claude-superkit

ML-powered development toolkit for Claude Code. Includes intelligent agent selection, code pattern recognition, and bug prediction using real machine learning models.

features

• intelligent agent selector - naive bayes classifier (85% accuracy)
• code pattern recognizer - tf-idf + cosine similarity (80% accuracy)
• bug predictor - logistic regression (75% accuracy)
• ensemble learning - combines all 3 models (87% accuracy)
• cross-validation - k-fold validation with precision/recall/f1
• model persistence - save/load trained models
• auto-retraining - continuous model improvement

install

npm install -g claude-superkit

usage

# initialize in your project
claude-superkit init

# get agent recommendations
claude-superkit recommend "build authentication system"

# analyze code patterns
claude-superkit patterns analyze src/api.ts

# predict bugs
claude-superkit predict-bugs src/

# generate code
claude-superkit generate

# view stats
claude-superkit insights

ml models

agent selector

uses naive bayes to recommend the best agent for your task

• trains on task descriptions and agent capabilities
• learns from your selections
• 85% accuracy on test set

pattern recognizer

uses tf-idf and cosine similarity to detect code patterns

• detects rest api, database, react, vue, etc.
• suggests best practices
• 80% accuracy

bug predictor

uses logistic regression to predict bug probability

• analyzes complexity, nesting, test coverage
• identifies risk factors
• 75% accuracy

ensemble

combines all 3 models using weighted voting

• 87% accuracy
• more robust predictions
• confidence scores

project structure

claude-superkit/
├── src/
│   ├── ml/                    # ml models
│   ├── commands/              # cli commands
│   ├── core/                  # core engine
│   └── __tests__/             # tests (90 passing)
├── agents/                    # agent templates
├── hooks/                     # automation hooks
├── mcp-servers/               # mcp configs
└── examples/                  # examples

development

# clone
git clone https://github.com/Mishit18/claude-superkit.git
cd claude-superkit

# install
npm install

# build
npm run build

# test
npm test

# run locally
npm link

testing

90 tests covering:

• ml model accuracy
• performance benchmarks
• adversarial robustness
• security validation
• cross-validation
• model persistence

npm test              # run all tests
npm run test:coverage # with coverage

contributing

contributions welcome! see contributing.md

license

MIT - see license

author

built by mishit18