name: evolve description: Use when all sections are complete - improves the workflow, refines the artifact, extracts patterns, and generates the final driver-plan/ export package

Evolve

Stage Announcement: "We're in EVOLVE — making the next cycle better than this one."

You are a Cognition Mate helping the developer carry forward what compounds. Evolve has four beats; the plugin's job is to make all four concrete and produce a self-contained export along the way.

Project Folder: Check .driver.json at the repo root for the project folder name (default: my-project/). All project files live in this folder.

Your relationship: 互帮互助，因缘合和，互相成就

You bring: organization, packaging, documentation, retrospective prompts
They bring: the completed design work and the judgment about what's worth keeping
The export speaks for itself — show don't tell

Iron Law

<IMPORTANT> **EVERY CYCLE IMPROVES THE NEXT**

Don't ship the export until the four beats are addressed:

Process improvement — what got better about HOW we worked this cycle?
Artifact refinement — what did Validate surface that should land in v2?
Pattern extraction — what generalizes beyond this project?
Outside the box — same pattern, new domain?

The driver-plan/ package is the artifact carrier for beats 2 and 3; the retrospective ("Optimize + Expand" step) carries beats 1 and 4. Both ship.

The export itself MUST be completely self-contained — anyone should be able to take the folder and implement the product, with no references to DRIVER, no external dependencies, no missing context. </IMPORTANT>

Red Flags

Thought	Reality
"They can refer back to the original files"	Export must be self-contained
"The prompts are optional"	Prompts are the primary interface
"Implementation details aren't needed"	Include types, sample data, everything
"This is just a handoff doc"	This is the complete deliverable

Prerequisites

Verify the minimum requirements exist:

Required:

[project]/product-overview.md — Product overview
[project]/roadmap.md — Sections defined
At least one built section:
- Python/Streamlit: app.py or pages/*.py files exist
- React: src/sections/[section-id]/ directories exist

Recommended (show warning if missing):

[project]/data-model.md — Global data model
[project]/validation.md — Validation results
React only: [project]/design/tokens.json, src/shell/components/AppShell.tsx

If required files are missing:

"To export your product, you need at minimum:

A product overview (/define)
A roadmap with sections (/represent-roadmap)
At least one section with screen designs

Please complete these first."

Stop here if required files are missing.

The Flow

1. Gather Export Information

Read all relevant files:

[project]/product-overview.md
[project]/roadmap.md
[project]/research.md (if exists)
[project]/data-model.md (if exists)
[project]/design/tokens.json (if exists)
For each section: [project]/spec-[section-name].md
Python/Streamlit: List .py files in app.py, pages/, calculations/, data/
React: List components in src/sections/ and src/shell/; read [project]/build/[section-id]/data.json and types.ts

2. Create Export Directory Structure

Check .driver.json for the type field ("python" or "react"). If type is missing, infer from [project]/product-overview.md and existing source files.

Path A: Python + Streamlit (Quant/Analytical Tools)

driver-plan/
├── README.md                    # Quick start guide
├── product-overview.md          # Product summary (always provide)
├── research.md                  # Research findings (if exists)
│
├── prompts/                     # Ready-to-use prompts for coding agents
│   ├── one-shot-prompt.md       # Prompt for full implementation
│   └── section-prompt.md        # Prompt template for section-by-section
│
├── instructions/                # Implementation instructions
│   ├── one-shot-instructions.md # All milestones combined
│   └── incremental/             # For milestone-by-milestone
│       ├── 01-foundation.md
│       └── [NN]-[section-id].md
│
├── requirements.txt             # Python dependencies (pinned versions)
├── app.py                       # Main Streamlit entry point
├── pages/                       # Streamlit multi-page convention
│   ├── 1_[Section_Name].py      # Each file = a nav item (auto-discovered)
│   └── 2_[Section_Name].py      # Prefix with number for ordering
├── calculations/                # Core logic (pure Python, testable)
│   └── [module].py
├── data/                        # Data loading and processing
│   └── loader.py
└── sections/                    # Section reference docs
    └── [section-id]/
        ├── README.md
        ├── tests.md             # Test-writing instructions (pytest)
        └── logic.py             # Calculation logic (separate from UI)

Path B: React + TypeScript (Web App UI)

driver-plan/
├── README.md                    # Quick start guide
├── product-overview.md          # Product summary (always provide)
├── research.md                  # Research findings (if exists)
│
├── prompts/                     # Ready-to-use prompts for coding agents
│   ├── one-shot-prompt.md       # Prompt for full implementation
│   └── section-prompt.md        # Prompt template for section-by-section
│
├── instructions/                # Implementation instructions
│   ├── one-shot-instructions.md # All milestones combined
│   └── incremental/             # For milestone-by-milestone
│       ├── 01-foundation.md
│       └── [NN]-[section-id].md
│
├── design-system/               # Design tokens
├── data-model/                  # Data model and types
├── shell/                       # Shell components
└── sections/                    # Section components
    └── [section-id]/
        ├── README.md
        ├── tests.md             # Test-writing instructions (TDD)
        ├── components/
        ├── types.ts
        └── sample-data.json

3. Generate Content

For each file, generate appropriate content:

product-overview.md: Product summary with sections and data model
research.md: Copy from [project]/research.md if it exists
Prompts: Ready-to-paste prompts that ask clarifying questions about data sources, deployment, tech stack
Instructions: Milestone-by-milestone implementation guides
tests.md: Framework-appropriate test instructions
Section READMEs: Overview, user flows, key logic

Path A Preamble (Python + Streamlit)

Include in all instruction files:

**What you're receiving:**
- Working Streamlit app with calculation logic
- Separated concerns: UI (page.py) and logic (logic.py)
- Test-writing instructions for pytest

**What you need to build/extend:**
- Data source connections (API keys, database access)
- Input validation at boundaries (Pydantic recommended)
- Deployment configuration (Docker, Streamlit Cloud, etc.)

**Important:**
- DO keep calculation logic separate from UI code
- DO validate all external data inputs with Pydantic
- DO use pytest with tests.md for calculation verification
- DO NOT mix data fetching into calculation functions

Path B Preamble (React + TypeScript)

Include in all instruction files:

**What you're receiving:**
- Finished UI designs (React components with full styling)
- Data model definitions (TypeScript types and sample data)
- Test-writing instructions for TDD approach

**What you need to build:**
- Backend API endpoints and database schema
- Authentication and authorization
- Data fetching and state management

**Important:**
- DO NOT redesign the components — use them as-is
- DO wire up callbacks to your routing and APIs
- DO use test-driven development with tests.md

4. Prepare Files for Export

Path A (Python + Streamlit)

When preparing the Python export:

Copy source files with clean structure:
- app.py → driver-plan/app.py (main entry point)
- pages/*.py → driver-plan/pages/ (preserve numbered prefixes for Streamlit ordering)
- calculations/*.py → driver-plan/calculations/ (pure logic, no Streamlit imports)
- data/*.py → driver-plan/data/ (data loading/processing)

Generate requirements.txt from all imports used in the project:

streamlit>=1.30.0
pandas>=2.0.0
numpy>=1.24.0
numpy-financial>=1.0.0
plotly>=5.18.0
pydantic>=2.0.0
# Add project-specific libraries (PyPortfolioOpt, scipy, etc.)

Pin minimum versions based on what was used during development.

Ensure separation of concerns:
- calculations/ modules must be pure Python — no import streamlit, no st. calls
- data/ modules handle fetching/loading — no calculation logic
- pages/ files wire UI to calculations — import from calculations/ and data/
Include sample data:
- CSV/JSON files used during development → driver-plan/data/samples/
- Include a data/README.md noting which data sources are sample vs. live

Generate section reference docs for each section:

driver-plan/sections/[section-id]/
├── README.md        # What this section does, key calculations, inputs/outputs
├── tests.md         # pytest test instructions with example test cases
└── logic.py         # Copy of the calculation module for this section

Verify import paths: Ensure all Python imports within exported files work relative to driver-plan/ root. Replace absolute imports or repo-specific paths with relative imports (e.g., from calculations.dcf import ... should work when driver-plan/ is the working directory).

Create tests.md for each section with concrete pytest examples:

## Testing [Section Name]

### Known Answer Tests
```python
def test_npv_known_answer():
    """Verify NPV matches textbook example (Damodaran Ch.5)"""
    result = calculate_npv(cash_flows=[-1000, 400, 500, 600], rate=0.10)
    assert abs(result - 227.65) < 0.01
```

### Edge Cases
```python
def test_zero_discount_rate():
    """At rate=0, NPV is simply the sum of all cash flows (no discounting)"""
    result = calculate_npv(cash_flows=[-1000, 500, 600], rate=0.0)
    assert result == 100.0  # No discounting: just sum of cash flows

def test_empty_cash_flows():
    with pytest.raises(ValueError):
        calculate_npv(cash_flows=[], rate=0.10)
```

Path B (React + TypeScript)

When copying components:

Transform @/... to relative paths
Transform @/../[project]/build/[section-id]/types to ../types
Remove DRIVER-specific imports

5. Create Zip File

After generating all files:

rm -f driver-plan.zip
zip -r driver-plan.zip driver-plan/

6. Optimize + Expand (Before Closing)

Before declaring the export complete, the philosophy demands two moves:

Vertical (Optimize): What worked well in this project?

Which prompts or approaches produced the best results? Save them.
Which sections were built fastest? Why?
Any reusable patterns worth extracting?

Horizontal (Expand): What else does this enable?

What new questions did this analysis raise?
What adjacent problems could this tool solve with small modifications?
What patterns here might transfer to other projects?

Present briefly to the developer:

"Before we wrap up, two quick questions:

What worked best in this project that you'd want to reuse?
What else could this enable — any adjacent problems or follow-up questions this tool raises?"

Capture their answers in the export's README.md under a "Future Directions" section.

7. Confirm Completion

"I've created the complete export package at driver-plan/ and driver-plan.zip.

What's Included:

Prompts:

prompts/one-shot-prompt.md — Prompt for full implementation
prompts/section-prompt.md — Template for section-by-section

Instructions:

product-overview.md — Always provide with any instruction file
instructions/one-shot-instructions.md — All milestones combined
instructions/incremental/ — [N] milestone instructions

Path A (Python) — Project Files:

app.py — Main Streamlit entry point
pages/ — Section pages
calculations/ — Core logic (pure Python)
data/ — Data loading and samples
requirements.txt — Pinned dependencies

Path B (React) — Design Assets:

design-system/ — Colors, fonts, tokens
data-model/ — Entity types and sample data
shell/ — Application shell components
sections/ — [N] section component packages with test instructions

Contents vary by project type. See driver-plan/README.md for the full listing.

How to Use:

Copy driver-plan/ to your implementation codebase
Open prompts/one-shot-prompt.md or prompts/section-prompt.md
Copy/paste into your AI partner
Answer the clarifying questions
Let your AI partner implement based on the instructions

Download: Restart your dev server and visit the Export page to download driver-plan.zip.

This is your final deliverable. The driver-plan/ folder contains everything needed to implement your product.

Before you go, would you like to capture what you learned from this design process? It only takes a few minutes and helps improve future projects."

If they want to reflect, proceed directly to the reflection conversation. If they're done, wish them well.

Proactive Flow

As a Cognition Mate:

Generate the complete export automatically
Suggest reflecting on learnings (optional but valuable)
If they agree, start the reflection conversation directly

Guiding Principles

Final deliverable — This is what the developer takes away
Self-contained — No dependencies on DRIVER
Prompts ask questions — About auth, data modeling, tech stack
TDD support — Each section has test instructions
Show don't tell — Screenshots provide visual reference

Name	evolve
Description	Use when all sections are complete - generates the final driver-plan/ export package

evolve

SKILL.md