oft3r/agentic-trading-desk

agentic-trading-desk is a personal trading analysis tool that pairs an AI agent with local Python scripts to evaluate stocks and ETFs for short-term buy and sell decisions. The core principle is a strict division of labor: the AI fetches market data through the Robinhood API and interacts with the user, Python scripts perform the actual calculations, and the user reviews and approves every order before it executes. The analysis is built around a three-pillar scoring framework where each asset receives a score from -6 to +6. Trend checks price position relative to three exponential moving averages and whether they are stacked in an upward order. Momentum combines the RSI, MACD histogram, and a triple-smoothed rate-of-change indicator called TRIX to measure the strength of recent price action. The third pillar, Macro-Sentiment, evaluates the broader market environment by comparing ETF ratios (small-caps versus large-caps, high-yield versus investment-grade bonds, equities versus treasuries) and the yield curve spread fetched from the web. The Python scripts run from the command line and accept JSON as input. They use only the Python standard library, so no extra packages are needed. The scoring script produces a full scorecard and recommends one of several predefined actions, such as "HOLD (ride the cycle)", "EXIT / TRIM", or "WAIT (do not chase)". The recommendation logic checks for specific indicator patterns, such as the RSI turning down from overbought territory or the MACD histogram shrinking, before settling on a verdict. The project integrates with Claude Code as a skill: when you ask the AI to analyze a ticker, it fetches data from Robinhood, runs the Python scripts, and presents the scorecard along with suggested next steps. Orders are never placed without the user's explicit confirmation. A SKILL.md file in the repository defines the behavioral guardrails for the AI agent.