Harness Bench

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

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)
• Per-agent model selectors sourced from the model config
• Dark, monospace-first UI with ghostty-web terminals

Quick Start

bun install
bun run dev

Open http://localhost:3000.

Commands

bun run dev      # UI + PTY server
bun run ui       # UI only (Vite on :3000)
bun run pty      # PTY websocket server on :4000
bun run build    # production build
bun run preview  # preview built app
bun run start    # start output server (.output/server/index.mjs)
bun run test     # vitest
bun run lint     # eslint
bun run format   # prettier
bun run check    # format + lint

Requirements

• Bun
• Agent CLIs installed and available on PATH: amp, droid, pi, codex, claude, opencode
• Git

Architecture

• Frontend: TanStack Start routes and root shell
• WebSocket client connects to ws://localhost:4000/vt
• Dashboard UI renders agent cards and live terminals
• Backend server spawns PTYs, streams base64 output, exposes /diff and /stop
• Theme system controls dark/light styles and tokens

Roadmap

• Wire model selection into backend runs
• Add per-agent stop controls (not just global STOP)
• Persist run metrics/logs for comparison history