AI Skill Hub 强烈推荐:PawBench 是一款优质的Agent工作流。AI 综合评分 8.0 分,在同类工具中表现稳健。如果你正在寻找可靠的Agent工作流解决方案,这是一个值得深入了解的选择。
PawBench 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
PawBench 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
# 方式一:pip 安装(推荐)
pip install pawbench
# 方式二:虚拟环境安装(推荐生产环境)
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install pawbench
# 方式三:从源码安装(获取最新功能)
git clone https://github.com/agentscope-ai/PawBench
cd PawBench
pip install -e .
# 验证安装
python -c "import pawbench; print('安装成功')"
# 命令行使用
pawbench --help
# 基本用法
pawbench input_file -o output_file
# Python 代码中调用
import pawbench
# 示例
result = pawbench.process("input")
print(result)
# pawbench 配置文件示例(config.yml) app: name: "pawbench" debug: false log_level: "INFO" # 运行时指定配置文件 pawbench --config config.yml # 或通过环境变量配置 export PAWBENCH_API_KEY="your-key" export PAWBENCH_OUTPUT_DIR="./output"
<p align="center"> <a href="README.md"><strong>English</strong></a> · <a href="README.zh-CN.md">简体中文</a> </p>
<p align="center"> <a href="#tasks"> <img alt="tasks" src="https://img.shields.io/badge/tasks-150-2ea44f"> </a> <a href="https://agentscope-ai.github.io/PawBench/"> <img alt="models" src="https://img.shields.io/badge/models-9-0969da"> </a> <a href="#harnesses"> <img alt="harnesses" src="https://img.shields.io/badge/harnesses-3-8250df"> </a> <a href="https://agentscope-ai.github.io/PawBench/"> <img alt="leaderboard" src="https://img.shields.io/badge/leaderboard-live-cf222e"> </a> <a href="https://github.com/agentscope-ai/OpenJudge"> <img alt="OpenJudge Ecosystem" src="https://img.shields.io/badge/ecosystem-OpenJudge-blue?logo=github&color=0969da"> </a> <a href="LICENSE"> <img alt="license" src="https://img.shields.io/badge/license-Apache%202.0-blue"> </a> </p>
<p align="center"> <strong>A Model × Harness co-evaluation benchmark for agentic AI.</strong><br> 150 agent tasks · 9 models · 3 harnesses · task slices · diagnostic traces </p>
---
The same model can behave very differently once it is placed inside a real agent runtime. A failure may come from model reasoning, missing tools, weak skill discovery, poor workspace awareness, brittle web access, or a completion check that is too loose. A single final pass rate cannot separate these causes.
PawBench is built around one claim:
$$\text{Agent Performance} = f(\text{Model}, \text{Harness})$$
[!NOTE] PawBench is part of the OpenJudge ecosystem. It shares OpenJudge's philosophy of evaluation-driven optimization, but focuses specifically on the interaction between LLMs and agent harnesses.
It evaluates the model and the harness together while keeping enough metadata to read both dimensions independently. v1.0 covers 9 models × 3 harnesses × 150 tasks, with public prompts, graders, task labels, submissions, and leaderboard slices.

With PawBench, you can:
Python 3.11+ and Docker are required. Node.js 20+ is only needed for the leaderboard site.
Install dependencies and add credentials. DashScope is the recommended provider for the default setup:
pip install -r requirements.txt
cat > .env <<'EOF'
DASHSCOPE_API_KEY=...
JUDGE_API_KEY=...
JUDGE_BASE_URL=...
EOF
For OpenAI-compatible or custom providers, set OPENAI_API_KEY / OPENAI_BASE_URL or CUSTOM_API_KEY / CUSTOM_BASE_URL as needed.
PawBench is intended to be used as a diagnostic benchmark, not just a ranking table.
| Goal | Recommended setup | What to inspect |
|---|---|---|
| Choose a model | Fix one harness, run multiple models | Overall score, text/multimodal split, cost and trace quality |
| Choose a harness | Fix one model, run multiple harnesses | Harness gap, task errors, tool-use traces, workspace artifacts |
| Debug a harness | Rerun targeted slices after a change | Capability/source/scenario deltas, failed graders, transcripts |
| Add a dataset | Add tasks with the five-label taxonomy | Coverage balance, grader reliability, task detail page |
| Submit results | Aggregate run logs into submissions/*.json | Leaderboard row, slice payloads, task error count |
💡 Optimize Your Evaluation Logic with OpenJudge To build your own evaluation system beyond the LLM × Harness vertical, you can leverage OpenJudge's 50+ production-ready graders (relevance, tool selection, trajectory, etc.) to evaluate and optimize your custom agents.
python run_bench.py \ --agents qwenpaw openclaw hermes \ --model dashscope/qwen3.6-plus \ --tasks T002 T006
高质量的LLM性能评估工具
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
✅ Apache 2.0 — 宽松开源协议,可商用,需保留版权声明和 NOTICE 文件,含专利授权条款。
总体来看,PawBench 是一款质量优秀的Agent工作流,在同类工具中具备一定竞争力。AI Skill Hub 将持续追踪其更新动态,建议收藏备用,结合自身场景选择合适时机引入使用。
| 原始名称 | PawBench |
| Topics | benchmarkharnessllmpython |
| GitHub | https://github.com/agentscope-ai/PawBench |
| License | Apache-2.0 |
| 语言 | Python |
收录时间:2026-06-09 · 更新时间:2026-06-09 · License:Apache-2.0 · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端