One email. No follow-up unless I ship something that fits.
Read firstNot financial advice. This is software that reads free public data and does arithmetic. It cannot see the future. Most people who day trade lose money, and you can too. Everything below assumes a paper account. Talk to a licensed professional if you need advice.
WHAT YOU NEED
Claude Code, or the Claude app with skills. No Claude Code? Paste one SKILL.md into a chat and it runs as a mode, without the Python tools.
Python 3.8 or newer. Nothing to pip install. Every tool is standard library Python.
A paper account if you want tickets to reach a broker. The pack is built around Alpaca's connector, which defaults to paper. You can skip this and never connect a broker at all.
No API key and no signup. Prices come from Yahoo Finance's free chart endpoint, symbols from Nasdaq Trader, filings from SEC EDGAR and the House Clerk.
Prove itThe tests run offline on saved real data and reproduce every number in the project's README example. If they pass, the maths works on your machine.
CONFIGURE
One file, and the lines that protect you
Copy trader-budget/config.example.json to trader/config.json in your project. These are the defaults, and why I keep them.
setting
default
why
paper
true
Tickets go to a paper account. Leave it on until you have weeks of paper results.
max_loss_per_trade_usd
50
Sets the share count, not the other way round. This is the most important line.
max_daily_loss_usd
150
Three bad trades and the risk check blocks the rest of the day.
max_trades_per_day
3
Stops revenge trading.
require_stop
true
No stop, no order.
no_entries_before / after
09:45 / 15:30
New York time. Skips the first wild minutes and the last hour.
allow_options, margin, short
false
Leave all three off.
sec_user_agent
placeholder
Put your name and email here. The SEC asks every tool that reads its filings to identify itself.
In Claude, set your brokerage tool to Needs approval (Customize, Connectors, tool permissions), never Always allow. That gives you a second yes on every order.
ONE TRADE, START TO FINISH
What I got on a real breakout
TSLA, Tuesday 20 October 2026, New York time, replayed after the close. Nothing was placed.
/trader-entrycaptured 21 Oct 2026
Opening range 9:30-9:45: high $370.49, low $363.64TRIGGERED 10:00 AM at $370.58Nothing was placed.
/trader-exitplan before entry
Stop $367.06 opening range midpointTarget $374.10 1R (one times the risk)Risk a share $3.52 · Flat by 15:55 ET
/trader-timemachine10,000 replays, seed 19
target first 51.3 %stop first 44.2 %neither 4.6 %History: fired 34 times in 59 real sessions: 13 target, 15 stop, 6 neither.Close to a coin flip.
/trader-budget then /trader-risk$5,000 paper budget
Size: 10 shares. Max loss at the stop: $35.20.BLOCK #1 buy TSLA 10 shareswindow: the market is closedBlocked orders do not go to the broker.
Read it right51 against 44 is not an edge you can bank. Notice the history: on the real sessions the same rule lost slightly more often than it won. That is the Time Machine doing its job. It talked me out of a trade, on paper, for free.
THE CHECKLIST
My 10 point pre trade checklist
I run this before any ticket, paper or not. If one line is a no, there is no trade.
1accountI am on the paper account. "paper": true in config.
2setupThe scanner or entry timer says triggered, not waiting or expired.
3newsI ran /trader-news. No earnings, deal or halt news I haven't read.
4stopThe stop is set before the entry, and I won't move it further away.
5targetThe target is at least 1R. Reward to risk is 1.0 or better.
6max lossThe loss at the stop is under my per trade limit.
7oddsTime Machine: target first beats stop first by at least 10 points, and the history agrees. A coin flip is a no.
8dayI'm under my trades per day and daily loss limits.
9risk/trader-risk says PASS for this exact order.
10exitAn alert is set for the stop, the target and flat by 15:55.
Why 10 pointsThat threshold is my rule for paper practice, not the project's and not a promise. The run above failed it.
THE TRAPS
What goes wrong, and the fix
"window: the market is closed"
The risk check blocks anything outside 09:30 to 16:00 New York. From Melbourne that is 12:30 AM to 7 AM right now, an hour later once US clocks change on 1 November. Practise on replays in the day, trade paper at night.
Prices look a bit off
Yahoo's chart endpoint is free and unofficial. It can lag, change or rate limit you. Check anything that matters against your broker.
The odds felt like a promise
The Time Machine is built from the last 60 days of 5 minute bars. Earnings, halts and headlines can do things those days never did.
The stop filled worse
A stop is an order, not a guarantee. Across a gap it can fill well past its price. The max loss shown is the loss at the stop price.
News and filings tools refuse
Set sec_user_agent to your name and email, or the SEC_USER_AGENT variable.
Wrong repo
The older trader-skill repo is a different pack (investor copying, screener, tax lots) with no Time Machine. You want day-trader-skill.
US onlyThe scanner covers about 5,900 US listed stocks. US day trading rules apply, and your broker may still enforce the old pattern day trader minimum during its phase in. Set "pdt_guard": true if so. Australian tax treats frequent trading differently again. Nothing here is tax advice.
THE SHAPE OF IT
The eleven
/trader-scannerscans the US market for setups /trader-newsheadlines and filings for your watchlist /trader-entrywatches a setup bar by bar /trader-exitstop and target before you enter /trader-timemachine10,000 replays plus the real history /trader-budgetsizes to your max loss /trader-riskPASS or BLOCK against your limits /trader-brokerthe ticket, paper by default, only on your yes /trader-alertsstop, target and the close /trader-analystreads earnings filings /trader-pelosireads Congress trade disclosures
Each skill writes to a trader/ folder in your project and the next one reads it. You can stop at any step.
That is the whole thing
Install in two minutes, keep it on paper, and let the Time Machine talk you out of the coin flips. Not financial advice.
I'm Raj Singh. I've led ecommerce at two Fortune 100 companies, and at one of them I ran about $8M a year in ads for four years. I've worked with AI since 2019. Now my engineering team and I ship AI products, including debtops.ai. Each guide here is written from a real install, with the output dated.
