针对 AI 编码代理日益融入软件开发流程的现状，有开发者提出了“任务即代码”（Tasks as Code）的工作流概念 1。该模式主张将开发任务和工单以 Markdown 文件的形式直接存储在 Git 仓库中，而非依赖传统的独立项目管理平台 1。通过这种方式，AI 编码代理可以在同一位置读取需求、规划实现并更新文档，确保所有任务变更均被版本化且可审查 1。这一理念借鉴了“文档即代码”的实践，同时也契合规范驱动开发（Spec-Driven Development, SDD）的思想，其中任务文件本身即可作为规格说明使用 1。
在实际操作层面，文章建议采用一套包含五个步骤的标准工作流：首先描述任务，接着分析需求，然后创建实施计划，经人工审查批准后，最后由代理执行实施并报告结果 1。为了保持数据所有权归 Git 而非特定平台，并实现访问权限隔离，架构上推荐使用三个独立的 Git 仓库分别存储代码、文档和任务；具体做法是将文档和任务仓库本地克隆至代码仓库的子目录中，并在 .gitignore 文件中予以排除 1。协作方面，开发者可利用 IDE、GitHub 以及基于浏览器的界面（如 Typemill）进行交互，文中提及的辅助工具还包括 Claude Code、OpenCode 和 Spec Kit 1。随着任务全面进入代码仓库，预计未来将出现针对此模型的新软件和界面，而传统工单系统中如看板、冲刺管理等功能，可能会因工作主要由 AI 代理主导而减少需求 1。
A concept titled "Tasks as Code" proposes storing development tasks and tickets directly within a Git repository as Markdown files 1. This approach enables AI coding agents to read requirements, plan implementations, and update documentation from the same location where the code resides 1. By treating task descriptions similarly to source code, the method ensures that all changes to work items are version-controlled and auditable, mirroring the established practice of "Docs as Code" 1. The model aligns with Spec-Driven Development (SDD), positing that task files themselves can serve as specifications for software projects 1.
To implement this strategy, developers are encouraged to follow a five-step process: describing the task, analyzing requirements, creating an implementation plan, reviewing and approving the plan, and finally executing the task while reporting results 1. The architecture recommends maintaining three separate Git repositories for code, documentation, and tasks; these can be cloned into subdirectories of the main code repository and excluded via .gitignore to preserve access control boundaries 1. Various tools such as Claude Code, OpenCode, Typemill, and Spec Kit are cited as potential enablers for this collaborative environment 1. Proponents argue that keeping data ownership within Git rather than specific platforms facilitates integration with IDEs, GitHub, and browser-based interfaces 1. As tasks migrate into repositories, the industry may see a decline in demand for traditional ticketing features like kanban boards and sprints, shifting focus toward agent-driven workflows 1.
评论
还没有评论，欢迎留下第一条。