{"repo":"reacher-z/HarnessBench","free":true,"listed":false,"github":"https://github.com/reacher-z/HarnessBench","clone":"git clone https://github.com/reacher-z/HarnessBench.git","description":"Benchmark for comparing agent harnesses on everyday online tasks — fixes the base model, varies the harness. Sister project of ClawBench, same scoring pipeline.","language":"Python","stars":40,"topics":["agent-evaluation","agent-harness","agentic-ai","ai-agents","benchmark","browser-agent","browser-automation","computer-use","evaluation","llm"],"license":"Apache-2.0","category":"ai-agents","readme_excerpt":"The Benchmark for Comparing Agent Harnesses on Everyday Online Tasks Read the Docs  ·  Harness Comparison  ·  Cloud Setup If you want to compare base models on a fixed harness, check out our sister project ClawBench  —  same pipeline, orthogonal axis. Install → List → Run.   Reuses ClawBench's pipeline.   Cloud harnesses opt-in via env vars. Which Harness Wins on the Same Task? Given one task (order food, book travel, apply for a job) and one fixed base model -- which agentic harness actually gets it done? Six named harnesses , four runtimes, one pipeline, one leaderboard. --- 6 harnesses  ·  4 runtimes (Python / Node / Rust / Web)  ·  153 shared tasks  ·  15 categories 中文   Plugin Entry-Points          One Container per Harness          Cloud Opt-in          Same Pipeline as ClawBench How It Works LLM Quick Start Point your coding agent (Claude Code, Cursor, Copilot, etc.) at AGENTS.md and prompt away. HarnessBench shares ClawBench's test-cases, container base image, and 5-layer recording stack -- if your agent already knows ClawBench, there is nothing new to learn about the pipeline, only a new harness axis. Human Quick Start Prerequisites: Python 3.10+, uv, and a container engine -- Docker or Podman. Same engine detection as ClawBench; force one with export CO","default_branch":null,"files":null,"tree":[],"storefront":"/r/reacher-z","claimed":false,"request_supported":{"post":"https://gitbuyer.com/r/reacher-z/HarnessBench/request-supported","requests":0},"note":"indexed from public GitHub; nothing is for sale on this page. Clone it from GitHub. Paid listings live at /search."}