ix-ml-builder - SKILL.md Agent Skill

name: ix-ml-builder description: Build ephemeral or persistent ML pipelines via ix MCP — auto-detects task type, selects models, handles preprocessing, evaluation, and caching

ML Pipeline Builder (via ix MCP)

Build complete ML pipelines from data to predictions in one tool call using the ix MCP server.

Prerequisites

The ix MCP server must be registered in your .mcp.json:

{ "ix": { "command": "cargo", "args": ["run", "--release", "-p", "ix-agent"], "cwd": "path/to/ix" } }

When to Use

When you have data (CSV, JSON, or inline) and want ML analysis — classification, regression, clustering, or dimensionality reduction.

Quick Start

ix_ml_pipeline({
  "source": { "type": "csv", "path": "data.csv", "target_column": "label" },
  "task": "auto",
  "model": "auto",
  "preprocess": { "normalize": true }
})

Task Auto-Detection

Signal	Task	Default Model
2 unique integer values in target	Binary classification	LogisticRegression
3-20 unique integers, low ratio	Multiclass classification	DecisionTree
Continuous target (>20 unique)	Regression	LinearRegression
No target column	Clustering	KMeans (k=3)

Model Selection

Data Size	Recommended
< 100 rows	KNN (classify) or LinearRegression (regress)
100-10k	DecisionTree (classify) or LinearRegression (regress)
10k+	RandomForest (classify) or LinearRegression (regress)

Persistent Pipelines

Train once, predict later:

ix_ml_pipeline({ ..., "persist": true, "persist_key": "my_model" })
ix_ml_predict({ "persist_key": "my_model", "data": [[1,2,3]] })

MCP Tools

ix_ml_pipeline — Full pipeline: load → preprocess → train → evaluate → persist
ix_ml_predict — Load cached model, predict on new data

Source

Full documentation: ix repo .claude/skills/ix-ml-builder/SKILL.md