Skillget

ai-trading-agent

agent
Guvenlik Denetimi
Uyari
Health Uyari
  • No license — Repository has no license file
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Gecti
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Gecti
  • Permissions — No dangerous permissions requested
Purpose
This project is a multi-symbol autonomous trading platform that uses Claude AI to analyze and execute trades on MetaTrader 5 across assets like GOLD, OIL, and BTC. It features a robust full-stack architecture with a Next.js frontend and a FastAPI backend.

Security Assessment
Overall risk: Medium. The system requires highly sensitive data to function, including MetaTrader 5 credentials, Anthropic API keys, and direct access to your financial brokerage account. The code scan (covering 12 files) found no dangerous patterns, hardcoded secrets, or malicious shell executions. It also implements strong internal defenses, such as AES-256-GCM encryption for its secrets vault and non-bypassable trading guardrails. However, because it makes external network requests to AI APIs and a Windows VPS to execute real financial transactions, it inherently handles high-stakes data.

Quality Assessment
The project is under active development, with its most recent push occurring today. Despite this, it has very low community visibility with only 5 stars, meaning it has not been broadly tested or peer-reviewed by a wide audience. Additionally, the repository lacks a defined license. This is a critical oversight for a financial tool, as it means no clear legal terms govern its usage, liability, or whether the code can be safely modified and distributed.

Verdict
Use with caution — the code itself appears clean and well-architected, but the combination of unlicensed software, low community trust, and autonomous financial execution introduces significant operational risk.
SUMMARY

Multi-symbol AI trading agent — Claude AI + FastAPI + Next.js + MetaTrader 5

README.md

AI Trading Agent

Multi-symbol autonomous trading platform powered by Claude AI agents.
Trades GOLD (XAUUSD), OILCash, BTCUSD, USDJPY via MetaTrader 5.

Architecture

Frontend (Next.js 16, Vercel)
    |
    | HTTPS + WebSocket
    v
Backend (FastAPI, Railway)
    |-- Auth Layer (Passkey WebAuthn + JWT)
    |-- Secrets Vault (AES-256-GCM encrypted)
    |-- Runner Manager (process/Docker sandbox)
    |   |-- Job Queue (Redis + DB)
    |   |-- Heartbeat Monitor
    |   '-- Agent Entrypoint
    |       |-- MCP Tool Server (36 tools)
    |       |-- Guardrails (non-bypassable limits)
    |       '-- Multi-Agent Pipeline
    |           |-- Reflector (Haiku) -- past trade review
    |           |-- Technical Analyst (Haiku) -- indicators
    |           |-- Fundamental Analyst (Haiku) -- sentiment
    |           |-- Risk Analyst (Haiku) -- portfolio risk
    |           '-- Orchestrator (Sonnet) -- final decision
    |-- Strategy Engine (5 strategies + ensemble)
    |-- ML Models (LightGBM per-symbol)
    |-- PostgreSQL + Redis
    '----HTTP----> Windows VPS
                   '-- MT5 Bridge + MetaTrader 5

Tech Stack

Layer Tech
Backend FastAPI 0.115, SQLAlchemy 2.0 (async), asyncpg, Redis, APScheduler
Frontend Next.js 16, React 19, Tailwind 4, Zustand, recharts
AI Agent Anthropic SDK (Claude Sonnet + Haiku), MCP tools, guardrails
ML LightGBM, scikit-learn, pandas
Auth WebAuthn (Passkey) + JWT httpOnly cookie
Trading MetaTrader 5 via HTTP Bridge
CI/CD GitHub Actions (ruff, pytest, tsc, build), Railway auto-deploy
DB PostgreSQL 15, Redis 7

Features

  • AI Agent Trading: Claude-powered multi-agent system that analyzes markets and executes trades autonomously
  • Passkey Auth: Passwordless login via WebAuthn (fingerprint/Face ID/YubiKey)
  • Secrets Vault: AES-256-GCM encrypted storage for API keys and tokens
  • Runner Management: Docker sandbox runners with live logs, metrics, and job queue
  • Gradual Rollout: Shadow -> Paper -> Micro-Live -> Live deployment modes
  • Self-Reflection: Agent reviews past trades to improve future decisions
  • Adaptive Strategy: Regime detection (trending/ranging/volatile) with automatic strategy selection
  • Session Memory: Redis-backed context that persists across trading sessions
  • Guardrails: Hard limits on lot size, daily loss, trade frequency (agent cannot bypass)
  • 5 Strategies: EMA Crossover, RSI Filter, Breakout, Mean Reversion, ML Signal + Ensemble
  • ML Models: Per-symbol LightGBM with 40+ features, drift detection, auto-retrain
  • Real-time Dashboard: Trading view, positions, P&L, AI insights, notifications

Local Development

1. Start databases

docker-compose up -d

2. Backend

cd backend
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt
pip install -r requirements-dev.txt  # test/lint deps
cp .env.example .env
alembic upgrade head
uvicorn app.main:app --reload --port 8000

3. MT5 Bridge (Windows VPS only)

cd mt5_bridge
pip install -r requirements.txt
cp .env.example .env  # MT5 credentials
uvicorn main:app --host 0.0.0.0 --port 8001

4. Frontend

cd frontend
npm install
cp .env.example .env.local
npm run dev

5. Run tests

cd backend
python -m pytest tests/ -v --no-cov  # 413 tests

Environment Variables

See backend/.env.example and mt5_bridge/.env.example.

Key variables for production:

  • SECRET_KEY — JWT signing key
  • VAULT_MASTER_KEY — Secrets vault encryption key
  • WEBAUTHN_RP_ID / WEBAUTHN_ORIGIN — Passkey config
  • CLAUDE_OAUTH_TOKEN — Claude Max subscription token (stored in Vault)
  • ROLLOUT_MODE — shadow / paper / micro / live
  • AGENT_MODE — single (Phase C) or multi (Phase D multi-agent)

Project Structure

backend/
  app/
    api/routes/      # REST endpoints (60+)
    bot/             # Trading engine, scheduler, health monitor
    strategy/        # 5 strategies + ensemble + regime detection
    risk/            # Risk manager, circuit breaker, correlation
    ml/              # LightGBM trainer, features, drift detection
    ai/              # Claude AI client, sentiment, optimization
    runner/          # Docker sandbox runner system
    middleware/      # Auth middleware
    db/              # SQLAlchemy models, migrations
  tests/             # 413 tests (unit + integration)
frontend/
  app/               # Next.js App Router pages
    dashboard/       # Trading dashboard
    runners/         # Runner management + logs + metrics
    secrets/         # Secrets vault UI
    login/           # Passkey login
  components/        # Shared UI components
  lib/               # API client, WebSocket, utilities
mcp_server/
  server.py          # FastMCP server (36 tools)
  guardrails.py      # Trading guardrails (non-bypassable)
  agent_config.py    # Agent loop + tool dispatch
  system_prompt.md   # Agent system prompt
  tools/             # 11 tool modules
  agents/            # 5 specialist agents + orchestrator
mt5_bridge/          # MetaTrader 5 HTTP bridge (Windows VPS)

Yorumlar (0)

Sonuc bulunamadi