Trading AI Agent Skills Suite 📈🤖

Quantitative & Multi-Market Intelligence Suite for AI Coding Agents

License: MIT Python 3.10+ Cross-Agent: Compatible Google TimesFM CFTC COT Data

Universal Agent Skills Standard: Koleksi kemampuan (skills) kuantitatif dan analisis pasar mandiri yang dapat di-instal ke semua AI coding assistant (Claude Code, OpenAI Codex, Antigravity, OpenCode, Cursor, Windsurf). Mendukung Saham Indonesia (IDX), Kripto, Valuta Asing (Forex), dan Komoditas Berjangka.

🌐 Language Quick Navigation / Pintasan Bahasa

🔥 What's New in v1.3.0 / Apa yang Baru di v1.3.0

Catatan rilis lengkap: WHATS_NEW.md | Release notes: GitHub Releases.

📦 What's New in v1.2.0 Archive

🇮🇩 Bahasa Indonesia

1. Ringkasan Eksekutif

Trading AI Agent Skills Suite bukanlah aplikasi tertutup (black-box SaaS) yang kaku, melainkan kumpulan modular Agent Skills terstandarisasi. Suite ini memberi AI Agent Anda kemampuan penuh untuk mencari dan mengunduh data pasar sendiri, menyadap siaran running trade langsung dari browser web trading pengguna, menganalisis saham menggunakan metode fundamental, teknikal, sentimen berita makro (NFP/CPI/BI-Rate), laporan posisi uang pintar CFTC COT, dan memproyeksikan lintasan harga probabilistik menggunakan model kecerdasan buatan Google Research TimesFM.

Semua sinyal eksekusi tunduk pada protokol Human-in-the-Loop (persetujuan eksplisit pengguna via Terminal CLI, Web Dashboard, atau Bot Telegram), dilengkapi simulasi Paper Trading lokal (SQLite) dengan ekspor otomatis ke jurnal spreadsheet Excel (.xlsx) dan visualisasi Web Chart Mobile-View First.

2. Cara Pemasangan Agent Skills (How to Install)

🌟 1. One-Click Standalone Installers (Paling Direkomendasikan)

Unduh installer mandiri sekali klik langsung dari halaman GitHub Releases:

2. Global Installation for AI Agents (One-Prompt Universal Setup)

Salin instruksi ini langsung ke prompt AI Assistant Anda (Google Antigravity, Claude Code, Cursor, Codex, ChatGPT) untuk memasang skills ini secara global:

ext Please install and register the skills globally from https://huggingface.co/spaces/lensetek/trading-ai-agent-skills so they are accessible across all my projects.

3. Menggunakan Perintah NPX Skills (Local Workspace)

Jika lingkungan Anda mendukung NPX skills manager:

npx skills install https://huggingface.co/spaces/lensetek/trading-ai-agent-skills.git

Atau pasang menggunakan CLI bawaan repositori ini:

# Pasang ke workspace lokal saat ini:
npx trading-ai-agent-skills

# Atau pasang secara global ke direktori agent Antigravity/Gemini:
npx trading-ai-agent-skills --global

💬 Metode B: Menggunakan Prompt Otomatis ke AI Agent

Cukup salin dan kirimkan prompt berikut ke AI Agent Anda (Claude Code, Codex, Antigravity, OpenCode, Cursor, Windsurf):

Prompt untuk AI Agent:
"Tolong pasang dan pelajari kumpulan agent skills dari repositori https://huggingface.co/spaces/lensetek/trading-ai-agent-skills ke dalam workspace ini. Kloning repositori tersebut, baca panduan README.md serta file SKILL.md di dalam direktori skills/, dan siapkan file konfigurasi .env dari .env.example untuk mode paper trading. Selain itu, buat file custom agent untuk OpenCode v2 di .opencode/agents/ dengan mengonversi setiap skills/*/SKILL.md ke format agent OpenCode (termasuk frontmatter description), sambil tetap mempertahankan folder skills/ untuk kompatibilitas lintas-agent."

AI Agent Anda akan langsung mengkloning repo, membaca spesifikasi instruksi tiap sub-skill, dan mengaktifkan kemampuan analisis pasar secara otomatis.

📦 Metode C: Kloning Manual via Git 

# 1. Kloning repositori publik
git clone https://huggingface.co/spaces/lensetek/trading-ai-agent-skills.git
cd "Trading-AI-Agent-Skills"

# 2. Pasang pustaka dependensi Python
pip install -r requirements.txt

# 3. Siapkan file konfigurasi aman
cp .env.example .env

3. Daftar 13 Modul Sub-Skills 

skills/
├── trading-orchestrator/          # Koordinator utama alur analisis & strategi
├── market-data-collector/         # Pengunduh data multi-market (IDX, Yahoo Fin, CCXT)
├── backtest-strategy-optimizer/   # Backtest vektorisasi, Genetic Algorithm, WFO, & Monte Carlo
├── browser-tape-interceptor/      # Sadap running trade web broker via Chrome CDP
├── orderflow-tape-scalper/        # Analisis tape reading, bid/ask depth, bandarmologi
├── timesfm-forecaster/            # AI Time Series Foundation Model Google TimesFM
├── cftc-cot-tracker/              # Pengunduh laporan CFTC COT & jembatan komoditas ke IDX
├── economic-news-trader/          # Kalender NFP, CPI, BI-Rate & pelindung slippage
├── technical-analyst/             # Price action, indikator klasik, SMC/ICT liquidity
├── fundamental-macro-analyst/     # Rasio lapkeu, valuasi emiten & korelasi makro (DXY/Yield)
├── paper-trading-engine/          # Simulator portofolio, trailing stop & ekspor Excel .xlsx
├── mobile-chart-server/           # Web dashboard lokal responsif (Mobile-View First)
└── agent-update-manager/          # Auto-updater 24 jam sekali dari GitHub resmi
Modul Peran & Kemampuan Utama
trading-orchestrator Router pusat yang menerima instruksi bebas, mendistribusikan tugas ke spesialis, dan menyusun laporan trading plan utuh.
market-data-collector Mengambil data harga historis dan intraday untuk Saham IDX, Crypto (Binance/Indodax via CCXT), Forex, dan Komoditas.
backtest-strategy-optimizer Backtesting vektorisasi cepat, optimasi parameter Genetic Algorithm (GA), validasi Walk-Forward (WFA), dan stress test 2.500+ iterasi Monte Carlo.
browser-tape-interceptor Menempelkan (attach) sesi browser pengguna via Chrome DevTools Protocol (CDP) untuk menangkap WebSocket running trade tanpa menyentuh password Anda.
orderflow-tape-scalper Membaca tape secara real-time: deteksi Hajar Kanan (Haka) volume masif, Iceberg Orders, Fake Bid/Ask, dan Broker Summary (Bandarmologi).
timesfm-forecaster Menjalankan model AI Google TimesFM (PyTorch) untuk membuat proyeksi pergerakan harga ke depan dan menghitung quantile probabilitas (P10, P50, P90).
cftc-cot-tracker Mengunduh data resmi CFTC Commitment of Traders tiap akhir pekan, menghitung COT Index (0-100%), dan mengaitkan posisi komoditas global ke saham tambang/energi IDX.
economic-news-trader Memantau rilis berita berdampak tinggi (NFP, Inflasi AS, Suku Bunga The Fed, BI 7-Day RR). Menerapkan Pre-News Shield dan Post-Release Momentum Scalping.
technical-analyst Menghitung Moving Averages, RSI, MACD, Bollinger Bands, Supply & Demand, dan peta likuiditas Smart Money Concepts (SMC).
fundamental-macro-analyst Menghitung rasio keuangan (PER, PBV, ROE, DER, DCF) dan menganalisis korelasi antar pasar (DXY vs Rupiah vs IHSG vs Yield US Treasury).
paper-trading-engine Menjalankan simulasi transaksi lokal (SQLite), kalkulasi fee dan slippage realistis, manajemen trailing stop, dan otomatis membuat berkas Excel (.xlsx).
mobile-chart-server Server dashboard lokal mandiri berbasis FastAPI + TradingView Lightweight Charts dengan tampilan responsif berorientasi Mobile-View First.
agent-update-manager Pengecekan pembaruan otomatis ke GitHub resmi 1x sehari (24h throttle cache) dengan proteksi kredensial .env dan konfirmasi pengguna.

4. Fitur Unggulan Institusional

A. Integrasi Google Research TimesFM

Menggunakan fondasi model time-series decoder-only dari Google untuk memberikan proyeksi pergerakan harga tanpa perlu pelatihan manual (zero-shot). Menghasilkan batas bawah pesimis (Quantile P10) sebagai trailing stop dinamis berbasis statistik, garis median (P50) sebagai target utama, dan batas atas optimis (P90) untuk profit target.

B. CFTC COT Institutional Tracker (Komoditas ke Saham IDX)

Melacak arah uang institusi hedge fund dunia pada kontrak berjangka Emas (COMEX), Minyak Mentah (NYMEX), dan Tembaga. Data ini langsung dihubungkan ke saham emiten komoditas Bursa Efek Indonesia seperti ANTM, MDKA, MEDC, ADRO, dan PTBA.

C. Mode Intersepsi Browser Aman (Zero-Password-Risk)

Agent dapat menempel ke sesi Google Chrome pengguna yang sudah berjalan melalui --remote-debugging-port=9222. Anda melakukan login secara mandiri dengan keamanan biometrik/2FA Anda sendiri. Agent hanya membaca frame WebSocket untuk mengambil data running trade secara pasif tanpa menyimpan sandi Anda.

D. Engine Paper Trading & Ekspor Jurnal Excel Otomatis

Setiap eksekusi simulasi dicatat rapi ke dalam basis data SQLite lokal. Anda cukup memberi perintah: "Buat jurnal trading minggu ini ke Excel", maka sistem akan menghasilkan spreadsheet .xlsx profesional dengan formula return, rasio Sharpe, Maximum Drawdown, dan kurva ekuitas.

5. Penggunaan Lintas Agen AI

A. Claude Code (Anthropic) 

claude
# Di dalam prompt:
> Analisis saham BBCA menggunakan metode teknikal dan cek proyeksi TimesFM untuk 1 minggu ke depan.

B. OpenAI Codex / Custom Agent 

codex --skill-path ./skills

C. Antigravity / Gemini CLI

Tambahkan path repositori ini ke dalam daftar plugin agent Antigravity Anda:

antigravity plugin add https://huggingface.co/spaces/lensetek/trading-ai-agent-skills.git

6. Keamanan Kredensial & Manajemen Risiko

Aturan Keamanan Baku:

  1. Tidak Ada Kredensial di Client-Side: Frontend web chart hanya menerima data harga dan grafik. Kunci API broker tidak pernah dikirim ke browser atau diakses dari client-side.
  2. Persetujuan Manusia Mutlak (Human-in-the-Loop): Tidak ada transaksi beli atau jual yang dieksekusi secara sepihak oleh AI tanpa konfirmasi eksplisit dari pengguna.
  3. Perlindungan Berkas Rahasia: Berkas .env, basis data transaksi *.db, dan berkas laporan *.xlsx secara ketat diabaikan oleh .gitignore.

🇬🇧 English

1. Executive Summary

Trading AI Agent Skills Suite is not a rigid monolithic SaaS application, but a standardized collection of modular Agent Skills. This suite equips your AI coding agent with the full autonomy to autonomously search and download market data, intercept real-time running trade streams directly from your web trading broker session, analyze multi-market instruments using fundamental, technical, macro news sentiment (NFP/CPI/Interest Rates), institutional positioning reports (CFTC COT), and zero-shot probabilistic forecasting via Google Research TimesFM.

All trade execution is strictly governed by a Human-in-the-Loop safety gate (explicit user authorization via CLI Terminal, Web Dashboard, or Telegram Bot), backed by an offline Paper Trading Engine (SQLite) with automated export to Excel (.xlsx) trade journals and a Mobile-View First Web Chart.

2. How to Install Agent Skills

🌟 1. One-Click Standalone Installers (Recommended)

Download the standalone installer directly from the GitHub Releases:

2. Global Installation for AI Agents (One-Prompt Universal Setup)

Copy and paste this instruction directly into your AI Assistant prompt (Google Antigravity, Claude Code, Cursor, Codex, ChatGPT) to install these skills globally:

ext Please install and register the skills globally from https://huggingface.co/spaces/lensetek/trading-ai-agent-skills so they are accessible across all my projects.

3. Via NPX Skills CLI (Local Workspace)

If using an agent skills manager:

npx skills install https://huggingface.co/spaces/lensetek/trading-ai-agent-skills.git

Or install using the bundled CLI script:

# Install to current local workspace:
npx trading-ai-agent-skills

# Or install globally to your system agent directory:
npx trading-ai-agent-skills --global

💬 Method B: Via AI Agent Prompt (One-Click)

Copy and paste this prompt directly into your AI coding assistant (Claude Code, Codex, Antigravity, OpenCode, Cursor, Windsurf):

Agent Prompt:
"Please install and inspect the agent skills suite from https://huggingface.co/spaces/lensetek/trading-ai-agent-skills into this workspace. Clone the repository, review the README.md and SKILL.md files inside ./skills/, and configure .env based on .env.example for paper trading mode. Also generate OpenCode v2 custom agents in .opencode/agents/ by converting each skills/*/SKILL.md to OpenCode agent format (include description in frontmatter), while keeping the skills/ folder intact for cross-agent compatibility."

Your AI Agent will clone the repository, index the skills, and enable autonomous multi-market analysis immediately.

📦 Method C: Manual Git Clone 

git clone https://huggingface.co/spaces/lensetek/trading-ai-agent-skills.git
cd "Trading-AI-Agent-Skills"
pip install -r requirements.txt
cp .env.example .env

3. Suite of 13 Modular Skills

Module Core Purpose & Capabilities
trading-orchestrator Master dispatcher receiving natural instructions, routing tasks to specialist skills, and compiling complete trading plans.
market-data-collector Fetches historical and intraday market data for IDX Indonesian Stocks, Crypto (via CCXT), Forex, and Global Commodities.
backtest-strategy-optimizer High-speed vectorized backtesting, Genetic Algorithm (GA) parameter optimization, Walk-Forward Analysis (WFA), and 2,500+ run Monte Carlo stress testing.
browser-tape-interceptor Attaches to user's existing Chrome browser session via Chrome DevTools Protocol (CDP) to sniff running trade WebSockets safely.
orderflow-tape-scalper Real-time tape reading: detects aggressive volume sweeps, Iceberg orders, Bid/Ask imbalances, and Broker Summary (Bandarmologi).
timesfm-forecaster Runs Google Research TimesFM (PyTorch) for zero-shot time-series forecasting and probabilistic quantile bands (P10, P50, P90).
cftc-cot-tracker Downloads weekly CFTC Commitment of Traders reports, computes COT Index (0-100%), and links global commodities to IDX mining/energy stocks.
economic-news-trader Monitors high-impact events (NFP, US CPI, FOMC, BI 7-Day RR). Enforces Pre-News Shield and Post-Release Momentum Scalping.
technical-analyst Computes MAs, RSI, MACD, Bollinger Bands, Supply & Demand levels, and Smart Money Concepts (SMC) liquidity sweeps.
fundamental-macro-analyst Evaluates financial ratios (P/E, P/B, ROE, D/E, DCF) and analyzes intermarket macro correlations (DXY vs IDR vs Treasury Yields).
paper-trading-engine Local trade simulator (SQLite), realistic fee/slippage modeling, dynamic trailing stop tracker, and automated Excel (.xlsx) journal generation.
mobile-chart-server Standalone local web server powered by FastAPI + TradingView Lightweight Charts optimized for Mobile-View First touch interfaces.
agent-update-manager Automated once-a-day update checker from GitHub (24h throttle cache) with strict credential isolation and human confirmation.

4. Institutional-Grade Features

A. Google Research TimesFM Forecasting

Leverages Google's state-of-the-art decoder-only foundation model trained on billions of time points. Generates probabilistic forecasts with uncertainty intervals without fine-tuning: P10 quantile acts as a dynamic trailing stop, P50 as median expectation, and P90 as high-probability take profit targets.

B. CFTC COT Institutional Tracking

Monitors the real positioning of global hedge funds and commercial hedgers across Gold (COMEX), Crude Oil (NYMEX), and foreign currencies. Direct correlation models bridge global commodity positioning into Indonesian domestic stocks (ANTM, MDKA, MEDC, ADRO, PTBA).

C. Zero-Password-Risk Browser Interception

Through Chrome's --remote-debugging-port=9222, the agent attaches passively to your existing authenticated browser session. You log in manually with your own biometric/2FA security. The agent sniffs WebSocket running trade packets without ever seeing or storing your password.

D. Paper Trading Engine & Automated Excel Journaling

All simulated trades are recorded in an encrypted local SQLite database. Simply ask: "Export this week's trading performance to Excel", and the engine outputs a formatted .xlsx workbook featuring Win Rate metrics, Sharpe Ratio, drawdown curves, and categorized trade logs.

5. Cross-Agent Usage

Each skill contains a standard SKILL.md declaration with semantic prompt parameters, enabling instant integration into:

6. Credential Safety & Risk Management

📄 License

This project is licensed under the MIT License.