---
spec_name: "Talk-to-Build Stack — Implementation Spec for Claude Code"
spec_version: "1.0"
last_updated: 2026-05-07
target_audience: "Designers, creative directors, and Shopify / WordPress / Webflow site builders setting up an AI-native build stack"
expected_outcome: "Mike's full Talk-to-Build stack installed and running on the user's project — Claude Code wired in, Claude Desktop talking to daily tools, the right platform AI selected, Netlify deploying, Claude Security gating launches"
ai_tools_supported: [claude_code, google_antigravity, cursor]
based_on: https://mikekwal.com/blog/talk-to-build-stack/
author: Mike Kwal
---

# Talk-to-Build Stack — Implementation Spec for Claude Code

## Instructions for the AI

You are helping a designer or creative director set up Mike Kwal's Talk-to-Build stack on their project. Read this spec end-to-end before touching anything. Confirm each step before moving on. The user runs client sites on Webflow, WordPress, Shopify, or a custom Netlify-hosted app — ASK which one before making platform-specific changes.

Tone: senior teammate explaining as you go. One sentence of context per change so the user learns the stack while you build it.

---

## Step 1: Install Claude Code and point it at the project

**What it does:** Claude Code is the terminal-based builder that reads the user's whole codebase, plans changes, and ships them. The engine of the stack.

**Tasks:**

1. Confirm the user has a paid Claude.ai plan (Pro or Max). Claude Code is bundled — see https://www.anthropic.com/pricing.
2. Install the CLI:
   ```bash
   # Mac / Linux
   curl -fsSL https://claude.ai/install.sh | sh

   # Windows (PowerShell)
   irm https://claude.ai/install.ps1 | iex
   ```
3. Run `claude login` and complete the OAuth flow.
4. `cd` into the project root and run `claude init` to generate a `CLAUDE.md`. Populate it with: stack, brand voice (one paragraph), deploy commands, and any "do not touch" files.

**Verify:** `claude` in the project directory responds with awareness of the project files.

---

## Step 2: Set up Claude Desktop as the daily co-pilot

**What it does:** Claude Desktop is the chat window that talks to Slack, Gmail, Google Calendar, Monday.com, Figma, and Fireflies through native MCP integrations. Where the day starts.

**Tasks:**

1. Download from https://claude.ai/download. Sign in with the same account as Claude Code.
2. Settings → Connectors. Enable only the integrations the user actually uses (Slack, Gmail, Google Calendar, Drive, Figma is the common set). Fewer connectors = faster Claude.
3. Test the wiring: *"Summarize my Slack DMs from the last 24 hours and pull anything client-related into a list."*

**Verify:** Claude Desktop returns real Slack data, not "I don't have access to Slack."

---

## Step 3: Pick the right platform AI for the site type

The decision tree. Route on the user's answer to "what platform is this site on?"

### Webflow
Use **Webflow AI Site Builder** (https://webflow.com/ai). Best for marketing sites and 5–15 page content builds.
- Webflow workspace → New Site → Build with AI. Feed it the brief, brand colors, audience.
- Schema, meta tags, and clean HTML come baked in.
- Drop into Claude Code for custom logic the visual builder can't reach (forms, integrations, custom AEO schema).

### Shopify
Use **Shopify Sidekick** (built into the admin). Best for product pages, theme tweaks, storefront copy.
- Click the Sidekick icon (top right of admin).
- Theme work: *"Edit theme.liquid to inject this schema in the head — [paste schema]."*
- Bulk copy: *"Rewrite all product descriptions in [collection] using inverted-pyramid AEO."*
- For anything Sidekick can't ship, fall back to Claude Code editing theme files locally.

### WordPress
Use **WordPress AI / Abilities API plugins** (WP Engine's AI Toolkit, or any WP 6.5+ Abilities-API plugin). Best for content sites where the client logs in and updates with natural language.
- Install the AI Toolkit plugin. Settings → AI → API Key (use the user's Anthropic key).
- Use it for content + SEO/AEO updates.
- For theme/plugin code, drop into Claude Code locally on `functions.php`, custom blocks, templates.

### Custom app (no CMS)
Skip platform AI. The whole build runs through Claude Code, Cursor, and Netlify.

**Verify:** The user completes one small task on their site using the platform AI alone (e.g., change the hero headline).

---

## Step 4: Wire up Netlify for deployment

**What it does:** Netlify is where Talk-to-Build apps live. Agent Runners auto-fix failed builds.

**Tasks:**

1. Sign up at https://app.netlify.com/signup (free tier covers most builds).
2. Connect the repo:
   ```bash
   netlify init
   ```
   Pick "Create & configure a new site." Confirm build command + publish directory (Claude Code can read `package.json` and answer this).
3. Enable Agent Runners: Site settings → Build & deploy → Agent Runners → ON.
4. Set environment variables:
   ```bash
   netlify env:set ANTHROPIC_API_KEY [key]
   ```
5. Push to git. Confirm the preview URL works.

**Verify:** Netlify preview URL loads. A small change auto-builds.

---

## Step 5: Establish the daily workflow patterns

Here's what I'd run on day 1 to lock in the rhythm.

**Morning brief (Claude Desktop, 10 min):**
```
Sweep Slack DMs, Gmail unreads, today's calendar. Pull client-related
items into a list. Flag same-day replies. Push action items to
Monday.com board [BOARD ID].
```

**Project work (Claude Code, 1–2 hour blocks):**
```
Read this project. I want to [DESCRIBE THE CHANGE].
Plan it first. Show me the plan. Ship after I approve.
```

**Bulk content (DeepSeek or Claude, batch):**
```
Rewrite all 50 product descriptions in [collection] using
inverted-pyramid AEO. Output as a CSV I can import.
```

**Copy review (GPT-5.5):** Paste hero/CTA into ChatGPT. Ask: *"Rewrite for tone — direct, designer audience, no fluff."* Pick the better of the two.

---

## Step 6: Run Claude Security before every launch

**What it does:** Claude Security (public beta on Opus 4.7) reads the codebase and flags leaked secrets, missing input validation, exposed admin routes. Free with any paid Claude plan.

**Tasks:**

1. In Claude Code: `claude security scan`. Or paste the codebase into Claude Desktop with: *"Run a security review. Flag everything I should fix before launch — secrets, auth gaps, input validation, rate limiting, exposed admin routes."*
2. Triage: fix critical/high before launch, defer low-priority into a backlog.
3. Re-run after fixes. Confirm clean.

**Verify:** No critical or high findings on the latest scan.

---

## Step 7: Document the stack in the project README

So the user (and any future teammate) can onboard fast. Add to `README.md`:

```markdown
## Stack
- Claude Code — main builder, terminal
- Claude Desktop — daily co-pilot, native integrations
- [Webflow AI / Shopify Sidekick / WordPress AI Toolkit] — site shell + content
- Netlify — deploy + Agent Runners
- Claude Security — pre-launch scan

## Daily commands
- `claude` — open Claude Code in this project
- `netlify deploy --prod` — ship to production
- `claude security scan` — pre-launch security pass
```

---

## Quick-start prompts for Claude Code / Google Antigravity / Cursor

### Prompt 1 — Initialize project memory
```
Read every file in this project. Generate a CLAUDE.md at the root
capturing: stack, framework, deploy target, brand voice, build/deploy
commands, and files I should never touch. Show me the result so I can edit.
```

### Prompt 2 — Audit my current AI stack
```
Look at my package.json, git config, and home directory for existing AI
tooling (Claude Code, Cursor, Codex, Copilot, Replit, Continue.dev, etc.).
Tell me what I have, what's redundant, and what to drop based on Mike Kwal's
Talk-to-Build Stack spec.
```

### Prompt 3 — Wire Netlify to this project
```
Read my project. Confirm framework, build command, publish directory.
Walk me through `netlify init` and create the site. Enable Agent Runners.
Confirm the preview URL works.
```

### Prompt 4 — Pick my platform AI
```
My client site is on [Webflow / Shopify / WordPress / custom]. Based on
the Talk-to-Build Stack spec, tell me which platform AI to use, what to
install, and walk me through one small win to confirm it works.
```

### Prompt 5 — Run the pre-launch security pass
```
Run Claude Security on this codebase. Flag every critical and high finding.
For each, explain the risk in plain English and propose the fix. Do NOT
auto-fix — show me the plan first.
```

### Prompt 6 — Build the morning brief routine
```
Build a Claude Desktop prompt I paste every morning that:
- Sweeps Slack DMs from the last 24 hours
- Sweeps Gmail unreads
- Pulls today's calendar
- Outputs one triaged list with same-day-reply flags
- Pushes action items to my Monday.com board [BOARD ID]
Save it as /prompts/morning-brief.md.
```

---

## Verification checklist

| Check | How |
|---|---|
| Claude Code installed + authed | `claude` in the project responds with project awareness. |
| `CLAUDE.md` exists at root | File describes stack, voice, deploy target. |
| Claude Desktop integrations live | Asks Claude Desktop for Slack DMs — returns real data. |
| Platform AI picked + tested | One small change shipped via Webflow AI / Sidekick / WP AI. |
| Netlify connected | `netlify status` shows the linked site. Preview URL loads. |
| Agent Runners enabled | Site settings → Build & deploy → Agent Runners is ON. |
| Claude Security ran clean | No critical or high findings on the latest scan. |
| README documents the stack | New teammate can read the README and know what's installed and why. |

---

## Need help?

Four ways to go deeper:

- **Free** — Join the [Talk-to-Build community](https://www.skool.com/talktobuild). Designers and creative directors shipping AI-built sites together. Weekly office hours, prompt library, build-along sessions.
- **Done-for-you** — [MK-Way](https://mk-way.com) installs and runs this stack on your client work. AEO-ready websites in 7 days.
- **Quick question** — DM Mike on [Instagram](https://www.instagram.com/mikekwal). He reads every message.
- **B2B / strategy** — Connect on [LinkedIn](https://www.linkedin.com/in/mikekwal/) for deeper conversations on AI in design and agency work.

---

*This spec is the actionable implementation companion to the Talk-to-Build Stack hub article: [https://mikekwal.com/blog/talk-to-build-stack](https://mikekwal.com/blog/talk-to-build-stack/). The hub explains the why and the what; this spec ships the how.*

*Last updated: 2026-05-07*