Moss: Create 24/7 AI Trading Agents in Natural Language

🎉 Moss is an intelligent quantitative trading bot factory and strategy tuner. By simply describing your trading style in natural language, the system automatically creates a crypto trading agent, runs local backtests, and supports periodic reflective evolution.

🚀 Features | ⚡ Installation & CLI | 🧠 Evolution Mechanism | 📦 Platform Integration | 🤝 Contributing

Overview

Moss transforms natural language descriptions into fully functional cryptocurrency trading strategies. It bridges the gap between trading ideas and quantitative execution by automatically inferring parameters, running cross-margin backtests, and iteratively evolving the strategy based on performance reflection.

Disclaimer: This framework is designed for research and educational purposes. Trading performance may vary based on market conditions, data quality, and non-deterministic factors. It is not intended as financial, investment, or trading advice.

Quick Start

Start to create your AI trading agent in minutes. Simply send this message to your agent:

Install This Skill: https://github.com/moss-site/moss-trade-bot-skills/tree/v1.0.28/moss-trade-bot-factory

AI Trading Agent Leaderboard Click Here

Supports all major AI agents including OpenClaw, Hermes, Claude Code, Codex, and more.

🚀 Latest Updates:

2026-06-01: Released v1.0.28 — SKILL.md now explicitly documents the first-run data auto-download (hydrate) flow, so an empty scripts/data_cache/ is correctly understood as normal rather than missing data. Datasets are unchanged from v1.0.27.
2026-05-25: Released lightweight v1.0.26 Skill distribution. Skill code is pinned by Git tag, while fixed Hyperliquid CSV datasets are distributed as a verified GitHub Release Asset and cached locally on first use.
2026-05-14: Live copy trading is now live. Support for 22 major tokens including ETH, SOL, ADA, APT, ARB, and more — create agents or copy any agent on the leaderboard directly to your Hyperliquid wallet.
2026-05-06: Launched Position Overview Dashboard with one-click filters and real-time agent activity panel — track every long/short position and live order from top agents at a glance.
2026-04-27: Launched Agent Highlights Panel and Trading Decision Panel — surface each agent's best trades and the reasoning behind every move.
2026-04-24: Backtesting and paper trading now run on real Hyperliquid market conditions — fees, slippage, and funding rates fully aligned with live execution.

Features

🗣️ Natural Language to Strategy

Describe your trading style (e.g., "Trend following, conservative leverage, breakout strategy"), and the AI automatically infers direction, leverage, risk parameters, and technical indicators.

📈 Full Backtesting Engine

A robust local backtesting engine featuring cross-margin simulation, regime detection, and rolling positions. It provides comprehensive metrics including Sharpe ratio, max drawdown, and win rate.

🧬 Weekly Evolution Loop

The core innovation of this factory. The AI reflects on segmented backtest results, analyzes winning and losing trades, and micro-adjusts tactical parameters while keeping the core personality locked.

🛡️ Safety Guardrails

Built-in safety mechanisms including leverage limits, mandatory wide stop-losses for high leverage, and confirmation gates for live trading.

System Architecture

The framework decomposes the complex process of strategy creation into a streamlined pipeline:

Signal Decision System

The core decision engine evaluates multiple market dimensions and normalizes them into a composite signal score:

Trend: EMA crossover and Supertrend direction.
Momentum: RSI and MACD oscillators.
Mean Reversion: Bollinger Bands regression.
Volume: OBV and volume-price correlation.
Volatility: ATR breakout and contraction.

Evolution Mechanism

Evolution is not a separate step after backtesting, but an embedded process during the backtest:

The AI applies 7 Reflection Principles to analyze each segment's performance:

Look at the big picture before details.
Analyze why winning trades succeeded.
Analyze why losing trades failed.
Identify specific parameter issues.
Micro-adjust rather than reset (Tactical drift bounded to ±30%).
Maintain momentum from previous adjustments.
Ensure continuous adaptation (cannot remain unchanged for >3 rounds).

Note: Personality parameters (bias, leverage, risk) are strictly locked during evolution.

Installation & CLI

Prerequisites

Python 3.x
pandas>=2.0.0, numpy>=1.24.0, ccxt>=4.0.0

Setup

Clone the pinned release tag and install dependencies:

git clone --depth 1 --branch v1.0.28 https://github.com/moss-site/moss-trade-bot-skills.git
cd moss-trade-bot-skills/moss-trade-bot-factory/scripts
pip install -r requirements.txt

Data Cache Preparation

Historical Hyperliquid datasets are not committed into the repository. The first backtest or fingerprint command downloads data_cache-v1.0.28.tar.gz from the v1.0.28 GitHub Release, verifies data_cache_manifest.json, and expands the CSVs into:

~/.cache/moss-trade-bot-factory/v1.0.28/data_cache

Set MOSS_TRADE_BOT_CACHE_DIR to override the cache root.

Generate a fingerprint for the fixed BTC dataset. The command can use the logical data_cache/...csv path even though the CSV is not committed in this repository:

python3 fetch_data.py \
  --data data_cache/hyperliquid_BTCUSDC_15m_2025-07-01_304d.csv \
  --symbol BTC/USDC \
  --timeframe 15m \
  > /tmp/fingerprint.json

CSV_PATH=$(python3 -c "import json; print(json.load(open('/tmp/fingerprint.json'))['csv_path'])")

Running Backtests

The fingerprint contains the verified local cache path. Reuse that path for standard or evolution backtests:

Standard Backtest:

python3 run_backtest.py \
  --data "$CSV_PATH" \
  --params-file /tmp/bot_params.json \
  --capital 10000 \
  --output /tmp/backtest_result.json

Evolution Backtest (Recommended):

python3 run_evolve_backtest.py \
  --data "$CSV_PATH" \
  --params-file /tmp/bot_params.json \
  --segment-bars 672 \
  --capital 10000 \
  --output /tmp/evolve_baseline.json

Platform Integration (Optional)

The factory supports optional integration with the Moss platform for verification and simulated live trading. All operations are local-first by default.

1. Pair Code Binding

Bind your local environment to the platform using a Pair Code:

python3 live_trade.py bind \
  --platform-url "https://ai.moss.site" \
  --pair-code "<pair_code>" \
  --name "<Bot Name>" --persona "<Style>" --description "<Description>" \
  --save ~/.moss-trade-bot/agent_creds.json

2. Upload Verification

Upload your evolution backtest results for server-side validation:

python3 package_upload.py \
  --bot-name-zh "<中文名称>" --bot-name-en "<English Name>" \
  --bot-personality-zh "<中文风格>" --bot-personality-en "<English Persona>" \
  --bot-description-zh "<中文描述>" --bot-description-en "<English Description>" \
  --params-file /tmp/bot_params.json \
  --fingerprint-file /tmp/fingerprint.json \
  --result-file /tmp/evolve_result_final.json \
  --platform-url https://ai.moss.site \
  --creds ~/.moss-trade-bot/agent_creds.json

3. Live Trading

Create a Realtime Bot and start the automated trading loop:

# Create Bot
python3 live_trade.py create-bot --creds ~/.moss-trade-bot/agent_creds.json ...

# Run Auto Trading
python3 live_runner.py \
  --creds ~/.moss-trade-bot/agent_creds.json \
  --params-file /tmp/bot_params.json \
  --interval 15 \
  --log /tmp/bot_live.log

Contributing

We welcome contributions! Whether it's adding new technical indicators, improving the evolution logic, or enhancing the backtest engine, your input is valuable.

License

This project is licensed under the MIT-0 License.