crap4ts

Inspired by Robert C. Martin (Uncle Bob)'s crap4java and crap4clj, adapted for TypeScript and TSX. His later crapper solution calculates CRAP scores for several languages, including TypeScript.

Find complex, poorly tested TypeScript functions and fail CI when their risk exceeds your limit. crap4ts combines cyclomatic complexity and test coverage into a per-function CRAP (Change Risk Anti-Patterns) score.

It analyzes TypeScript/TSX with Istanbul JSON or LCOV coverage and produces text or JSON reports. Supports per-file thresholds, independent monorepo package groups, optional branch scoring, Express route labels, and Git change selection.

Quick start

Requires Node.js >=20.19.0 <25. The package bundles binaries for Linux (glibc 2.35+) and macOS on x64/arm64, plus Windows x64.

npm install --save-dev @crap4ts/crap4ts
# Generate coverage with your test runner first, then:
npx crap4ts --coverage coverage/coverage-final.json src

For LCOV or JSON output:

npx crap4ts --coverage coverage/lcov.info --coverage-format lcov src
npx crap4ts --coverage coverage/coverage-final.json src --format json --threshold 12
npx crap4ts --coverage coverage/coverage-final.json src --coverage-metric branch
npx crap4ts --coverage coverage/coverage-final.json src --changed-since main

Run coverage once with the Vitest/Jest recipes. Generated commands have a configurable five-minute timeout.

Using crap4ts in an agent workflow

Give a coding agent concrete feedback on complex, poorly tested functions: generate coverage, run crap4ts, inspect flagged functions, improve tests or simplify the code, then rerun the gate. JSON reports support automated feedback; the same thresholds can run in CI.

See the agent workflow recipe for configuration and reusable agent instructions.

Scoring and CI

CRAP = complexity² × (1 − coverage)³ + complexity, where coverage is a fraction from 0 to 1. Lower scores are better.

The default threshold is 8. Exit codes: 0 passes, 2 means a score exceeds its threshold, 1 means invalid input or analysis failure. Missing or ambiguous coverage fails by default; --report-only keeps unknown scores visible while still enforcing thresholds on measured scores.

Tests, declarations, and generated output are excluded by default. JavaScript sources and raw V8 coverage are unsupported.

Usage and configuration · How it’s made · MIT license