← Back to list
walk-forward-validation Checked
Walk-forward validation framework for trading strategies and ML models with time-series-aware spl…
8.2Overall rating
Installation
🤖 Install via AI
Copy the prompt below and send it to your AI assistant (e.g. Claude Code) — it will follow the instructions and install automatically.
Install the "walk-forward-validation" skill by following the instructions at https://skill123.me/install/walk-forward-validation.
⌨️ Command line install
Run in your terminal — downloads and installs to ~/.claude/skills/.
curl -fsSL https://skill123.me/install/walk-forward-validation.sh | bash
📦 Download ZIP
Download the zip and extract it into your skills directory (e.g. ~/.claude/skills/), then restart your session.
⬇ Download v1.0.0 · 24 KBAbout this skill
Overview
Walk-forward validation framework for trading strategies and ML models with time-series-aware splits, overfit detection, and regime-aware validation
Source
- Repo: https://github.com/agiprolabs/claude-trading-skills
- Path: skills/walk-forward-validation
Score breakdown
Trigger
4.0
Description covers What with strong domain terms (time-series-aware splits, overfit detection, regime-aware validation) but no When/When-Not or trigger variants: 3+2+3+0=8/20
Structure
9.3
212-line SKILL.md, three well-scoped references with inline load cues, two scripts with demo modes: 4+5+3+2=14/15
Workflow
8.0
Theory-plus-engine design; purging/embargo act as explicit leakage guards; overfit detection (DSR/PBO) forms a genuine verification loop; crypto-specific window tables aid application: 4+5+5+3+3=20/25
Content
10.0
Concise, mathematically precise, runnable quick-start and CLI examples: 10/10
Engineering
10.0
Valid frontmatter; scripts have shebangs and declared deps (numpy/pandas); all referenced paths exist: 10/10
Security
10.0
Fully offline; no eval/base64/destructive ops, no secrets, no injection, no external URLs of concern; capabilities match declared scope: 20/20
Rigorous time-series validation skill: walk-forward engine (rolling/expanding, purging, embargo) plus overfit detector (Deflated Sharpe, PBO), all offline with synthetic --demo modes. Capability manifest: no file/env/credential reads, no writes, no network egress at all, no persistence — capabilities match declared scope and are strictly narrower than it.