Harness Bench

CLI agent benchmarker dashboard. Run multiple coding agents on the same task, watch their terminals live, and compare each output using reviewer.

harness-bench-demo-github.mp4

Highlights

• Run amp, opencode, claude, codex, pi, droid in parallel
• WebSocket-driven PTY streaming for live terminal output
• Explicit global stop path via POST /stop with shutdown ladder (Ctrl-C, Ctrl-C, SIGTERM, SIGKILL)
• Dark, monospace-first UI with ghostty-web terminals (optionally xtermjs, restty alternatives)
• Per agent Git worktree set up with cleanup control

Quick Start

Requirements:

• Git
• OPENROUTER_API_KEY (add to PATH or provide in UI)
• Bun: curl -fsSL https://bun.sh/install | bash
• Amp: curl -fsSL https://ampcode.com/install.sh | bash
• Droid: curl -fsSL https://app.factory.ai/cli | sh
• OpenCode: curl -fsSL https://opencode.ai/install | bash
• Codex: bun i -g @openai/codex
• Pi: bun i -g @mariozechner/pi-coding-agent
• Claude Code: curl -fsSL https://claude.ai/install.sh | bash

Run locally without installing:

bunx @frixaco/hbench

bunx @frixaco/hbench runs the prebuilt dist server bundle shipped in the package.

Development:

bun install
bun run dev

Commands

bun run dev                     # UI + PTY + REST server
bun run build                   # build fullstack server bundle into dist/
bun run start                   # run dist bundle (production runtime)
bun run lint                    # eslint
bun run format                  # prettier
bun run check                   # format + lint
bun publish --access public     # publish new version