← Back to list
feature-engineering Checked
Feature construction from market data for ML trading models including price, volume, on-chain, an…
9.0Overall 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 "feature-engineering" skill by following the instructions at https://skill123.me/install/feature-engineering.
⌨️ Command line install
Run in your terminal — downloads and installs to ~/.claude/skills/.
curl -fsSL https://skill123.me/install/feature-engineering.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 · 23 KBAbout this skill
Overview
Feature construction from market data for ML trading models including price, volume, on-chain, and microstructure features
Source
- Repo: https://github.com/agiprolabs/claude-trading-skills
- Path: skills/feature-engineering
Score breakdown
Trigger
7.0
Description states What (feature construction for ML trading models, four data families) with good domain keywords; no when/when-not in description; body carries a useful integration map to complementary skills.
Structure
9.3
299-line SKILL.md, two references (feature catalog, pitfalls), two scripts; Files section maps roles including demo vs live modes; no redundant files.
Workflow
9.2
Excellent workflow: no-lookahead guarantee as hard rules, stationarity transform table, rolling-normalization with WRONG/CORRECT code contrast, ordered 4-step selection pipeline, label creation with thresholds; demo mode default and ADF quality report provide verification; scripts enforce the guards.
Content
9.0
Dense and imperative with concrete formulas and runnable snippets; short 'Why Features Beat Models' preamble is mild background but earns its place.
Engineering
10.0
Valid frontmatter; scripts have shebangs, dependency and env-var docs, graceful optional-deps handling (statsmodels); all referenced paths exist.
Security
10.0
6.0 manifest: data reads BIRDEYE_API_KEY/TOKEN_MINT env (declared); side effects none; network only to official Birdeye API in opt-in --live mode; credentials not sent anywhere except Birdeye auth; persistence none. Capabilities match declared scope. No injection, no destructive commands.
High-quality ML feature-engineering skill for trading models: 7 feature categories with formulas and lookbacks, stationarity (ADF) testing, no-lookahead rules stated as absolute constraints with correct/incorrect code contrast, 4-step feature-selection pipeline, label creation guidance, and two scripts (demo default with synthetic data; --live via Birdeye). Cross-skill integration map is well done. 6.0 manifest: network destination public-api.birdeye.so only, in explicitly opt-in --live mode, with BIRDEYE_API_KEY read from env and sent only to Birdeye itself as auth header - declared in both SKILL.md and script docs; no local file writes, no persistence. Capabilities match declared scope. No injection, no secrets hardcoded.