Claude Audit

Audit once, prompt perfectly.
Idea Stage JBMD Creations LLC 2026-02-24

The Problem

Developers waste hours feeding AI context about their UI

The typical AI-assisted UI audit looks like this: take 10 screenshots, paste them one at a time, explain what each shows, AI hallucinates about components that don't exist, correct it, re-explain, re-screenshot. Repeat every session.

2-4h
Wasted per audit
50K+
Tokens on context
5-10
Manual screenshots
3-5x
Correction rounds

The Solution

One URL + repo = a complete, annotated visual brief

Claude Audit packages the capture-analyze-brief pipeline into a single product. Feed it a URL, connect a repo, get back a structured board with labeled screenshots, component inventories, dead code maps, and cross-reference tables. Everything an AI needs in one image.

URL + Repo ──> Capture Engine ──> Analysis Engine ──> Brief Composer ──> Board │ │ │ Multi-breakpoint AST scanning Dark-themed HTML Auth-aware Import graph Base64 images Real device merge Dead code detect Audit tables Component map Cross-references

What It Produces

Screenshot Boards

Multi-breakpoint captures across mobile, tablet, desktop. Auth-aware. Real device photos merged alongside automated captures.

Component Inventories

Every control labeled, cross-referenced against source code. Numbered, per-state visibility (blank/active/mobile/desktop).

Dead Code Maps

Unused components flagged with "superseded by" mappings. Migration paths. Safe-to-delete confidence scoring.

Rendering Audit

Which components render which data types. Stream pipeline mapping. Unimplemented/unwired component flags.

Cross-Reference Matrix

Control x State visibility grid. Desktop vs mobile control mapping. Chrome resize vs real device discrepancies.

AI-Ready Export

Self-contained HTML, PNG, or structured JSON. One board = complete context. Feed it to Claude and start working.

Core Features by Phase

Phase 1: Capture Engine (MVP)

  • 6 default breakpoints (375-1920px)
  • Auth-aware: JWT, cookies, OAuth
  • Full-page + viewport modes
  • CSS selector targeting
  • Real device photo upload + merge
  • JSON manifest with metadata

Phase 2: Analysis Engine

  • AST-based component scanning (ts-morph)
  • Import graph builder
  • Dead code detector with migration paths
  • Stream rendering audit map
  • Cross-reference matrix generator
  • Framework auto-detection

Phase 3: Brief Composer

  • Dark-themed responsive HTML boards
  • Base64-embedded images (self-contained)
  • Audit tables: controls, rendering, dead code
  • Letter-labeled captures (A, B, C...)
  • Export: HTML, PNG, JSON, Markdown

Phase 4-5: Web App + API

  • Dashboard with project list + history
  • One-click re-audit, historical diff
  • GitHub integration (AST from repo)
  • REST API for programmatic access
  • MCP server for Claude Code
  • GitHub Action for PR-triggered audits

Technical Stack

LayerTechnologyWhy
FrontendNext.js 16 + React 19 + Tailwind v4Existing expertise, SSR for boards
BackendCloudflare Workers (Hono)Edge-first, existing infra
CapturePuppeteer in CF Sandbox containersIsolated, scalable, auth-safe
AnalysisTypeScript AST (ts-morph)Full type-aware component scanning
StorageR2 (images) + D1 (metadata)CF-native, cost-effective
AuthBetter AuthExisting pattern, OAuth support
PaymentsStripeExisting JBMD Creations LLC setup

Pricing

Free

$0
  • 3 audits/month
  • Public repos only
  • 6 breakpoints
  • No auth capture

Team

$49/seat
  • Everything in Pro
  • Shared dashboard
  • CI/CD integration
  • PR-triggered audits
  • Slack notifications

Enterprise

Custom
  • Self-hosted option
  • SSO/SAML
  • Custom analysis rules
  • Dedicated support
  • SLA guarantees

Competitive Landscape

ProductWhat It DoesGap We Fill
StorybookComponent catalogNo cross-reference to live UI state
ChromaticVisual regressionScreenshots only, no analysis
PercyVisual testingCI-focused, no AI-optimized output
Figma Dev ModeDesign specsDesign to code, not code to audit
LighthousePerformance auditPerformance only, no component mapping
Claude AuditFull UI audit briefVisual + structural + AI-ready

Implementation Timeline

Sprint 1 — Week 1-2

Core Capture Engine

  • Project scaffolding (Next.js + CF Workers)
  • Puppeteer multi-breakpoint capture
  • Auth injection (JWT, cookies)
  • R2 storage + basic board composer
Sprint 2 — Week 3-4

Analysis Engine

  • ts-morph AST setup + React scanner
  • Import graph + dead code detection
  • Cross-reference matrix generator
  • Audit table HTML templates
Sprint 3 — Week 5-6

Web App MVP

  • Dashboard UI + Better Auth
  • Audit trigger flow + board viewer
  • Export (HTML, PNG, JSON)
Sprint 4 — Week 7-8

Polish + Open Beta

  • Real device photo upload
  • Stripe integration + free tier
  • Product Hunt launch
Sprint 5-6 — Week 9-12

Pro Tier + Integrations

  • Private repos, API, webhooks
  • GitHub Action, MCP server
  • Team tier, Slack notifications

Revenue Targets

$950
Month 3 MRR
$4.8K
Month 6 MRR
$10K+
Month 12 MRR
~80%
Gross Margin

Go-To-Market

Phase 1: Dogfood (Week 1-4)

Use on VaporForge, Clarity, WP Dispatch, DevTools. Refine pipeline from real usage. Document edge cases.

Phase 2: Open Beta (Week 5-8)

Product Hunt launch. Free tier only. Target Claude Code, Cursor, Windsurf communities. Reddit + Twitter/X.

Phase 3: Paid Launch (Week 9-12)

Pro tier via Stripe. Blog posts, video demos. Partner with AI coding tool newsletters.

Phase 4: Team + API (Month 4-6)

Team tier. CI/CD integration. API docs site. Agency partnerships.

Prior Art (Already Built)

AssetWhat It Proves
/brief skillCapture + compose + export pipeline works end-to-end
capture-authed.mjsJWT localStorage injection for auth-aware Puppeteer
VaporForge UI Audit planThe exact audit table format users need
VaporForge prompt-kit migrationDead code detection + migration mapping pattern
DevTools Hub plans boardHTML board pattern for plan visualization

The VaporForge UI audit session IS the MVP demo. Everything built tonight becomes the foundation.