Svelte5 + Cloudflare

Use this skill for any SvelteKit-on-Cloudflare task.

What To Do First

1. Identify runtime model before touching config:

• Worker service: use wrangler secret put and wrangler deploy.
• Pages project: use wrangler pages secret put and Pages build output config.

2. Run scripts/doctor.sh from this skill and follow fixes in order.
3. If user asks for “latest” behavior/tools, verify with official docs before advising.

Guardrails (Do Not Skip)

1. Do not use mode: 'no-cors' to hide integration errors.
2. For SvelteKit API routes, prefer same-origin POST from frontend to /api/....
3. Validate server input and treat upstream logical errors as failures.
4. Keep secrets in Cloudflare secrets, never in git or [vars] for sensitive values.
5. For Worker deploys of SvelteKit cloudflare output, deploy both worker entry and assets.

Standard Implementation Pattern

A) SvelteKit Cloudflare config

• Adapter: @sveltejs/adapter-cloudflare
• Build output used for manual worker deploy: .svelte-kit/cloudflare
• Worker entry file: .svelte-kit/cloudflare/_worker.js

B) Waitlist/form backend pattern

• Frontend posts to same-origin /api/waitlist with URLSearchParams.
• API route validates fields, enforces honeypot, forwards upstream server-side.
• API route returns structured errors:
  • 400 validation
  • 503 missing env
  • 502 upstream/network failure

C) Cloudflare secrets and deploy

• Worker service secret:
  • npx wrangler secret put GOOGLE_APPS_SCRIPT_URL --name <worker-name>
• Worker deploy for SvelteKit output:
  • npx wrangler deploy .svelte-kit/cloudflare/_worker.js --name <worker-name> --assets .svelte-kit/cloudflare

Decision Tree: Pages vs Worker

• If dashboard route is workers.dev and shows Worker versions, treat as Worker service.
• If using Pages project/deployments pipeline, treat as Pages.
• If unsure, run both list commands and use whichever returns the target.

Auto-Maintenance Protocol (Living Skill)

Run scripts/refresh-skill-state.sh whenever:

• SvelteKit/adapter/wrangler versions change
• Deploy command changes
• Pages/Worker mode changes
• A new incident is discovered

Then update references/playbook.md with:

1. New failure symptom
2. Root cause
3. Detection command
4. Minimal fix

When this skill is invoked, agents should proactively run scripts/doctor.sh and patch drift before feature work.

References

• Operational playbook: references/playbook.md
• Drift/state snapshot: references/state.json