzer:dcf 600519 (DCF valuation). CLI also works: python3 run.py <ticker> --no-browser. More options (e.g., --depth lite/deep, --school A-I) are in the README.
How does this skill compare with similar options?
Compared to Anthropic's financial-services-plugins, UZI-Skill is specifically adapted to the A-share market and uses free data sources, while Anthropic's plugin targets US stocks and requires paid data.