Strategy Writer

Auto-invokes to create QuantConnect Lean trading strategies.

Project Structure

Each strategy project in algorithms/{name}/ contains:

code

algorithms/{name}/
├── main.py           # Strategy implementation
├── spec.md           # Strategy specification (goals, rules, parameters)
├── results.md        # Backtest results summary (updated after each run)
└── research.ipynb    # Research notebook for analysis

Keep all files in sync - update spec.md when strategy changes, update results.md after backtests.

Workflow

1. Parse Requirements

Parse the user's request for:

•Strategy type: momentum, mean-reversion, breakout, trend-following, pairs
•Asset class: crypto (preferred), equity, forex, futures
•Timeframe: intraday, daily, weekly
•Indicators: specific indicators or signals mentioned
•Risk parameters: position size, stops, targets mentioned

2. Ensure Data Exists

Prefer Binance crypto data - the lp download tool fetches it directly.

Check Available Data

bash

ls /Lean/Data/crypto/binance/         # Binance crypto (preferred)
ls /Lean/Data/crypto/coinbase/        # Coinbase crypto
ls /Lean/Data/equity/usa/daily/       # US Equity (SPY, AAPL, etc.)

Download Crypto Data (Binance)

Use lp download to fetch Binance data in Lean format:

bash

# Basic - last 10 days, 1m and 1h intervals
lp download --symbols BTCUSDT

# Multiple symbols, 30 days
lp download --symbols BTCUSDT ETHUSDT --days 30

# Specific date range
lp download --symbols BTCUSDT --start-date 2024-01-01 --end-date 2024-06-01

# Daily data only
lp download --symbols BTCUSDT --intervals 1d

# Futures data (perpetuals)
lp download --symbols BTCUSDT --account-type usdt_future

# All intervals
lp download --symbols BTCUSDT --intervals 1m 5m 15m 1h 4h 1d

Options:

•--symbols - Trading pairs (BTCUSDT, ETHUSDT, SOLUSDT, etc.)
•--intervals - 1s, 1m, 5m, 15m, 30m, 1h, 4h, 1d
•--days - Days of history (default: 10)
•--start-date / --end-date - Specific date range (YYYY-MM-DD)
•--account-type - spot (default), usdt_future, coin_future

Data is written to /Lean/Data/crypto/binance/.

3. Load Relevant Knowledge

Read all files from .claude/skills/strategy-writer/knowledge/:

bash

# List and read all knowledge files
ls .claude/skills/strategy-writer/knowledge/

For each knowledge file:

•Filter entries by tags matching strategy type, asset class, timeframe
•Prioritize high-confidence entries (confidence > 0.7)
•Load both patterns (to apply) and anti-patterns (to avoid)

4. Design Strategy

Based on requirements and knowledge:

•Select entry signal approach
•Design exit rules
•Define position sizing
•Add risk management (stops, limits)

5. Create Project

bash

lp create algorithms/{name}
# Or from example:
lp create algorithms/{name} --from {ExampleAlgorithm}

6. Generate Code

Create the strategy following Lean conventions:

•Location: algorithms/{strategy_name}/main.py
•Import: from AlgorithmImports import *
•Class inherits from QCAlgorithm
•Implement initialize() and on_data()
•Use snake_case method names

7. Write Strategy Files

main.py - Strategy Implementation

python

from AlgorithmImports import *


class StrategyName(QCAlgorithm):
    """
    Strategy description.

    Type: {type}
    Entry: {entry_logic}
    Exit: {exit_logic}
    Knowledge: [{applied_ids}]
    """

    def initialize(self):
        self.set_start_date(2024, 1, 1)
        self.set_end_date(2024, 6, 1)
        self.set_cash(100_000)

        # Crypto (Binance) - preferred
        self.symbol = self.add_crypto("BTCUSDT", Resolution.HOUR, Market.BINANCE).symbol
        # Equity alternative:
        # self.symbol = self.add_equity("SPY", Resolution.DAILY).symbol

        # Indicators
        self.sma_fast = self.sma(self.symbol, 10)
        self.sma_slow = self.sma(self.symbol, 30)

        # Risk parameters
        self.position_size = 1.0

        # Warm up
        self.set_warm_up(30)

    def on_data(self, data: Slice):
        if self.is_warming_up:
            return
        if not data.contains_key(self.symbol):
            return
        if not self.sma_fast.is_ready or not self.sma_slow.is_ready:
            return

        # Entry logic
        if not self.portfolio.invested:
            if self.sma_fast.current.value > self.sma_slow.current.value:
                self.set_holdings(self.symbol, self.position_size)

        # Exit logic
        else:
            if self.sma_fast.current.value < self.sma_slow.current.value:
                self.liquidate(self.symbol)

spec.md - Strategy Specification

markdown

# {Strategy Name}

## Overview
Brief description of the strategy concept.

## Hypothesis
What market behavior does this strategy exploit?

## Rules

### Entry Conditions
- Condition 1
- Condition 2

### Exit Conditions
- Condition 1
- Condition 2

### Risk Management
- Position size: X%
- Stop loss: X%
- Take profit: X%

## Parameters
| Parameter | Value | Description |
|-----------|-------|-------------|
| fast_period | 10 | Fast MA period |
| slow_period | 30 | Slow MA period |

## Data Requirements
- Symbol: BTCUSDT
- Resolution: Hour
- Date range: 2024-01-01 to 2024-06-01

## Knowledge Applied
- [ID] Description of applied pattern

results.md - Backtest Results

markdown

# Backtest Results: {Strategy Name}

## Latest Run
- **Date**: {timestamp}
- **Period**: {start} to {end}

## Performance Metrics
| Metric | Value |
|--------|-------|
| Total Return | X% |
| Sharpe Ratio | X.XX |
| Max Drawdown | X% |
| Win Rate | X% |
| Total Trades | N |

## Summary
Brief analysis of the results.

## History
| Date | Sharpe | Return | Drawdown | Notes |
|------|--------|--------|----------|-------|
| ... | ... | ... | ... | ... |

8. Run Backtest

bash

lp backtest algorithms/{name}

9. Update Results

After backtest completes:

•Read results/{name}/{timestamp}/main-summary.json
•Update algorithms/{name}/results.md with new metrics
•Add entry to history table

10. Output

Provide:

•The complete strategy code
•Brief explanation of how knowledge was applied
•Suggested backtest command: lp backtest algorithms/{strategy_name}
•Reminder to run /evolve after viewing results

Browse Examples

bash

lp browse              # List all ~500 examples
lp browse macd         # Search for MACD examples
cat /Lean/Algorithm.Python/MACDTrendAlgorithm.py  # Read example

Lean API Quick Reference

Adding Securities:

python

self.add_crypto("BTCUSDT", Resolution.HOUR, Market.BINANCE)  # Preferred
self.add_equity("SPY", Resolution.DAILY)
self.add_forex("EURUSD", Resolution.HOUR, Market.OANDA)

Resolutions: Resolution.TICK, SECOND, MINUTE, HOUR, DAILY

Built-in Indicators:

python

self.sma(symbol, period)      # Simple Moving Average
self.ema(symbol, period)      # Exponential MA
self.rsi(symbol, period)      # Relative Strength Index
self.macd(symbol, fast, slow, signal)  # MACD
self.bb(symbol, period, k)    # Bollinger Bands
self.atr(symbol, period)      # Average True Range
self.adx(symbol, period)      # Average Directional Index

Position Management:

python

self.set_holdings(symbol, 1.0)   # 100% long
self.set_holdings(symbol, -0.5)  # 50% short
self.liquidate(symbol)           # Close position
self.portfolio.invested          # Bool: has positions?

File Maintenance

After strategy changes:

•Update spec.md with new parameters/rules
•Update docstring in main.py

After each backtest:

•Parse results/{name}/{latest}/main-summary.json
•Update results.md with metrics and history entry