Investor Panel — 65-AI Judge Jury for Stock Analysis
Get structured buy/sell signals and vote tallies from 65 simulated investors across 9 schools of investing.
Evidence shows: The skill only reads input files and writes output files, no system-level operations; the skill requires the user to output strict JSON, but lacks explicit user confirmation steps. Permissions are moderate, but no least-privilege principle or extra security measures are mentioned. Data flow is transparent, no sensitive data handling involved. Dependencies are common financial libraries with no known vulnerabilities. External effects limited to generated report files, rollback possible. Author identity is explicit (FloatFu-true), but publisher is not verified. Main deductions: no user confirmation, undefined permission boundaries, unverified publisher identity.
Evidence shows: The skill has clear execution steps and output format, covering multiple investment styles, dependencies (akshare, yfinance, etc.) are common financial libraries, but no test cases provided. The skill describes some error handling (e.g., data gap handling), but not systematic. Key path (generating panel.json) cannot be verified under static review. Deductions: no test evidence, incomplete error handling, undefined behavior outside happy path.
Evidence shows: The skill targets a clear scenario (panel analysis), with well-defined inputs and outputs, clear semantic trigger conditions (e.g., user requests 'panel/65 masters' view). Environment fit: dependent data sources may be unreachable in mainland China (e.g., Yahoo Finance), but the skill provides fallback chains and optional MX_APIKEY, and README has network-constrained environment instructions. Deductions: core function depends on external data sources, but reachability issues are explicitly mentioned; boundary conditions (e.g., missing data) handling is not detailed enough.
Evidence shows: The skill document is well structured, with background, execution steps, output format, completion checks, and good information layering. Version (3.9.2) and changelog (in README) exist, MIT license is explicit, author name present. But lacks FAQ, clear known-limitation disclosure (though partly in README), install instructions are in README not in skill file. Deductions: no install/dependency notes in skill, no FAQ, incomplete known-limitations disclosure.
Evidence shows: The skill generates structured panel.json with multiple investor signals and vote statistics, core task clear. But output format requires strict JSON, which may be user-unfriendly. Value proposition clear, but no example output or verification of generated results. Deductions: static review cannot verify output availability, no sample output, limited value evidence.
Evidence shows: The skill claims based on multiple sources (e.g., Buffett's letters), but no internal verification materials. README mentions 685 tests, but no test suite directly relevant to this skill. Source citations are not explicit (e.g., 'Source: Berkshire Hathaway Shareholder Letters' but no link). Deductions: no third-party execution evidence, no cross-source verification, sources not traceable.
- The skill output may be used for investment decisions; it should be explicitly stated as analytical reference only, not investment advice.
- Several external data sources (e.g., Yahoo Finance, some social media APIs) may be unreachable from mainland China; proxy or fallback is needed.
- The role-play style comments generated (e.g., youzi style) carry a risk of misleading; should be clearly marked as simulated content.
- Publisher identity is unverified; assess risk before use.
What does this skill do, and when should you use it?
This skill, part of the UZI-Skill monorepo, turns pre-generated financial data into a multi-perspective investment panel. It reads dimensions.json and raw_data.json for a given ticker, then runs 65 simulated investors — spanning classic value, growth, macro-hedge, technical, China value, A-share hot-money, quant, tech-leader, and AI-chokepoint styles — each scoring the stock according to their own methodology and producing a Pydantic Signal with signal, confidence, score, verdict, and comment. The output is a panel.json with consensus and vote distributions. It emphasizes authenticity by requiring each comment to mimic the investor's real public speaking style.
Loads investor metadata from scripts/lib/investor_db.py and seat_db.py, and reads dimensions.json and raw_data.json from .cache/{ticker}/. For each investor, it extracts relevant features based on their field whitelist, applies their methodology (e.g., value rules, growth criteria) to generate a Signal, and calibrates confidence based on rule hits. For 22 hot-money investors (Group F), it first checks if the ticker falls within their trading style range using seat_db.py; if not, it outputs a neutral signal with verdict '不适合' and high confidence. Finally, it aggregates all signals into panel.json, including bullish percentage, vote/signal distributions, and a list of investors.
- When a user asks 'What does the panel think of this stock?' to get a broad bull/bear overview.
- When a user wonders 'Would Buffett buy this?' and wants a role-played answer from a specific investor.
- When a user wants to see how different schools (value, growth, macro) diverge on the same stock before making a decision.
- When a user wants to check if a hot-money trader would even consider a stock based on its style range.
- When a user needs a structured, JSON-based panel output for further analysis or report generation.
What are this skill's strengths and limitations?
- Covers 9 distinct investment schools with 65 simulated investors, offering diverse perspectives.
- Each investor has unique rules and speaking style, making outputs engaging and persona-accurate.
- Hot-money investors are style-filtered, avoiding irrelevant assessments for out-of-range stocks.
- Structured JSON output (Pydantic Signal) is machine-readable and integrates well with reporting pipelines.
- Requires pre-generated dimensions.json and raw_data.json, so it cannot run standalone without the larger UZI-Skill pipeline.
- Signals are AI-simulated and do not represent the real investors' views.
- No independent test suite is provided for this skill; robustness depends on upstream data quality.
- For non-hot-trend stocks, the hot-money group may largely return 'unsuitable', which users must interpret correctly.
How do you install this skill?
This skill is part of the UZI-Skill repository (MIT licensed). Clone the repository or install via your agent's marketplace, ensuring the skills/investor-panel directory is accessible. Full installation steps are in the repo README.
How do you use this skill?
First, ensure input data exists: run the UZI-Skill analysis to generate .cache/{ticker}/dimensions.json and raw_data.json. Then, in your agent, trigger the panel with a prompt like 'Run the investor panel on 600519' or 'What does the 65-investor jury say?'. The agent will execute the skill's scripts and produce panel.json. To focus on one school, use the CLI: python run.py <ticker> --school <A-I>.
How does this skill compare with similar options?
The skill borrows the Pydantic Signal format from virattt/ai-hedge-fund, which is a comparable alternative. ai-hedge-fund also generates signals from multiple AI investors but may focus more on US equities and use different data sources.