laiso/ts-bench

Provides dual benchmark datasets—Exercism TypeScript exercises (v1) and Docker-based SWE-Lancer monorepo tasks (v2)—with a CLI interface supporting multiple AI agents and providers. Built as a Bun-based tool with reproducible baselines, GitHub Actions workflows for both datasets, and structured spec documentation for methodology and agent-specific caveats. Includes web UI generation for SWE-Lancer task visualization and frozen release tags to ensure reproducible comparisons across runs.

210 stars.