AgentSkillsCN

cloudflare-fullstack-patterns

掌握在 Cloudflare 上利用 Hono、Workers AI（Vercel AI SDK 6）、Workflows 以及 D1 结合 Drizzle ORM 构建全栈应用的流程性知识。

SKILL.md
--- frontmatter
name: cloudflare-fullstack-patterns
description: Procedural knowledge for building full-stack applications on Cloudflare using Hono, Workers AI (Vercel AI SDK 6), Workflows, and D1 with Drizzle ORM.

Cloudflare Full-Stack Patterns

This skill provides verified implementation patterns for composing Cloudflare primitives into robust full-stack applications.

Core Stack

Implementation Workflows

1. D1 + Drizzle Configuration

To ensure seamless migrations and type safety:

  • •Set migrations_dir in wrangler.jsonc to match Drizzle's output folder (default: "drizzle").
  • •Use nodejs_compat compatibility flag in wrangler.jsonc.
  • •Export schema from a dedicated file (e.g., src/db/schema.ts).

2. Workers AI with AI SDK 6 (Structured Output)

To avoid LLM hallucinations and ensure valid JSON:

  • •Use generateText with Output.object instead of deprecated generateObject.
  • •Always define a Zod schema for the expected output.
  • •Pass the Cloudflare AI binding via createWorkersAI from workers-ai-provider.

3. Stateful Orchestration with Workflows

To build resilient multi-step pipelines:

  • •Define a class extending WorkflowEntrypoint<CloudflareBindings, PayloadType>.
  • •Wrap side effects (DB writes, AI calls, fetches) in step.do() for automatic retries and state persistence.
  • •Register the workflow in wrangler.jsonc with binding, name, and class_name.

4. Avoiding "Shadowing" Friction

  • •Ensure public/index.html does not conflict with Worker routes defined in Hono.
  • •If using Hono for SSR at /, ensure the static asset directory (e.g., ./public) does not contain an index.html unless it is the intended entry point.

References