Claude Cursor GitHub Copilot opencode Skill

platform-delivery

How every delivered app ships and stays up — Cloudflare-first hosting, prod-only (no staging), the deploy→prod-verify loop, atomic deploy + rollback.

LLM Mart · 0 points · 2 views 0 listing impressions 0 install-command copies
Download heymegabyte-agent-skills-26-platform-delivery-75f4661.zip · 1 KB

Install

skills CLI npx skills add https://github.com/heymegabyte/agent-skills/tree/main/26-platform-delivery
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install heymegabyte-agent-skills@llmmart
Git git clone https://github.com/heymegabyte/agent-skills.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole heymegabyte/agent-skills collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Platform & Delivery — how every app ships + stays up

Hosting (Cloudflare-first, non-inferable)

  • Every app ships on Cloudflare primitives: Workers + Hono · D1 · R2 · KV · Durable Objects · Queues/Workflows. Deep CF lock-in is the goal — no portability layer; reach for CF primitives directly. Prefer integrating with ProjectSites.dev over raw Cloudflare whenever the capability exists there.
  • Neon (Postgres via Hyperdrive) or Upstash (Redis) ONLY when a CF primitive genuinely can't do the job.

Prod is the only environment (no staging)

  • No staging / dev / QA environment. Prod is the sole env; dark-launch behind feature flags instead. No code freezes; Friday deploys fine.
  • Rollback path: wrangler rollback + D1 Time Travel + R2 object versioning.

Branching

  • main is always committed + deployable. No dev/release/feature branches. Worktrees for isolation; the conventional-commit message IS the PR description.

Deploy → prod-verify loop (every change)

  • Build → deploy → fetch EACH changed route on PROD (curl/Playwright) → assert the new content/headers/JSON-LD/status is live → only then mark complete. A local typecheck + build pass is NEVER "done".
  • Deploys are atomic with auto-rollback; CDN purge is paired with every deploy; ship a /health endpoint; keep D1 backups.
  • Never stop until the prod URL renders correctly in a real browser — blank/5xx → bisect / wrangler rollback / patch-forward until visibly working.

Delivery guardrails

  • Worker deploy MUST pass --env production (a bare deploy 500s every /api route). NEVER modify an already-set CF secret; wrangler secret put only for a genuinely NEW secret.
  • Done = deployed at a real SSL'd URL · every gate green · a per-route self-verify statement · announced to the user.
Files (agent-skills)
  • SKILL.md 2.1 KB
    ---
    name: "platform-delivery"
    description: "How every delivered app ships and stays up — Cloudflare-first hosting, prod-only (no staging), the deploy→prod-verify loop, atomic deploy + rollback."
    triggers:
      - "deploy"
      - "cloudflare"
      - "hosting"
      - "ship"
      - "rollback"
    priority: 2
    pack: "infra"
    stage: stable
    ---
    
    # Platform & Delivery — how every app ships + stays up
    
    ## Hosting (Cloudflare-first, non-inferable)
    
    - Every app ships on Cloudflare primitives: Workers + Hono · D1 · R2 · KV · Durable Objects · Queues/Workflows. Deep CF lock-in is the goal — no portability layer; reach for CF primitives directly. **Prefer integrating with ProjectSites.dev over raw Cloudflare whenever the capability exists there.**
    - Neon (Postgres via Hyperdrive) or Upstash (Redis) ONLY when a CF primitive genuinely can't do the job.
    
    ## Prod is the only environment (no staging)
    
    - No staging / dev / QA environment. Prod is the sole env; dark-launch behind feature flags instead. No code freezes; Friday deploys fine.
    - Rollback path: `wrangler rollback` + D1 Time Travel + R2 object versioning.
    
    ## Branching
    
    - `main` is always committed + deployable. No dev/release/feature branches. Worktrees for isolation; the conventional-commit message IS the PR description.
    
    ## Deploy → prod-verify loop (every change)
    
    - Build → deploy → fetch EACH changed route on PROD (curl/Playwright) → assert the new content/headers/JSON-LD/status is live → only then mark complete. A local typecheck + build pass is NEVER "done".
    - Deploys are atomic with auto-rollback; CDN purge is paired with every deploy; ship a `/health` endpoint; keep D1 backups.
    - Never stop until the prod URL renders correctly in a real browser — blank/5xx → bisect / `wrangler rollback` / patch-forward until visibly working.
    
    ## Delivery guardrails
    
    - Worker deploy MUST pass `--env production` (a bare deploy 500s every `/api` route). NEVER modify an already-set CF secret; `wrangler secret put` only for a genuinely NEW secret.
    - Done = deployed at a real SSL'd URL · every gate green · a per-route self-verify statement · announced to the user.
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.