Agent Skills: Cloudflare Workers

Cloudflare Workers

Use this skill to deliver Cloudflare Workers tasks with production-safe defaults and minimal setup churn.

Execute the workflow

1. Classify the request.
2. Bootstrap or inspect the project.
3. Configure wrangler.jsonc and bindings.
4. Implement handlers and business logic.
5. Validate locally and with tests.
6. Deploy and verify in logs.

Classify the request

Identify the primary task before making changes:

• New project or scaffold
• Binding configuration or migration
• Runtime API usage
• Framework integration
• Deployment or CI/CD
• Debugging and observability

Use the routing table below to load only the needed reference file.

Route to references on demand

Load only one file at a time unless the task clearly spans multiple domains.

• references/bindings.md: KV, R2, D1, Durable Objects, Queues, AI, Vectorize, Service Bindings, Browser Rendering, Hyperdrive, Analytics
• references/runtime-apis.md: fetch, scheduled, queue, Cache API, HTMLRewriter, WebSockets, Streams, Crypto, Node compatibility
• references/frameworks.md: React/Vite, Vue/Vite, Next.js, Astro, SvelteKit, Hono, React Router, Nuxt, Solid Start, Express adapters
• references/patterns.md: API patterns, caching, auth, rate limiting, validation, repository patterns, RAG, performance, security

Bootstrap quickly

Prefer the bundled scaffold when the user wants a starter project:

bash scripts/scaffold-worker.sh <project-name> [api|spa|fullstack|hono]

Then run:

cd <project-name>
npm install
npx wrangler dev

Use npm create cloudflare@latest when the user explicitly asks for official templates.

Configure Wrangler correctly

Use wrangler.jsonc and always set:

• name
• main
• compatibility_date

Apply these rules:

• Keep non-sensitive config in vars
• Store secrets with wrangler secret put
• Keep bindings in wrangler.jsonc aligned with the Env interface in code
• Regenerate bindings types after changes with npx wrangler types

Implement handlers with clear boundaries

Start with fetch; add scheduled and queue only if needed.

• Keep routing thin
• Move logic to modules/services
• Use ctx.waitUntil for non-blocking async work
• Use prepared statements for D1
• Return explicit status codes and content types

Validate before deploy

Run a tight verification loop:

npx wrangler dev
npm test
npx wrangler deploy --dry-run

If bindings depend on remote resources, test with remote mode when necessary.

Deploy safely

Follow staged deployment:

1. Deploy to staging environment.
2. Run smoke checks on key routes.
3. Deploy to production.
4. Tail logs and confirm no regression.

Useful commands:

npx wrangler deploy
npx wrangler deploy -e staging
npx wrangler tail

Debug systematically

When debugging, execute this order:

1. Confirm request path and method handling.
2. Confirm binding names and IDs in wrangler.jsonc.
3. Confirm Env interface matches configured bindings.
4. Confirm secrets and environment-specific config.
5. Re-run type generation and restart dev server.
6. Tail logs and inspect failing requests.

Use production guardrails

• Never store secrets in vars.
• Validate external input before use.
• Keep queue consumers idempotent.
• Define cache behavior intentionally.
• Limit per-request work and parallelize I/O with care.

Return useful outputs

When responding to users, provide concrete artifacts:

• Minimal patch-ready config snippets
• Exact command sequence for local run and deploy
• Clear list of required bindings and secrets
• Verification checklist (what to test and where to look)