# Ghost Auto Trader

> Architect and deploy the Ghost Auto-Trader framework: a zero-DTE options trading pipeline using TradingView webhooks, AI-gate validation, and broker execution.

- Skill: `mphinance/ghost-auto-trader` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add mphinance/ghost-auto-trader`
- Raw SKILL.md: https://api.skillmd.com/api/skills/mphinance/ghost-auto-trader/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: DevOps & Infra
- Author: mphinance (https://skillmd.com/u/mphinance)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/mphinance/ghost-auto-trader

---


# Ghost Auto-Trader Architecture

This skill helps you deploy the Ghost Alpha Trading System for 0DTE options.

## Architecture Pipeline
1. **Signal Generation:** TradingView fires a "Ghost Alpha Grade A" alert based on momentum squeeze metrics.
2. **Webhook Receiver:** A Python FastAPI/Flask backend receives the signal payload containing ticker, timeframe, grade, and relative volume.
3. **The AI Gate:** The payload is sent to an LLM (e.g., Gemini Flash) for immediate contextual validation (checking macro alignment and news sentiment).
4. **Execution:** If the AI gate approves, the system automatically buys the ATM/OTM 0DTE option via the broker API (e.g., Tradier).
5. **Position Management:** A rigid 30-second monitor loop enforces +50% Take Profit, -40% Stop Loss, and a hard 3:00 PM ET time-based exit.

## Choosing the execution broker

This pipeline places **live orders on a tool call** — the AI gate is the only thing
between a signal and a filled position, so the broker you wire in matters as much as the
strategy. Before hardcoding a broker API:

- Use the [`broker-mcp-selector`](../broker-mcp-selector/) skill to choose one, or check
  the broker directly in
  **[awesome-broker-mcp](https://github.com/mphinance/awesome-broker-mcp)**.
- The default here, [Tradier](https://github.com/mphinance/awesome-broker-mcp/blob/main/brokers/tradier.md),
  is *live on a tool call* — there is no broker-enforced approval step, so your risk
  parameters and the AI gate are the entire guardrail. Size and test accordingly.
- **Paper/sandbox first.** Run the full pipeline against a paper account for far longer
  than feels necessary before pointing it at real money. Prompt injection into the
  signal or news feed reaches the order in this design.

## Usage
When the user asks to "set up a trading bot" or "build an auto-trader":
1. Scaffold the `main.py` webhook listener.
2. Scaffold the `auto_trader.py` execution engine with the AI gate.
3. Ensure strict risk management parameters are hardcoded.