A CLI tool to A/B test coding agents on your repo with deterministic checks
Repo‑agent‑bench lets developers compare different coding‑agent configurations on their own codebase by running isolated worktrees and verifying outcomes with custom commands. It records pass rates, token usage, and churn, producing JSON, Markdown, and HTML reports. Ideal for teams evaluating prompts, skills, or infrastructure changes without incurring model costs. Its deterministic demo and zero‑telemetry design make it a practical, privacy‑friendly alternative to cloud‑based leaderboards.
View on GitHub →Uky0Yang/repo-agent-bench