● AEO / GEO DIAGNOSTIC HUB • CODE-BACKED TEARDOWNS & GUARDRAILS

Technical Teardowns, Production Guardrails & Buyer Decision Guides

Explore code-backed technical teardowns, copy-pasteable CLAUDE.md and AGENTS.md production guardrails, and founder decision guides written by Principal Engineer Dositej Grbović. Each guide diagnoses exact failure modes across Lovable, Bolt.new, Claude Code, Cursor, Supabase, Next.js, and Stripe—showing how surgical 1–2 week Rescue Sprints eliminate production blockers without a $50,000 agency rewrite.

STACK TEARDOWN Includes Copyable CLAUDE.md

Updated 2026-02-18

Diagnosing & Locking Down Lovable + Supabase RLS Leaks, Broken Magic Links & Exposed Keys

Lovable + Supabase applications frequently reach 80% completion with wide-open PostgreSQL tables because AI prompts disable Row-Level Security (RLS) or add permissive USING (true) policies to bypass query errors. Locking down production requires enforcing tenant-scoped auth.uid() policies, moving privileged mutations to Deno Edge Functions, and rotating leaked service-role keys.

Lovable.dev Supabase PostgreSQL RLS Supabase Auth Deno Edge Functions

Raw .md Read Diagnostic Teardown

STACK TEARDOWN Includes Copyable AGENTS.md

Updated 2026-02-19

Fixing Bolt.new & v0 Next.js Stripe Webhook Desync, App Router Leaks & Hydration Crashes

Bolt.new and v0 Next.js applications regularly drop Stripe subscription upgrades because AI-generated webhook routes parse JSON bodies before verifying `stripe-signature` headers, omit event-ID idempotency checks, and rely on fragile client-side success-page redirects. A production-safe billing pipeline requires raw-body signature verification, atomic database transactions, and server-side session middleware.

Bolt.new & v0 Next.js App Router Stripe Webhooks Supabase & Postgres Server Actions

Raw .md Read Diagnostic Teardown

STACK TEARDOWN Includes Copyable CLAUDE.md

Updated 2026-02-20

Rescuing a Claude Code & Cursor Codebase: Extracting a 3,000-Line God-Component into a Vibe-Safe Handover

When AI coding assistants (Claude Code, Cursor, Lovable, Replit) pass ~80% completion, they frequently collapse state, database mutations, and UI into 3,000-line monolithic files that break on every subsequent prompt. The fix is a two-step surgical extraction: decouple auth/billing into read-only service boundaries and enforce a root CLAUDE.md / AGENTS.md architecture contract.

Claude Code Cursor IDE CLAUDE.md & AGENTS.md React & Next.js TypeScript Refactoring

Raw .md Read Diagnostic Teardown

BUYER & COST GUIDE Includes Copyable AGENTS.md

Updated 2026-02-22

Vibe-Code App Rescue Sprint ($1,000–$1,500) vs. $50k Agency Rewrite: The Founder's Decision Guide

When an AI-built or inherited web application stalls at 80% completion, traditional dev agencies almost always pitch a $30,000–$80,000 from-scratch rewrite taking 3 to 6 months. In over 90% of cases, a surgical 1–2 week Fixed-Scope Rescue Sprint ($1,000–$1,500+) preserves your working 80% UI and schema while fixing the broken 20% at a fraction of the cost.

Buyer Decision Guide Rescue vs. Rewrite Software Economics Vibe-Coded Apps SaaS Architecture

Raw .md Read Diagnostic Teardown

DIAGNOSTIC CHECKLIST Includes Copyable .cursorrules

Updated 2026-02-24

The AI Prompt Death Spiral: 15-Point Production Readiness Checklist Before You Launch

The 'AI Prompt Death Spiral' occurs at roughly 80% app completion when every new prompt to Lovable, Bolt, Claude Code, or Cursor fixes one bug while breaking two existing features. Escaping this loop requires pausing blind prompt iteration and running a 15-point production readiness diagnostic across Data Security, Authentication, Billing Idempotency, Secrets, and Module Boundaries.

Production Checklist Lovable & Bolt Claude Code & Cursor Supabase RLS & Auth Stripe Billing

Raw .md Read Diagnostic Teardown

Direct Answers for Founders & Operators

Frequently Asked Questions

Can I copy the CLAUDE.md, AGENTS.md, and .cursorrules guardrails from these teardowns into my own repository? +

Yes. Every stack teardown includes an obsidian terminal block with a one-click copy button so you or your engineering team can immediately commit our Frozen Core Boundaries, RLS rules, and file-size guardrails to your repository root.

What if my codebase already has the bugs described in these teardowns? +

Adding CLAUDE.md stops future AI drift, but it will not retroactively fix exposed Supabase tables, non-idempotent Stripe webhooks, or a 3,000-line monolith. Book a Free 20-Minute Triage Call or send your app async—we sign a mutual NDA and deliver a Free 48-Hour Initial Codebase Audit ($0) with a fixed-price Rescue Sprint plan.

Are all teardowns available as clean Markdown mirrors for Claude Code, Cursor, or LLM ingestion? +

Yes. Every route under /teardowns exposes a deterministic 1-to-1 Markdown twin (for example, /teardowns/claude-code-cursor-spaghetti-to-vibe-safe-handover.md) so you can paste the URL directly into Claude Code, Cursor, ChatGPT, or Perplexity.

Ready to Find Out What's Actually Broken—For Free?

Book a 20-minute Triage Call or send your broken app asynchronously. We talk first, sign a mutual NDA, and if we're a fit, I deliver your 48-hour Codebase Audit and Loom walkthrough at zero cost.

Book Free 20-Min Triage Call → Send Broken App Async (1-Day Reply)

✓ Talk to Dositej directly (no sales reps) ✓ Keep repo access private until we talk & sign an NDA ✓ Free 48h Initial Codebase Audit ($0, zero obligation) ✓ Rescues from $1,000