四层能力架构Four-Layer Capability Architecture
基础层Foundation Layer
理解前端、Git、API、数据文件和部署的基本概念，重点是会用 Codex、Claude Code、Antigravity、Cursor 读项目、改页面、跑命令、查错误。Understand basic concepts of front-end, Git, APIs, data files, and deployment, with the focus on using Codex, Claude Code, Antigravity, and Cursor to inspect projects, edit pages, run commands, and debug errors.
交付层Delivery Layer
把需求澄清、Prompt to Product、Agent 工作流、Demo、README、测试清单和演示脚本连成一条可交付链路。Connect discovery, Prompt to Product, agent workflow, demos, README files, test checklists, and demo scripts into one delivery chain.
企业层Enterprise Layer
学习企业现场常见概念：数据来源、权限、安全、日志、评估和回滚。目标不是成为云架构师，而是能指挥 AI 工具完成可验收交付。Learn common field concepts: data sources, access, security, logs, evaluation, and rollback. The goal is not to become a cloud architect, but to direct AI tools toward acceptable delivery.
行业层Industry Layer
把跨境电商、文旅、教育和企业 AI 转型诊断做成可复用行业实验室，每个行业都有案例、数据、Prompt、Demo 和评分标准。Turn commerce, culture-tourism, education, and enterprise AI diagnosis into reusable labs with cases, data, prompts, demos, and assessment standards.
COURSE ENTRY
如果你已经理解能力全景，下一步进入 Learn，按阶段完成课程和项目。 Once you understand the capability map, move to Learn and complete lessons and projects by stage.
进入学习路径Diagnostic Mindset：FDE 的三问Diagnostic Mindset: The Three Whys
事实来源在哪里？Where is the system of record?
如果核心数据只在某个人的 Excel、微信群截图或本地 ERP 导出里，项目风险已经出现。FDE 要先确认数据来源、责任人、更新时间和可信度。If core data lives in one person's Excel file, chat screenshots, or local ERP exports, the project is already at risk. The FDE must verify source, owner, freshness, and trustworthiness.
不做这件事的代价是什么？What is the cost of inaction?
如果没有明确代价，项目很容易变成好看的 Demo。FDE 要问清楚它影响收入、成本、时间、错误率、体验还是合规。Without a clear cost, the project becomes a nice demo. The FDE asks whether it affects revenue, cost, time, error rate, experience, or compliance.
第 2 天谁来维护？What does Day 2 look like?
上线之后谁看日志、谁改 Prompt、谁补数据、谁处理投诉？没有内部 owner 的 AI 项目，通常会在 FDE 离开后失效。After launch, who checks logs, edits prompts, fills data gaps, and handles complaints? Without an internal owner, AI projects often fail after the FDE leaves.
Forward Deployment Discovery ChecklistForward Deployment Discovery Checklist
组织与政治Organization and Politics
- 项目 Champion 是谁？谁真的愿意推动试点？Who is the champion who will fight for the pilot?
- 潜在 Blocker 是哪个部门？IT、法务、财务还是业务主管？Which function may block it: IT, legal, finance, or business owner?
- 成功指标是什么：更快响应、更高转化、更低错误率还是更少人工？What is success: faster response, higher conversion, lower error rate, or less manual work?
数据与安全Data and Security
- 数据分级是什么？是否包含个人信息、客户资料、合同、财务或敏感业务数据？How is the data classified? Does it include personal, customer, contract, finance, or sensitive business data?
- 数据如何进入系统：本地 Excel/CSV、ERP 导出、数据库只读账号、飞书表格还是 API？How does data enter the system: local Excel/CSV, ERP export, read-only database account, Feishu table, or API?
- 是否需要脱敏、最小权限、访问日志和人工复核？Do we need masking, least privilege, access logs, and human review?
基础设施Infrastructure
- 先让 AI 工具基于样本文件做数据审计，还是已经有安全的企业数据接口？Should AI tools audit sample files first, or is there already a safe enterprise data interface?
- 最小可行架构是静态站 + API，还是阿里云/本地服务 + 数据库？Is the MVA static site + API, or cloud/local service + database?
- 能否让 Codex、Claude Code、Antigravity 或 Cursor 帮你检查构建、环境变量、日志和回滚步骤？Can Codex, Claude Code, Antigravity, or Cursor help inspect builds, env vars, logs, and rollback steps?
AI 与评估AI and Evaluation
- 有没有 Golden Dataset 或历史样本用于评估？Is there a golden dataset or historical sample for evaluation?
- 工具调用路径是否可追踪？失败时能否回放？Can tool calls be traced and replayed after failure?
- 上线前谁验收？上线后如何发现模型退化、幻觉或数据漂移？Who accepts before launch, and how do we detect degradation, hallucination, or drift after launch?
Case Study 参考Case Study References
Artifact TemplatesArtifact Templates
TEMPLATE LIBRARY
完整模板集中放在 Artifact Templates 页面，避免同一批模板在 Roadmap、Resources 和 Templates 多处重复维护。 The full template library lives on the Artifact Templates page to avoid maintaining the same templates across Roadmap, Resources, and Templates.
Open Artifact Templates