使用方法
- 复制安装命令,在自己的智能体环境安装前先检查源码包。
- 从原始文档核对所需工具、凭据和支持的宿主。
- 说明输入材料与期望交付物,并对照来源检查输出。
npx skills add https://github.com/promptdriven/pdd --skill pdd-graph开始前需了解
- 安装不包含模型订阅、API 额度或访问文件的权限。
- 仓库星数和安装量反映活跃情况,不等于独立质量或安全测试。
按源仓库内容呈现,保留标题、案例、代码、表格、链接以及原文引用的演示图片。
Use the supported CLI and its saved model. Do not redraw or infer the graph with a model.
- "Show dependency graph": run
pdd graph --openin the chat repository. - "Why is issue 42 waiting?": run
pdd graph --issue 42and explain its saved reasons with issue/evidence links. - "What is next?" or "What can run?" in dependency context: run
pdd graph; separate dependency eligibility from the current owner's scheduling, room and approval checks.
An explicitly named repository uses graph-only --repo OWNER/REPO. Without a checkout or a named repository, ask for that repository. Do not choose a folder or start setup/login/clone.
On a headless VM, relay the exact laptop command returned by PDD. Do not claim the VM opener opens a laptop. Cached, stale, unchecked and offline facts stay labeled. Stale review means that something the review depends on changed: an input it used, an issue it names as a prerequisite, a link that a person changed to or from its issue, a person's answer to its question, or a trusted impact that names it as affected; an input that cannot be read also makes it stale. A new or unrelated issue, or a link that PDD writes for another issue's review, never causes it. A missing review is not a completed empty review.
A graph request is read-only. Never invoke dependency prepare/complete, AI analysis, GitHub mutation or a new workflow to answer it. If analysis is needed, explain the separate intake action and leave the read result honest.