中文版

day-trader-skill（GitHub：https://github.com/Jakeschincariol/day-trader-skill ，2026-10-20 创建，今晨抓取时已 5,991★）是 11 个 Claude 日内交易技能：扫描成千上万张图表、跟踪突发新闻、掐准每一笔进出场时机，并且在你花一分钱之前先把结果重播一遍。

README 的说法是：免费、MIT 协议、不用注册、不用 API 密钥，工具就是读免费公开数据的纯标准库 Python（雅虎财经图表、纳斯达克交易员股票清单、SEC EDGAR 与众议院书记官处的公开文件）。11 个技能各司其职：trader-scanner 扫描全市场找开盘区间突破、旗形、超卖反弹等形态；trader-news 跟踪你关注股票的突发新闻；trader-entry 逐根 K 线盯着，告诉你进场的那一刻；trader-exit 在进场前就定好止损与目标；trader-timemachine（时光机）把当天剩余时段按这只股票自己的历史 K 线重播 10,000 次，再跑一遍历史上所有同形态出现的日子，给出诚实的胜率；trader-risk、trader-budget 分别卡风控上限与仓位大小；trader-broker 负责下单；trader-alerts、trader-analyst、trader-pelosi 分别管价格提醒、财报解读和国会议员交易披露跟踪。

工作流是一条链：scanner → news → entry → exit → timemachine → budget → risk → broker，中间每个技能读上一个写下的结果。README 反复强调三条底线：默认模拟盘（paper trading）；没有任何一笔交易能在你亲口说 yes 之前成交，broker 技能还要再过 Claude 自己的“允许一次 / 拒绝”确认卡；以及“这不是投资建议”，时光机是模拟不是预测，大多数日内交易者是亏钱的。

安装方法：把仓库链接粘贴给 Claude 说 install skill，或在 Claude Code 里用 /plugin marketplace add Jakeschincariol/day-trader-skill 安装插件；也可以直接把 trader-* 文件夹拷进 ~/.claude/skills/。

为什么值得了解

• 它把“技能 + 小工具”这套模式推到了很极致的地步：11 个技能像流水线一样串起来，每个只干一件事，数据全部来自免费公开接口，连 pip 都不用装
• “下单前先重播一万遍”的时光机思路值得所有做 agent 的人学：让模型在行动之前先看到诚实的胜率分布，而不是只看到最乐观的那条路
• 三层把关（budget 定仓位、risk 卡上限、broker 要你亲口确认）是 agent 碰钱时应有的样子：再聪明的 agent，碰到真金白银也该默认模拟盘、默认要人点头

English Version

day-trader-skill (GitHub: https://github.com/Jakeschincariol/day-trader-skill , created 2026-10-20, 5,991 stars at this morning’s fetch) is a pack of 11 Claude skills for day trading: scan thousands of charts, track breaking news, time every entry and exit, and replay every outcome before you risk a dollar.

Per the README, it is free, MIT licensed, with no signup and no API key; the tools are plain standard-library Python reading free public data (Yahoo Finance charts, Nasdaq Trader’s symbol lists, SEC EDGAR and the House Clerk’s public filings). The eleven skills each do one job: trader-scanner scans the whole market for opening-range breakouts, flags and oversold bounces; trader-news tracks breaking news on every stock you watch; trader-entry watches a setup bar by bar and tells you the moment to get in; trader-exit sets your stop and target before you enter; trader-timemachine replays the rest of the session 10,000 times from the stock’s own recent bars, plus every past day the same setup fired, and reports the honest odds; trader-risk and trader-budget enforce your limits and position sizing; trader-broker executes orders; trader-alerts, trader-analyst and trader-pelosi cover price alerts, earnings analysis and tracking Congress members’ stock-trade disclosures.

The workflow is a chain: scanner → news → entry → exit → timemachine → budget → risk → broker, each skill reading what the last one wrote. The README stresses three bottom lines: paper trading is the default; no trade goes through until you say yes, with Claude’s own “Allow once / Deny” card as a second check; and “this is not financial advice” — the Time Machine is a simulation, not a prediction, and most people who day trade lose money.

To install, paste the repo link into Claude and say “install skill”, or in Claude Code run /plugin marketplace add Jakeschincariol/day-trader-skill; you can also copy the trader-* folders straight into ~/.claude/skills/.

Why it’s worth knowing

• It pushes the “skills + small tools” pattern about as far as it goes: eleven skills chained like an assembly line, each doing exactly one thing, all data from free public endpoints, nothing to pip install
• The “replay the day 10,000 times before ordering” Time Machine idea is worth learning for anyone building agents: let the model see the honest distribution of outcomes before acting, not just the most optimistic path
• The triple gate (budget sizes the position, risk enforces the limits, broker needs your spoken yes) is how an agent should behave around real money: even the smartest agent should default to paper trading and default to asking a human first