agentscope-agent-test-harness

根据现有 AgentScope 智能体模块生成轻量测试脚本，用假模型替代真实模型调用，断言工具调用、消息格式与状态转移，可用 pytest 离线运行，并输出通过/失败报告。

热度 7.64

最近更新 2026-10-03 10:10:30

分类：开发依赖：代码执行 / 读写文件 / 内部调用大模型应用场景：代码生成 / 调试排错 / 自动化脚本

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add agentscope-agent-test-harness

Metadata

name: agentscope-agent-test-harness
description: Create and run lightweight tests for AgentScope agents without hitting external APIs. Use this whenever the user wants to validate agent wiring, tool binding, or prompt formatting before live model calls.
version: 0.1.0
metadata: {"openclaw":{"requires":{"env":["OPENAI_API_KEY"],"bins":["python3"]},"primaryEnv":"OPENAI_API_KEY","envVars":[{"name":"OPENAI_API_KEY","required":true,"description":"OpenAI or compatible API key."},{"name":"AGENTSCOPE_TEST_MODE","required":false,"description":"When set, skip live model calls and run offline assertions only."}],"emoji":"🧪","homepage":"https://github.com/agentscope-ai/agentscope"}}

This skill helps you test AgentScope agents with mocked models and offline assertions.

Run the harness generator from this skill folder:

python3 scripts/generate_harness.py --agent /path/to/agent_module.py --output /path/to/test_harness.py

Then run:

python3 -m pytest /path/to/test_harness.py