Jakeschincariol/day-trader-skill

The problem

Day trading is mostly a chain of small computations: scanning charts for setups, checking headlines, timing entries, sizing positions against a max loss, and keeping a stop honest. Retail traders usually glue this together from broker screeners, spreadsheets and chat tabs, or pay for a signal service. This repo packages eleven Claude skills that read free public data (Yahoo Finance charts, Nasdaq Trader symbol lists, SEC EDGAR, House Clerk disclosures) and hand the output from one step to the next inside a chat session.

What's different

Most retail day-trading tooling is either a broker-integrated scanner with its own UI, or an indicator suite inside a charting platform; both assume you live in that product. Here the workflow is a sequence of commands an AI assistant can call, and the artifacts are plain JSON files in a trader/ folder that the next skill reads. Two pieces stand out: the Time Machine, which replays the rest of the session thousands of times from the stock's own recent bars and also re-runs the setup rule over past sessions, and a hard gate where a risk check must pass and you must type yes before anything reaches a broker. The trade-off is honest: everything runs on unofficial, delayed-ish public endpoints rather than a paid feed, and the AI reads and reasons but the arithmetic lives in standard-library Python.

Who it's for

Discretionary day traders who already have a method and want a repeatable, auditable checklist without paying for another platform, and developers who want to read or modify the strategy logic. It also suits people learning how a trade plan is constructed, since the paper-trading default and the simulation make the reasoning visible. Look elsewhere if you need real-time tick data, options, automated strategy execution at scale, or a compliance-grade audit trail; and if you want a hands-off signal feed, this is the opposite of that.

Cost of adoption

Python 3.8+, no third-party packages, no API keys. The SEC tools require you to set a name and email user-agent. Broker order execution depends on a separate Alpaca MCP connector, with paper accounts as the default. The main risks are data quality and reliability: the README itself warns that the Yahoo chart endpoint can lag, change or rate-limit, and that bars are 1- or 5-minute, not ticks. The README is unusually explicit about limits (simulation is not prediction, stops can gap), which is a good sign, but this is a four-day-old repo with no documented track history, and it is not financial advice.

Verdict

This is a brand-new project riding a fast spike: 4 days old, about 1,115 stars and roughly 251 per day at its board peak, never on the daily trending board and with no Hacker News discussion. The design is more thoughtful than the average trading-script dump, particularly the explicit approval gate and the willingness to tell you a setup is close to a coin flip. But 11 skills and an offline test suite on fixtures prove the code runs, not that the edge exists. Read it as a well-structured automation and education scaffold, paper-trade it for a while, and do not mistake the polish for a proven system.