菜单
CKS production stack — 26-layer reference: design tools, build, infrastructure, security, observability, AI/LLM, payments.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/cardinalconseils/claude-starter --skill cks-stack命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
Ecosystem monitoring — breaking changes, deprecations, security alerts, opportunity tracking, bulletin generation.
Project kickstart — idea to scaffolded project: research, monetization, brand, ERD, architecture, deploy config.
Unified Dashboard SaaS Build Sequence — 12-stage methodology for multi-role SaaS products (super admin/admin/user/vendor). Enforces single application with role-gated dashboard visibility, never separate dashboards per role, generated from one permissions matrix. Use for multi-role SaaS, admin/user/vendor roles, role-gated dashboard, permission matrix, tenant roles, super admin panel design.
| name | cks-stack |
| description | CKS production stack — 26-layer reference: design tools, build, infrastructure, security, observability, AI/LLM, payments. |
| allowed-tools | Read |
The CKS recommended tech stack for production web applications. 26 layers covering every production concern a vibecoder needs to own.
What vibe coders think "full-stack" means: Frontend + Backend. What production actually requires: every layer below, owned by you.
Cost principle: Free tiers are for validation only. The moment you have real users, assume you're paying. Design to minimize what scales with usage: egress, function invocations, LLM tokens, storage reads.
| Group | Layers | Focus |
|---|---|---|
| A — Build & Design | 1–5 | Design, Frontend, APIs, Database, Auth |
| B — Infrastructure | 6–9 | Hosting, CI/CD, Cloud, API Testing |
| C — Security | 10–13 | RLS, Rate Limiting, Caching, Scaling |
| D — Observability | 14–16 | Error Tracking, AI Tracing, E2E Testing |
| E — AI & LLM | 17–20 | LLM Gateway, Providers, Fast Inference, Voice |
| F — Payments | 21 | Stripe |
| G — Communication | 22–23 | Email, Internal Comms |
| H — Automation | 24 | n8n |
| I — Sales | 25 | Apollo.io |
| J — Dev Tools | 26 | Claude Code |
Read references/stack-layers.md for tool choices, cost notes, and consolidation decisions per layer.
| Maturity | Required Groups |
|---|---|
| Prototype | A + B (build and deploy working) |
| Pilot | A + B + C security basics + D error tracking (Sentry) |
| Candidate | A + B + C + D + E (if AI) + F (if payments) + G email |
| Production | All relevant groups — no uncovered critical layers |
| Rationalization | Reality |
|---|---|
| "We'll add monitoring later" | You will not notice outages until users report them. Set up Sentry before first real user. |
| "Rate limiting can wait" | One abuse incident costs more than a year of Upstash. Wire it at Pilot. |
| "We don't need email yet" | Magic links, receipts, and password resets are day-one features. Resend takes 10 minutes. |
| "I'll figure out the LLM costs later" | Token costs are the #1 bill-spike risk. Model them before writing LLM code. |
基于 SOC 职业分类