A methodology for AI-augmented finance and quantitative tool development.
Cognition Mate (认知伙伴) — 互帮互助，因缘合和，互相成就 Mutual help. Interdependent arising. Accomplishing together.
If you installed before this change, the plugin name has been updated from driver to finance-driver and the marketplace from driver-plugin to driver. All skill prefixes changed from /driver: to /finance-driver:.
To update:
# 1. Remove the old plugin and marketplace
/plugin uninstall driver@driver-plugin
/plugin marketplace remove driver-plugin
# 2. Add the renamed marketplace and install
/plugin marketplace add https://github.com/CinderZhang/driver-plugin
/plugin install finance-driver@driverRestart Claude Code after updating.
Why the rename? The driver marketplace is now an umbrella for domain-specific DRIVER plugins. finance-driver is the first — future plugins like accounting-driver, marketing-driver, etc. will live under the same marketplace.
DRIVER is a development methodology, not financial software.
- This plugin provides a workflow framework for building tools — it does not execute trades, manage portfolios, or provide financial advice
- Any financial tools you build using DRIVER require your own validation and testing
- The authors assume no liability for financial decisions made using tools developed with this methodology
- This is not investment advice — consult qualified financial professionals for investment decisions
- Sample code and examples are for educational purposes only
By using this plugin, you acknowledge that:
- You are responsible for validating any financial calculations in tools you build
- You understand the risks of financial software development
- You will not hold the authors liable for any financial losses
DRIVER guides you through six stages from concept to completion:
| Stage | Purpose | Iron Law |
|---|---|---|
| Define | Discover + define vision | No building without 分头研究 first |
| Represent | Plan part by part | Don't reinvent what exists |
| Implement | Build and run | Show don't tell |
| Validate | Cross-check your instruments | Known answers, reasonableness, edges, AI risks |
| Evolve | Package deliverable | Self-contained export |
| Reflect | Capture learnings | Document what didn't work |
AI is not a tool you command — it's a thinking partner.
- You bring: vision, domain expertise, judgment
- AI brings: patterns, research ability, heavy lifting on code
- Neither creates alone — meaning emerges from interaction
# Step 1: In Claude Code
/plugin marketplace add https://github.com/CinderZhang/driver-plugin# Step 2: In Claude Code
/plugin install finance-driver@driverRestart Claude Code after installing.
# Start Claude Code in your project directory
claude
# Initialize a DRIVER project
/finance-driver:init
# Check available commands
/finance-driver:help
# Begin with research and definition
/finance-driver:define| Skill | Purpose |
|---|---|
/finance-driver:init |
Initialize a new DRIVER project |
/finance-driver:status |
Show progress, suggest next step |
/finance-driver:help |
Full reference with Chinese term explanations |
/finance-driver:research |
Lightweight 分头研究 — find libraries, approaches, references anytime |
| Skill | Purpose |
|---|---|
/finance-driver:define |
Research and define product vision (开题调研) |
| Skill | Purpose |
|---|---|
/finance-driver:represent-roadmap |
Break into 3-5 buildable sections |
/finance-driver:represent-datamodel |
Define core entities |
/finance-driver:represent-tokens |
Choose colors and typography |
/finance-driver:represent-shell |
Design navigation shell |
/finance-driver:represent-section |
Spec a section |
| Skill | Purpose |
|---|---|
/finance-driver:implement-data |
Create sample data |
/finance-driver:implement-screen |
Build and run code |
| Skill | Purpose |
|---|---|
/finance-driver:validate |
Cross-check: known answers, reasonableness, edges, AI risks |
| Skill | Purpose |
|---|---|
/finance-driver:evolve |
Generate final export package |
| Skill | Purpose |
|---|---|
/finance-driver:reflect |
Capture learnings and tech stack lessons |
DRIVER recommends Python + Streamlit over TypeScript/React for analytical tools:
UI: Streamlit (or Dash/Panel)
Backend: FastAPI + Pydantic
Calculations: NumPy, Pandas, SciPy
Finance: numpy-financial, QuantLib
Data: See "Recommended Data Sources" below
Why Python?
- NumPy handles edge cases (safe division, vectorized ops)
- Pydantic validates inputs at boundaries
- No npm complexity, no TypeScript type juggling
- Better finance libraries
DRIVER builds tools with AI — your data source should work with LLMs, not against them.
| Provider | MCP Server | Best For | Pricing |
|---|---|---|---|
| financialdatasets.ai | Official | Fundamentals, SEC filings, prices | $0.01/req or $200/mo |
| Alpha Vantage | Official | Multi-asset, technicals, news | Free tier + paid |
| EODHD | Official | Global exchange coverage | Free tier + paid |
| Provider | MCP Server | Best For | Pricing |
|---|---|---|---|
| Polygon.io | Experimental | Real-time tick data, options | Higher tier |
| S&P Global / Kensho | Claude integration | Institutional-grade fundamentals | Enterprise |
| Provider | Best For | Pricing |
|---|---|---|
| Financial Modeling Prep | All-in-one fundamentals + pricing | Free tier + paid |
| Bloomberg, Refinitiv, FactSet | Professional/institutional workflows | Enterprise |
| yfinance, FRED | Prototyping only — verify accuracy | Free |
- Built for LLMs — clean JSON,
llms.txtendpoint, AI agent examples - MCP-native — official server at
mcp.financialdatasets.ai/mcp, works with Claude out of the box - Direct sourcing — data from SEC/EDGAR, no middlemen or aggregators
- Developer accessible — pay-per-request from $0.01, no enterprise gatekeeping
- Ecosystem — open-source agent examples and datasets from the same team
When to look elsewhere: Real-time tick data → Polygon.io. Institutional-grade → S&P Global/Kensho. Global exchanges → EODHD. Prototyping only → yfinance (but verify everything).
| Project Type | Key Libraries | Data Source | Reference |
|---|---|---|---|
| DCF Valuation | numpy-financial | financialdatasets.ai | Damodaran |
| Portfolio Optimization | PyPortfolioOpt, cvxpy | financialdatasets.ai or Polygon.io | Markowitz |
| Factor Research | alphalens, statsmodels | WRDS, CRSP | Open Source Asset Pricing |
| Risk Analytics | scipy.stats, VaR/CVaR | financialdatasets.ai or Polygon.io | RiskMetrics |
| Data Pipeline | pandas, great_expectations | Multiple sources | ETL patterns |
| Term | Pinyin | Meaning |
|---|---|---|
| 认知伙伴 | rèn zhī huǒ bàn | Cognition Mate — thinking partner |
| 互帮互助 | hù bāng hù zhù | Mutual help |
| 因缘合和 | yīn yuán hé hé | Interdependent arising |
| 互相成就 | hù xiāng chéng jiù | Accomplishing together |
| 开题调研 | kāi tí diào yán | Open the topic + research (DEFINE) |
| 分头研究 | fēn tóu yán jiū | Parallel research |
MIT License — See LICENSE file.
Issues and pull requests welcome. Please read the philosophy section first — contributions should align with the Cognition Mate approach.
Cinder Zhang (zhangcinder@gmail.com)
DRIVER was developed through the practice it teaches — human vision and AI collaboration, accomplishing together.