A GitHub repository called claude-trading-skills is drawing attention — around 556 stars today — for packaging a full retail investing workflow into Claude "skills", the SKILL.md bundles that extend Claude and Claude Code with reusable prompts, reference material, and helper scripts.

The catalog is broad: sector-rotation and market-breadth analyzers, a technical chart analyst, economic- and earnings-calendar fetchers, screeners for Minervini-style VCP and O'Neil-style CANSLIM setups, an options strategy advisor built on Black-Scholes, a pair-trade screener using cointegration tests, a position sizer, and a portfolio manager that reads live holdings through an Alpaca MCP server.

More unusual is the maintenance machinery. A daily macOS launchd job round-robins through the skills, scores each one from 0 to 100 with a "dual-axis reviewer", and — if the score falls below 90 — invokes Claude via the command line to improve the skill and open a pull request for human review. A weekly pipeline mines the author's own Claude Code session logs for patterns worth turning into new skills, then designs, scores, and proposes them as pull requests. The repository stresses that its material is for "educational and research purposes" and that several skills need paid data APIs or a brokerage account.

It is one of several trading-skill repositories now trending, as the agent-skills format spreads from software engineering into finance.