Included with purchase

CLAUDE.md Skill Files

Seven Markdown files written for Claude Code to consume. Append them to your CLAUDE.md and Claude instantly understands the full methodology — no copy-pasting documentation into every prompt.

install_skills.sh 
# After download, merge skills into your project CLAUDE.md
unzip ndx-pit-dataset.zip -d ~/ndx-pit-data/
cp ~/ndx-pit-data/*.parquet crucible/data/

# Append all skill files to CLAUDE.md
cat ~/ndx-pit-data/skills/SKILL_*.md >> crucible/CLAUDE.md

# Claude Code reads CLAUDE.md on every session start
# Now Claude knows the full methodology — without any prompt engineering
claude "build a momentum strategy using triple barrier labels and CPCV"
How it works

Each file is a methodology specification — concrete code patterns the model matches against, not documentation it ignores. Read how CLAUDE.md changes what Claude generates →

01

SKILL_strategy_research.md

López de Prado research process

The thinking framework behind every strategy. Economic hypothesis first, simple primary signal second, meta-label model third. Covers NDX-specific alpha sources, the iteration loop, and red flags that distinguish real edge from data mining.

Hypothesis-firstPrimary signalMeta-labelingIteration loop
02

SKILL_pit_dataset.md

Point-in-Time filtering

Teaches Claude the exact API for PIT universe construction. Includes correct vs wrong code patterns, the 216 bps bias measurement, and every anti-pattern that introduces survivorship bias.

PIT filteringUniverse constructionSurvivorship biasAnti-patterns
03

SKILL_triple_barrier.md

Labels, meta-labeling & sample weights

Triple barrier labeling from AFML Chapter 3 plus the meta-labeling layer — a second ML model that learns when the primary signal is worth trading. Covers CUSUM events, sample uniqueness weights, and the primary→meta→Kelly pipeline.

Triple barrierMeta-labelingCUSUM eventsSample weights
04

SKILL_cpcv.md

Combinatorial Purged Cross-Validation

The full CPCV methodology. Purging overlapping labels, embargo periods, C(N,K) path generation, and how to interpret a Sharpe distribution rather than a single lucky number.

PurgingEmbargoC(6,2)=15 pathsSharpe distribution
05

SKILL_feature_engineering.md

Stationary features with economic rationale

Production features with stationarity guarantees and the WHY behind each one. Cross-sectional z-scoring, fractional differentiation, momentum/vol/microstructure taxonomy, feature-to-hypothesis mapping, and look-ahead failure modes.

CS z-scoringFeature WHYFrac diffHypothesis map
06

SKILL_position_sizing.md

Kelly, meta-prob, vol targeting

Meta-label probability feeds directly into Kelly as win_prob. Half-Kelly rationale, signal-proportional sizing, inverse-vol weights for equal risk contribution, and drawdown-based position scaling.

Meta-prob → KellyHalf-KellyVol targetingDrawdown scalar
07

SKILL_regime_detection.md

200d MA, vol regime, HMM

Binary 200d MA regime filter, volatility expansion scalar, 2-state HMM for advanced regime detection, and the T+1 lag rule that prevents look-ahead in regime signals.

200d MA filterVol regimeHMM statesT+1 lag

Get all 7 skills with the dataset

The skill files are only available as part of the NDX PIT Dataset purchase.

Not sure yet? Read how CLAUDE.md skill files change what Claude Code generates →