Free pack3 Claude skills for traders

Stop guessing. Start testing.

I passed two prop firm evaluations with rule-based systems, and my AI-built bot now trades two funded accounts. This is the process I use to research and test every system, packed into three Claude skills: a literature search, a feature table, and a six-step backtest with validation. Free, and no coding needed.

You get a confirmation email, then the download link, a short series of emails on how I build and test systems, and news about my courses. Unsubscribe with one click at any time. Details in the privacy policy.

Validation · step 4 of 6In the backtest

NQ Asia Reversion

MNQ · 1-min bars · Nov 2021–Nov 2026 · costs included

	Whole	In-sample	Out-of-sample
Trades	834	498	336
Sharpe	1.15	1.27	1.06
Profit factor	1.22	1.24	1.19
Max drawdown	−5.6%	−5.6%	−5.1%

8 of 8 checks · hypothetical Passed

Where it ledMy prop accounts, as of November 2026

2evaluations passedSmall caps, five rule-based systems, traded by hand

$1,351first payoutOctober 2026, after the second passed evaluation

2funded bot accounts150K futures accounts, FUNDED, run by a bot I built with Claude

16systems in publicThe 2027 Experiment: every payout, every failed account, every cost

What it cost to get there: 17 bot evaluations for $4,848.50, 13 of them failed. I show those numbers too. Results not typical. Past results do not predict future results.

The process

From an idea to tested rules

Most trading ideas fail a fair test. The skills make that test the default: every step is fixed, Claude asks before each decision, and you see why an idea passes or fails.

01ResearchWhat do papers say about the market and the setup?

02FeaturesWhat was known each day before the trade?

03GridA few values per parameter, in-sample only.

04ValidationOut-of-sample, cost stress, Monte Carlo.

05FilterOne filter at a time, or none.

06ReportOne standard PDF of the final system.

What's in the pack

Three skills, one ZIP each

You add each ZIP to Claude once. Then you describe what you want in plain words and Claude leads you through the steps with one question at a time.

Idea search

tm-idea-search

Searches eight free research sources, from Google Scholar and SSRN to NBER and arXiv
Reads the papers and labels each source peer-reviewed or not
Turns every usable finding into an idea you can test

You getOne report: search log, evidence table with links, idea cards, a suggested test order

Feature calc

tm-feature-calc

An Excel table for your asset, one row per trading day
Your features plus the ones Claude proposes from the idea and the research, each with the time it is known
Optional: which features go with your good and bad trades, in five quintiles

You getThe feature table, and a short PDF with proposed filter levels

Backtest

tm-backtest

A standard Python file from your idea, with fixed costs, slippage and sizing
A small grid, then validation: in-sample vs out-of-sample, +1 tick stress, 1,000-run Monte Carlo, eight checks
Filters tested one at a time, at most one in the final system

You getExcel workbooks, charts and a standard PDF report

IncludedWorked example

One real system, all six steps

An NQ mean-reversion idea for the Asian session, taken through the whole process. Every file is in the pack: the strategy, the grids, the validation, the filter test and the final PDF report.

Backtest file from the idea and a literature search, standard costs and sizing
Grid of two free parameters, 3 values each, 9 sets, in-sample only
Plateau chosen, not the single best set; the edge flag says where to widen
Validation: all eight checks pass, out-of-sample still positive with +1 tick on every fill
Filter grid: two filters, five levels each, one recommended level
Report: verdict, rules, figures by year, Monte Carlo, limits

Validation	Whole	In-sample	Out-of-sample
Trades	834	498	336
Sharpe (daily)	1.15	1.27	1.06
Profit factor	1.22	1.24	1.19
Win rate	39%	40%	38%
Max drawdown	−5.6%	−5.6%	−5.1%
Costs, share of gross	33%	36%	29%
t-stat, average trade	2.49	2.11	1.46

MNQ micro futures, 1-minute bars from Massive, 15 Nov 2021 to 10 Nov 2026. 0.50 USD commission per side, 1 tick slippage on market orders, 2 on stops. Monte Carlo P90 drawdown −11.9%. The years are uneven: 2023 and 2025 are close to zero. Hypothetical results from a backtest.

Who it's for

Built for systematic traders

For you if

You already trade futures, prop accounts or small caps
You want rules instead of gut feeling
You have ideas but no clean way to test them
You want Claude to do the coding, and you want to see every number

Not for you if

You are looking for signals or trades to copy
You want a strategy that is promised to work
You have never placed a trade
You want a coding course

What you need

Three things, set up once

01 · Claude

A Claude account with skills

Skills work on every Claude plan, the free one included. Code execution (“Create and edit files”) must be switched on in the settings; the idea search also uses web search.

02 · Python

Python 3.10 or newer

For the feature table and the backtest. Claude writes the files; you click Run in PyCharm or paste one command into a terminal.

03 · Data

Price data

From Massive (formerly Polygon), Databento, or your own CSV or Parquet file. Your API key stays in your own files; the pack contains none.

PHOTO
440 × 440

Who builds this

Hi, I'm Florian

I'm an engineer and a maths guy, and I'm not a programmer. For more than 12 years I optimised industrial processes, data first. A trading system is the same job on a different machine.

So I build my trading the same way: an idea, a fair test, a validated system, and only then a prop account. That path took me to two passed evaluations by hand, a payout, and a bot that trades two funded accounts. Claude closed the coding gap: it does the research, the backtests and the bots with me.

In 2027 I run 16 systems on two prop firms in public, with every result on the channel. Later I'll teach the whole method in a course, and the people on this list hear about it first.

Watch on YouTube

Questions

Before you sign up

Is the pack really free?

Yes. You give your email address, you get the three skills and a short series of emails on how I use them. You can unsubscribe with one click and keep the skills.

Do I need to know how to code?

No. Claude writes the Python files and tells you how to run them. You need Python installed and you need to read the results; the skills explain every figure they show.

Which markets does it work for?

Any market with bar data: futures such as NQ, ES, YM and RTY, stocks, forex and crypto. The worked example uses MNQ micro futures.

Does it work with the free Claude plan?

Skills are available on the free plan. Long research runs and large backtests use a lot of messages, so a paid plan is more comfortable for regular use. Plans and limits change; Anthropic's help pages have the current state.

Where does my data go?

The backtests run on your own computer. Your price data and API keys stay in your files. Claude sees what you share in the chat.

Is there a paid course?

Yes, later. The course will teach the whole method: building systems with Claude, testing them, and running them on prop accounts. The free pack is the first part of it. People on the list hear about the course first.

Can you promise me results?

No one honestly can. I show you what happened on my own accounts and in my tests, with every cost. What you get out of it depends on your systems, your risk and your discipline.

Free pack3 Claude skills

Build and test your systems the way I do

Three Claude skills and one worked example, free for your email address.

Get the free pack ↑