Visual Flow — CoddexaBUILD

◈ CoddexaBUILD — Site Factory V3 Pipeline

Entry Point

Business Intake Submitted

Boss completes the form and clicks BUILD. A job is created instantly with a unique ID. Pipeline starts in background.

Input Fields

business_name * subdomain * niche / industry address / city language reference_site (URL) phone description features[ ] checkboxes

Job Created

job_id (hex 8 chars) status: running intake.json saved to disk

Phase A Reasoning Agent — Business Intelligence phase_id: reasoning

Step 1 — Reference Fetch

HTTP Download

Firecrawl / raw HTTP

Downloads reference site HTML. Timeout 10s. Extracts first 3000 chars of text content for style context.

Output

reference_text (3000 chars)

Step 2 — Business Analysis

DeepSeek V3 — Structured Inference

deepseek-chat

Analyzes business context against reference site style. Infers voice, location metadata, niche type, and complete build directives. Returns structured JSON.

Output → intake_structured.json

voice: name, tagline, tone location: city, country, postcode niche_meta: type, schema_org audit_priorities: platforms[ ] build_directives: structure, features, style, color_mood

Gate 1 Boss Reviews Reasoning Output status: waiting_approval · timeout: 30 min

Human Approval Checkpoint

Review: Business Voice & Build Directives

Boss reviews the inferred business profile. Can edit critical fields (tagline, tone, color mood, site structure) and trigger re-inference. Pipeline is paused until approval.

Actions

POST /approve-reasoning → pipeline continues POST /edit-reasoning → re-runs DeepSeek with corrections POST /cancel → job cancelled

Phase B Deep Research — Business Online Presence phase_id: research

Research Agent

Discovery — Contacts, Socials, Reviews

deep_research.py

Discovers the business's online footprint: Google Business listing, social media profiles, existing website URL, sample reviews. Builds the scrape target list for Phase C.

Output → research.json

business_name, address, phone, hours social_media: instagram, facebook website: existing URL photos_found[ ] (URLs) reviews_sample[ ] scrape_priorities[ ]

Phase C Scraping Agent — Data Collection phase_id: scraping · parallel + sequential

⟳ Parallel — 4 independent data sources

Source 1

Google Business

Apify · compass/google-maps-reviews-scraper

50 reviews. Place info: rating, hours, address. Timeout 90s.

Output

google_business.json google_reviews.json

Source 2

Instagram

Apify · apify/instagram-scraper

30 posts. Profile info, captions, image URLs. Timeout 120s.

Output

instagram.json

Source 3

Facebook Page

Apify · apify/facebook-pages-scraper

Page info, about text, cover photo. Timeout 30s.

Output

facebook.json

Source 4

Existing Website

Firecrawl API

Full markdown content + screenshot. Timeout 60s.

Output

existing_site.json

Sequential — after parallel sources complete

Photo Download

HTTP download

Downloads all discovered photo URLs. Max 30 photos. 30s timeout per photo. Skips files under 10KB. Saves to scraped/photos/.

Output

scraped/photos/ (image files) photos_manifest.json (metadata) scrape_summary.json

Phase D Compiler Agent — Content Curation & Copywriting phase_id: compile · model: Claude Opus 4.7

Compiler

Claude Opus 4.7 — Photo + Review + Copy

claude-opus-4-7 → claude-sonnet-4-6 (fallback)

Reads all scraped data. Scores photos by source weight and size. Uses LLM to select the 12 best reviews (min 4★, 80+ chars, max 24 months old, mix of types). Generates premium copy for every section of the site.

Photo Scoring

size ≥ 10KB filter google_business weight ×3 instagram weight ×2 existing_site weight ×1 top 12 selected

Output → compiled.json

selected_photos[12] + slot_suggested selected_reviews[12] + featured flag hero_tagline, hero_subtitle about_story, CTA, meta_description schema_org (Restaurant/Hotel/Store…) build_directives finalized

Gate 2 Boss Reviews Compiled Content status: waiting_approval · timeout: 2h

Human Approval Checkpoint

Review: Photos, Reviews, Copy, Build Directives

Boss reviews the compiled content on the approve.html page. Can edit compiled.json directly (swap photos, change copy, adjust directives). Approval unlocks the 60-minute build phase.

Actions

POST /approve-compile → build starts PUT /compiled → edit content, re-review POST /cancel → job cancelled

Phase E Builder Agent — Astro Site Generation phase_id: build · timeout: 60 min

Builder

Claude Code Subprocess — Astro Build

claude-code (subprocess invocation)

Invokes Claude Code as a subprocess with a 200+ line build prompt. Claude Code has full filesystem access to the site directory and executes pnpm build at the end.

4 sequential steps

Prepare Site Directory

Copy coddexa-base Astro template → site/. Remove dist/, node_modules/, .git. Write intake.json. Copy photos/ → site/public/images/.

pnpm install --frozen-lockfile

Install Astro dependencies. Timeout 2 min. Must succeed before Claude Code starts.

Claude Code: Design tokens → Content injection → Astro build

Extracts design tokens from reference site (colors, fonts, radius). Injects compiled.json into Astro components. Integrates photos into gallery. Builds niche-specific sections. Runs pnpm run build → dist/. Timeout 60 min.

Preview Deploy

Copy dist/ → /var/www/coddexa-build-preview/{job_id}/. Telegram: "Build complete, preview ready at /sfv3/jobs/{id}".

Output

site/dist/ (Astro static build) logs/{job_id}-build.log

Phase F Deployer Agent — DNS · Nginx · SSL · Lighthouse phase_id: deploy · 7 steps

7 sequential deployment steps

Porkbun DNS — A-record

POST to Porkbun API: create A-record {slug}.build.coddexa.media → 178.104.210.200. Skips if record already correct. Timeout 30s.

DNS Propagation Wait

nslookup poll loop. Max 60s. Continues even if timeout (certbot will retry).

Nginx HTTP Config

Write /etc/nginx/sites-available/{fqdn} (HTTP only, for certbot challenge). sudo nginx -t + reload.

Copy dist/ → Web Root

rsync site/dist/ → /var/www/coddexa-sites/{slug}.build/. Site is now LIVE on HTTP.

Certbot SSL (Let's Encrypt)

sudo certbot --nginx -d {fqdn}. Auto-updates nginx config to HTTPS redirect + SSL termination. Timeout 120s. Site is now LIVE on HTTPS.

Verification (4 checks)

① curl -I {url} → 200 OK? ② nslookup DNS → resolves to 178.104.210.200? ③ SSL cert file exists? ④ npx lighthouse --output json → perf / a11y / best-practices / SEO scores.

Telegram Notification

✅ Live URL + Lighthouse scores. Or ⚠️ with issues list. complete event pushed to SSE stream.

Output

deployment.json lighthouse.json live_url stored in job dict

Output

Site Live — https://{slug}.build.coddexa.media

Static Astro site live with SSL, custom subdomain on build.coddexa.media. Full content from scraped sources. Lighthouse verified. Telegram notification sent.

Final State

status: complete live_url: https://{slug}.build.coddexa.media ssl: true (Let's Encrypt) lighthouse.perf lighthouse.a11y lighthouse.seo completed_at timestamp all job files on disk → jobs/{id}/

Agents & Tools Reference

Phase	Agent	LLM / Tool	API / Binary	Timeout	Output File
A	Reasoning	DeepSeek V3 (deepseek-chat)	api.deepseek.com + HTTP	30s LLM · 10s ref-fetch	intake_structured.json
B	Deep Research	Web research	search / scrape APIs	variable	research.json
C	Scraper	Apify (3 actors) + Firecrawl + HTTP	api.apify.com · api.firecrawl.dev	30–120s / source	scraped/ directory
D	Compiler	Claude Opus 4.7 → Sonnet 4.6	api.anthropic.com	120s LLM	compiled.json
E	Builder	Claude Code (subprocess)	claude binary · pnpm · astro	3600s total	site/dist/
F	Deployer	Porkbun API + certbot + nginx + Lighthouse	porkbun.com · Let's Encrypt · npx lighthouse	120s certbot · 90s lighthouse	deployment.json

Approval Gates

Gate	After Phase	Timeout	Boss Can	On Timeout
Gate 1	Phase A — Reasoning	30 min	Approve · Edit fields + re-run DeepSeek · Cancel	Job cancelled with error event
Gate 2	Phase D — Compile	2 hours	Approve · Edit compiled.json · Cancel	Job cancelled with error event

Job Directory Layout

Disk · /opt/coddexa-build/jobs/{job_id}/

intake.json intake_structured.json ← Phase A research.json ← Phase B scraped/ ← Phase C scraped/google_business.json scraped/google_reviews.json scraped/instagram.json scraped/existing_site.json scraped/photos/ + photos_manifest.json compiled.json ← Phase D site/ (Astro project) ← Phase E site/intake.json site/public/images/ site/dist/ (built output) deployment.json ← Phase F lighthouse.json ← Phase F

DeepSeek V3

Claude / Anthropic

Apify

Firecrawl

Infrastructure (DNS · Nginx · SSL)

Approval gate / human checkpoint