Skip to content

About

No description, website, or topics provided.

Resources

Readme
MIT license

Code of conduct

Code of conduct

Contributing

Contributing

Security policy

Security policy
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository
BranchesTags

Repository files navigation

⚡ GLD LEAP Option Agent

GLD Agent License: MIT

An agentic trading system that runs a 7-prompt analysis framework using Claude AI with live web search, real-time option chain data (yfinance), and TradingView technical analysis to evaluate GLD (Gold ETF) LEAP options and deliver concise recommendations via Telegram — scheduled automatically through cron-job.org → GitHub Actions.

System Architecture

cron-job.org (external scheduler)
  ↓ POST to GitHub API (workflow_dispatch)
  ↓ run_type=full (8:30 AM CST) or run_type=intraday (hourly)

GitHub Actions
  ↓ python agent.py full | intraday

Data Sources (run in parallel before AI analysis)
  ├── yfinance        → Live GLD option chain (real bid/ask prices, 15-min delayed)
  └── TradingView     → RSI, MACD, SMA50/200, technical bias (authenticated)

7-Prompt Analysis Framework (Claude AI)
  Prompt 1 (Haiku)  → IV & Options Chain Analysis    [full only]
  Prompt 2 (Haiku)  → IV Entry Timing Gate            [full only]
  Prompt 3 (Sonnet) → News Sentiment                  [always]
  Prompt 4 (Haiku)  → Event Risk Calendar             [full only]
  Prompt 5 (Sonnet) → ETF Fundamental Scorecard       [full only]
  Prompt 6 (Haiku)  → Strike & Expiry Optimizer       [full only]
  Prompt 7 (Sonnet) → Master Synthesis                [always]
  ↓
Telegram Bot → concise recommendation to your group/channel
Detail Value
📅 Schedule Mon–Fri, full at 8:30 AM CST + hourly intraday 9:30 AM–2:30 PM CST
💬 Delivery Telegram (plain text, group or channel)
☁️ Host GitHub Actions (triggered via cron-job.org)
💰 API Cost ~$3–5/month

How the Analysis Works — Step by Step

Each full run executes 7 sequential AI prompts, each building on the previous. Here's exactly what happens:

Step 1 — Fetch Live Market Data (no AI)

Before any AI call, the agent collects real data:

  • yfinance: Fetches the GLD option chain for all expirations 12+ months out. For each expiry, it pulls strikes in the $380–$470 range with bid, ask, IV, and volume. Marks which strikes are within the $6,500/contract budget.
  • TradingView (authenticated): Fetches 100 daily candles for GLD. Computes RSI(14), MACD, SMA50, SMA200, and a technical bias (BULLISH/NEUTRAL/BEARISH) locally in Python — no AI needed.

Prompt 1 — IV & Options Chain Analysis (Haiku, no web search)

Input: The raw yfinance option chain data
Task: Analyze the chain without any web search since real data is already provided.

  • Calculates IV Rank (IVR) and IV Percentile — cheap (<30) vs expensive (>70)
  • Compares IV vs 30-day Historical Volatility to assess if options are over/underpriced
  • Analyzes IV skew across strikes (ATM vs OTM)
  • Estimates Delta, Gamma, Theta, Vega for 3 candidate strikes
  • Identifies institutional positioning from Open Interest and Volume
  • Outputs top 3 strike/expiry combos ranked by cost-efficiency (ITM delta ~0.70–0.80, ATM ~0.50)

Prompt 2 — IV Entry Timing Gate (Haiku, web search)

Input: Web search for current GLD IV metrics
Task: Determine if now is a good time to enter a LEAP position.

  • Retrieves current IV, 52-week high/low, IV Rank, IV Percentile, 30-day HV
  • Applies a simple gate: IVR < 30 = BUY NOW, 30–50 = CAUTION, >50 = WAIT
  • Calculates IV/HV ratio (>1.2 means options are expensive)
  • Estimates the implied ±1 standard deviation move by LEAP expiration
  • Outputs a single verdict: ✅ BUY NOW / ⚠️ WAIT / ❌ AVOID

Prompt 3 — News Sentiment (Sonnet, web search) — runs every run

Input: Web search for GLD/gold news from the past 5 trading days
Task: Score the macro news environment for a 12–24 month LEAP thesis.

  • Searches for: Fed policy signals, DXY moves, real yields (TIPS), central bank gold demand, geopolitical events, ETF flow data
  • For each story: sentiment (Bullish/Bearish/Neutral), impact window, and LEAP relevance
  • Outputs a net sentiment score from −5 (very bearish) to +5 (very bullish)
  • Identifies the #1 macro risk that could invalidate the thesis and the #1 upside catalyst

Prompt 4 — Event Risk Calendar (Haiku, web search)

Input: Web search for upcoming macro events
Task: Map out the event landscape over the next 3–18 months.

  • Searches for: FOMC meeting dates, CPI/PCE releases, DXY key levels, central bank reports, geopolitical risk events
  • For each event: estimated IV impact (High/Medium/Low), directional bias for GLD, and LEAP implication (enter before / after / avoid)

Prompt 5 — ETF Fundamental Scorecard (Sonnet, no web search)

Input: TradingView technical data (RSI, MACD, SMAs, price) fetched in Step 1
Task: Score GLD as a LEAP candidate across 4 dimensions.

  • Structure: AUM, daily options volume, bid/ask spread quality, expense ratio
  • Macro: Assesses current macro regime (risk-off/on), DXY trend and correlation, real yield direction, Fed policy stance
  • Technical: Uses live TradingView data — price vs SMA50/200, RSI(14), MACD signal, trend, key support/resistance
  • LEAP Suitability Score: Rates GLD 1–10 weighted equally across IV environment (25%), trend (25%), macro (25%), and liquidity (25%)

Prompt 6 — Strike & Expiry Optimizer (Haiku, no web search)

Input: The raw yfinance option chain data
Task: Find the best specific trade within the $6,500 budget using real ask prices.

  • Uses only real ask prices from the chain (never estimates premiums)
  • Models 3 scenarios: ITM (delta ~0.80), ATM (delta ~0.50), OTM (delta ~0.30)
  • For each: cost per contract (ask × 100), breakeven at expiry (strike + ask), probability of profit, max contracts within $6,500
  • Recommends the best strike for risk-adjusted return
  • Provides a roll strategy: when/how to roll if the position moves 20% against

Prompt 7 — Master Synthesis (Sonnet, no web search) — runs every run

Input: All outputs from Prompts 1–6
Task: Synthesize everything into a single Telegram-ready recommendation.

  • Weighs all 6 analysis components against the constraints: $6,500 budget, 2–3x target return, 12–18 month horizon, 50% max loss tolerance
  • Enforces pricing rules: only recommends strikes where ask × 100 ≤ $6,500
  • If no strike fits the budget, outputs a WAIT message with the cheapest available option and the price level to watch
  • Delivers a fixed-format output directly usable as a Telegram message

Intraday Runs (news + synthesis only)

For hourly intraday runs, only Prompt 3 (news) and Prompt 7 (synthesis) execute. The synthesis prompt is aware it's an intraday run and bases the recommendation primarily on news and macro context, clearly noting it's an intraday update rather than a full analysis.

Cost Optimization

Technique Saving
Model routing: Haiku for P1,P2,P4,P6 ~70% cheaper than Sonnet for those calls
No web search on P1, P5, P6 Eliminates ~3 unnecessary search calls
max_uses=2 per search (was 5) Halves search token consumption
Per-prompt max_tokens (350–900) Cuts wasted output budget vs 2048–3000 default
Intraday = 2 prompts only ~$0.01/run vs ~$0.16/full run

Estimated monthly cost: ~$3–5 for 22 trading days (1 full + 5 intraday runs/day)

To verify caching is working, check logs for cache_read_input_tokens > 0.

Project Structure

option_analyzer/
├── agent.py               # 7-prompt analysis agent with model routing & prompt caching
├── tv_analysis.py         # TradingView data fetcher (RSI, MACD, SMA — no AI)
├── telegram_sender.py     # Telegram bot sender (plain text)
├── requirements.txt       # Dependencies
├── CLAUDE.md              # Implementation spec for Claude Code
├── .env.example           # Template for environment variables
├── .github/
│   └── workflows/
│       └── gld_agent.yml  # workflow_dispatch only (triggered by cron-job.org)
└── README.md

Quick Start

Prerequisites

Installation

git clone https://github.com/sunilravuri/option_analyzer.git
cd option_analyzer

python -m venv .venv
source .venv/bin/activate      # macOS/Linux
# .venv\Scripts\activate       # Windows

pip install -r requirements.txt

Configure Environment

cp .env.example .env
# Edit .env:
#   ANTHROPIC_API_KEY=sk-ant-...
#   TELEGRAM_BOT_TOKEN=...
#   TELEGRAM_CHAT_ID=...         # negative number for group/channel
#   TRADINGVIEW_USERNAME=...
#   TRADINGVIEW_PASSWORD=...

Run Locally

# Full run (all 7 prompts — use at market open)
python agent.py full

# Intraday run (news + synthesis only)
python agent.py intraday

Deploy to GitHub Actions

  1. Push this repo to GitHub.

  2. Go to Settings → Secrets → Actions and add:

    Secret Where to Get It
    ANTHROPIC_API_KEY console.anthropic.com → API Keys
    TELEGRAM_BOT_TOKEN Telegram → @BotFather → /newbot
    TELEGRAM_CHAT_ID Use negative ID for groups/channels (e.g. -1003358222047)
    TRADINGVIEW_USERNAME Your TradingView login email
    TRADINGVIEW_PASSWORD Your TradingView password

  3. To test: Actions → GLD LEAP Agent → Run workflow → set run_type=full.

cron-job.org Setup

Create two jobs at console.cron-job.org:

Job 1 — Full run (market open)

Field Value
Schedule 30 14 * * 1-5 (8:30 AM CST = 14:30 UTC, Mon–Fri)
URL https://api.github.com/repos/sunilravuri/option_analyzer/actions/workflows/gld_agent.yml/dispatches
Method POST
Header Authorization: Bearer YOUR_GITHUB_PAT
Header Accept: application/vnd.github+json
Body {"ref":"main","inputs":{"run_type":"full"}}

Job 2 — Intraday runs (hourly)

Field Value
Schedule 30 15-20 * * 1-5 (9:30 AM–2:30 PM CST, Mon–Fri)
URL same as above
Method POST
Headers same as above
Body {"ref":"main","inputs":{"run_type":"intraday"}}

The GitHub PAT needs repo or actions:write scope.

Telegram Bot Setup

  1. Open Telegram → search @BotFather → tap "Start"
  2. Send /newbot, give it a name and username
  3. Copy the BOT_TOKEN BotFather gives you
  4. For a group: Add the bot to your group, then get the group's negative chat ID
  5. For a channel: Add the bot as admin, use the channel's -100XXXXXXXXXX ID
  6. To find the chat ID: send a message in the group/channel, then visit https://api.telegram.org/bot<TOKEN>/getUpdates and look for "chat":{"id": -100XXXXXXXXXX}

Sample Output

📊 GLD LEAP — 2026-03-16 11:50 CST
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
💰 GLD: $458.46 | Tech: NEUTRAL | Macro: FAVORABLE | IV: FAIR

✅ GO RECOMMENDATION
Strike: $460 | Expiry: Mar 2027 | Ask: $57.80/share → $5,780/contract
Contracts: 1 | Break-even: $517.80 | Delta: ~0.52

📌 Why: Highest volume, ATM delta, 13-month runway, dollar weakness tailwind
⚡ Conviction: MEDIUM (6/10) — FOMC overhang resolved post March 18-19
🎯 Target: $520 | 🛑 Stop: GLD < $435
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
⚠️ Educational only. Not financial advice.

License

MIT — see LICENSE.

⚠️ Disclaimer: This agent is for educational and automation purposes only. The GLD LEAP option analysis generated does not constitute financial advice. Always consult a licensed financial advisor before making investment decisions. Options trading involves significant risk of loss.

About

No description, website, or topics provided.

Resources

Readme
MIT license

Code of conduct

Code of conduct

Contributing

Contributing

Security policy

Security policy
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages