OpenSpace 教程：用 Skills、MCP 和 Lineage 构建自进化 AI 智能体

在本教程中：https://github.com/MARKTECHPOST-AI-MEDIA-INC/AI-Agents-Projects-Tutorials/blob/main/Agentic%20AI%20Codes/HKUDS_OpenSpace_Self_Evolving_AI_Agents_Marktechpost.ipynb，我们构建并检查一个 OpenSpace 流程：https://github.com/HKUDS/OpenSpace，从环境设置和稀疏仓库克隆开始，到实时任务执行、技能演化以及基于 MCP 的代理集成。我们配置模型凭证和工作区变量，以可编辑模式安装项目，调用异步 Python API，并检查 OpenSpace 如何在 SQLite 中存储演化能力，包括版本控制和血统元数据。我们还创建了自定义的 SKILL.md，连接主机代理技能，测试暖任务重用，启动可流式 HTTP MCP 服务器，并分析展示数据库以了解 FIX、DERIVED 和 CAPTURED 技能如何支持低成本、可重用的代理行为。

我们验证 Python 运行时，定义所需的 API 凭证，并配置 OpenSpace 模型及可选的云访问设置。我们使用稀疏检出克隆仓库，以可编辑模式安装包，并确认 OpenSpace 命令行工具可用。然后我们创建工作区和技能目录，编写环境配置文件，导出所需变量，并检测是否启用实时 LLM 执行。

我们在 Google Colab 中初始化异步执行，并定义一个可重用函数，通过 OpenSpace Python API 提交任务。我们运行一个初始的工资生成任务，并检查 OpenSpace 在后续执行分析中演化的任何技能。我们还检查 SQLite 数据库结构，显示存储的记录，并验证技能注册表和类型定义是否可以通过程序访问。

我们提交一个相关的工资任务，以观察 OpenSpace 如何重用或派生之前生成的技能能力。我们创建一个自定义 SKILL.md，指导代理分析 CSV 文件并生成结构化 Markdown 报告。然后我们安装 OpenSpace 主机技能，生成演示数据集，并通过相同的演化代理工作流程执行自定义能力。

我们使用可流式的 HTTP 传输启动 OpenSpace MCP 服务器，并将其绑定到本地 Colab 端点。我们探测该端点以确认服务器进程正在运行，即使一个基本的 HTTP 请求返回的是 MCP 特定的响应状态。我们还生成了一个示例 MCP 主机配置，外部代理可以使用它来访问 OpenSpace 工作区和技能目录。

当存在有效的云 API 密钥时，我们会有条件地将自定义技能上传到 OpenSpace 云社区。我们检查了存储库的 showcase SQLite 数据库，以研究存储的技能、元数据、质量信息以及完整的进化谱系。最后，我们按照来源类型汇总技能，并总结整个教程中建立的 Colab 环境、自定义能力、MCP 集成和自我进化的工作流程。

总之，我们建立了一个实际的 OpenSpace 环境，展示了代理能力如何被执行、持久化、重用以及逐步改进。我们直接使用 Python API、本地技能目录、SQLite 支持的注册表、MCP 传输以及可选的云端技能共享命令，从而能够了解用户端的工作流程以及底层技能引擎的架构。我们还验证了相关任务如何重用之前进化的知识，自定义技能如何在运行时变得可发现，以及谱系记录如何揭示每个能力的开发历史，为在 Colab 中构建自我进化、成本高效的代理系统奠定了坚实的基础。

需要与我们合作推广您的 GitHub 仓库 OR Hugging Face 页面 OR 产品发布 OR 网络研讨会等吗？请联系： https://forms.gle/wbash1wF6efRj8G58

Sana Hassan，是 Marktechpost 的咨询实习生，同时也是 IIT Madras 的双学位学生，热衷于将技术和 AI 应用于解决现实世界问题。他对解决实际问题充满兴趣，为 AI 与现实生活解决方案的交汇带来了新的视角。

In this tutorial,：https://github.com/MARKTECHPOST-AI-MEDIA-INC/AI-Agents-Projects-Tutorials/blob/main/Agentic%20AI%20Codes/HKUDS_OpenSpace_Self_Evolving_AI_Agents_Marktechpost.ipynb we build and examine an OpenSpace ：https://github.com/HKUDS/OpenSpace workflow, progressing from environment setup and sparse repository cloning to live task execution, skill evolution, and MCP-based agent integration. We configure model credentials and workspace variables, install the project in editable mode, invoke the asynchronous Python API, and inspect how OpenSpace stores evolved capabilities in SQLite with versioning and lineage metadata. We also create a custom SKILL.md, connect host-agent skills, test warm-task reuse, launch the streamable HTTP MCP server, and analyze the showcase evolution database to understand how FIX, DERIVED, and CAPTURED skills support lower-cost, reusable agent behavior.

We verify the Python runtime, define the required API credentials, and configure the OpenSpace model and optional cloud access settings. We clone the repository with sparse checkout, install the package in editable mode, and confirm that the OpenSpace command-line tools are available. We then create the workspace and skill directories, write the environment configuration files, export the required variables, and detect whether live LLM execution is enabled.

We initialize asynchronous execution in Google Colab and define a reusable function to submit tasks via the OpenSpace Python API. We run an initial payroll-generation task and inspect any skills that OpenSpace evolves during post-execution analysis. We also examine the SQLite database structure, display stored records, and verify that the skill registry and type definitions are accessible programmatically.

We submit a related payroll task to observe how OpenSpace reuses or derives capabilities from previously generated skills. We create a custom SKILL.md that instructs the agent to analyze CSV files and produce structured Markdown reports. We then install the OpenSpace host skills, generate a demonstration dataset, and execute the custom capability through the same evolving agent workflow.

We start the OpenSpace MCP server using the streamable HTTP transport and bind it to a local Colab endpoint. We probe the endpoint to confirm that the server process is running, even when a basic HTTP request returns an MCP-specific response status. We also generate an example MCP host configuration that external agents can use to access the OpenSpace workspace and skill directories.

We conditionally upload the custom skill to the OpenSpace cloud community when a valid cloud API key is available. We inspect the repository’s showcase SQLite database to study the stored skills, metadata, quality information, and complete evolution lineage. We finally aggregate skills by origin type and summarize the Colab environment, custom capability, MCP integration, and self-evolving workflow that we establish throughout the tutorial.

In conclusion, we established a practical OpenSpace environment that demonstrates how agent capabilities are executed, persisted, reused, and progressively improved. We worked directly with the Python API, local skill directories, SQLite-backed registries, MCP transports, and optional cloud skill-sharing commands, giving us visibility into both the user-facing workflow and the underlying skill-engine architecture. We also verified how related tasks can reuse previously evolved knowledge, how custom skills become discoverable at runtime, and how lineage records expose the development history of each capability, leaving us with a strong foundation for building self-evolving, cost-efficient agent systems in Colab.

Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us ：https://forms.gle/wbash1wF6efRj8G58

Sana Hassan, a consulting intern at Marktechpost and dual-degree student at IIT Madras, is passionate about applying technology and AI to address real-world challenges. With a keen interest in solving practical problems, he brings a fresh perspective to the intersection of AI and real-life solutions.