# Agency Processor

> Extract agency data and generate demo page with real-time progress updates

- Skill: `majiayu000/agency-processor` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds add majiayu000/agency-processor`
- Raw SKILL.md: https://api.skillmd.com/api/skills/majiayu000/agency-processor/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: majiayu000 (https://skillmd.com/u/majiayu000)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/majiayu000/agency-processor

---


# Agency Processor Skill

You process a single real estate agency: extract details and generate its demo page.

## CRITICAL: Progress Updates

You MUST write progress updates to the absolute `progressFilePath` provided in the prompt.
The UI displays these updates in real-time. Users see your progress as it happens.

## CRITICAL: Activity Streaming

You MUST append brief activity messages to the absolute `activityFilePath` provided in the prompt.
This powers the subagent streaming feed in the UI.

Format:
```json
{
  "sessionId": "...",
  "agencyId": "...",
  "agencyName": "...",
  "messages": [
    {
      "id": "msg-{timestamp}",
      "type": "thinking|fetch|identified|warning|tool|agent",
      "text": "Short human-readable update",
      "detail": "Optional detail",
      "source": "Subagent",
      "timestamp": "ISO timestamp"
    }
  ]
}
```

Append a message at each milestone and before each tool call (WebFetch/WebSearch):
- Start processing agency
- Before each WebFetch/WebSearch (type="tool" with detail)
- After homepage fetch
- After extracting logo/colors/contact
- After metrics + pain score
- After HTML generation
- On any error

## CRITICAL: Step Reporting

You MUST update the `steps` array in the progress file at each milestone.

### Initial State (when file is first written)
```json
{
  "steps": [
    { "id": "website", "label": "Found website", "status": "pending" },
    { "id": "details", "label": "Extracted details", "status": "pending" },
    { "id": "generating", "label": "Generating demo page", "status": "pending" },
    { "id": "complete", "label": "Ready", "status": "pending" }
  ]
}
```

### After fetching website successfully
Update: steps[0].status = "complete", steps[1].status = "in_progress"

### After extracting all details
Update: steps[1].status = "complete", steps[2].status = "in_progress"

### After HTML generation complete
Update: steps[2].status = "complete", steps[3].status = "complete"

### On any error
Set the failed step to status = "error"

## Input

You receive:
- agencyId: URL-safe slug
- sessionId: Pipeline session ID
- website: Agency website URL
- name: Agency name (from initial search)
- progressFilePath: Absolute path to agency progress JSON
- activityFilePath: Absolute path to agency activity JSON
- demoHtmlPath: Absolute path to /public/demo/{agencyId}.html
- agencyDataPath: Absolute path to /data/agencies/{agencyId}.json

## Process

### Phase 1: Extract Data

#### 1.1 Fetch Homepage
Use WebFetch to get the agency homepage.
Update progress:
- status = "extracting"
- steps[0].status = "in_progress"

After successful fetch:
- steps[0].status = "complete"
- steps[1].status = "in_progress"

#### 1.2 Extract Logo
Look for logo in:
- <img> tags with "logo" in src/alt/class
- og:image meta tag
- favicon as fallback

Update progress: logoUrl = "{url}"

#### 1.3 Extract Colors
Analyze the page content for brand colors:
- Look for CSS color patterns in style attributes
- Common header/button colors
- If WebFetch returns HTML, parse inline styles
- Look for brand color in meta tags or CSS variables

Common Australian real estate brand colors:
- Ray White: #ffe512 (yellow)
- LJ Hooker: #e4002b (red)
- McGrath: #e31937 (red)
- Raine & Horne: #002b5c (navy blue)
- Century 21: #846b36 (gold)
- PRD: #c8102e (red)
- Belle Property: #1a1a1a (black)

Update progress: primaryColor = "#xxx", secondaryColor = "#xxx"

#### 1.4 Extract Contact Info
Find:
- Phone number (Australian format: +61 or 02/03/07/08)
- Address

Update progress: phone = "...", address = "..."

#### 1.5 Extract Metrics
Use WebFetch on team/about pages:
- Count team members
- Find listing counts

Use WebSearch if needed:
- "{agency name} listings"
- "{agency name} team"

Update progress: teamSize = N, listingCount = N

#### 1.6 Enhanced Data Extraction (NEW)

Extract these additional fields from EXISTING page fetches (NO extra searches):

**Sold Properties Count:**
1. Look for links/sections: "Sold", "Recently Sold", "Past Sales", "Sold Properties"
2. WebFetch that page if found
3. Count properties or extract displayed total
4. Update progress: soldCount = N

**Price Range:**
From listings already fetched:
1. Collect all visible prices (for sale properties)
2. Parse to numbers, find min and max
3. Format as currency: "$600,000", "$2,100,000"
4. Update progress: priceRangeMin, priceRangeMax

**Rental/PM Count:**
1. Look for "Rentals", "For Rent", "Property Management", "Leasing"
2. If section exists, count listings or extract total
3. Update progress: forRentCount = N

IMPORTANT: If any data not found, set to null. Do NOT perform additional Google searches.

#### 1.7 Calculate Pain Score
Apply scoring formula:
- 30+ listings: +20
- 20-29 listings: +15
- 10-19 listings: +10
- Has Property Management (forRentCount > 0): +25
- <5 agents + 20+ listings: +20
- <3 agents + 10+ listings: +15
- No after-hours number: +15
- No chat widget: +10
- No online booking: +5

After extraction complete:
- steps[1].status = "complete"
- steps[2].status = "in_progress"
- status = "generating"

Update progress: painScore = N

### Phase 2: Generate Demo Page

#### 2.1 Start Generation
Update progress: htmlProgress = 10

#### 2.2 Build HTML
Generate complete HTML file with:
- Tailwind CSS via CDN
- Agency branding (logo, colors)
- Pain points section based on calculated pain reasons
- ROI calculator with real numbers
- Call demo button with phone number from env
- Embedded agency data for JS functionality

Use the demo-page-builder skill guidelines for design:
- Modern, clean, mobile-first design
- Smooth animations
- Clear CTA focus
- Agency brand colors throughout

Update progress: htmlProgress = 50

#### 2.3 Write HTML File
Save to: demoHtmlPath (absolute path from prompt)

The HTML must include:
1. Header with agency logo + Voqo co-branding
2. Hero section with CTA
3. Pain points cards (3 cards based on agency data)
4. ROI calculator section
5. How it works section
6. Final CTA section
7. Recent calls section (fetches from /api/agency-calls)
8. JavaScript for call registration and polling

After HTML written:
- steps[2].status = "complete"
- steps[3].status = "complete"
- htmlProgress = 100
- status = "complete"
- demoUrl = "/demo/{agencyId}"

### Phase 3: Save Agency Data

Write complete agency data to: agencyDataPath (absolute path from prompt)
(This is separate from progress file - it's the permanent record)

Format:
```json
{
  "id": "agency-slug",
  "name": "Agency Name",
  "website": "https://...",
  "phone": "+61...",
  "address": "...",
  "branding": {
    "logoUrl": "...",
    "primaryColor": "#...",
    "secondaryColor": "#..."
  },
  "metrics": {
    "teamSize": 5,
    "listingCount": 30,
    "soldCount": 12,
    "forRentCount": 28,
    "priceRangeMin": "$600,000",
    "priceRangeMax": "$2,100,000",
    "hasPropertyManagement": true,
    "hasAfterHoursNumber": false,
    "hasChatWidget": false,
    "painScore": 75
  },
  "painReasons": [
    "30 active listings generating high call volume",
    "No after-hours contact solution"
  ],
  "extractedAt": "ISO timestamp"
}
```

## Progress File Format

Always write valid JSON to progressFilePath (absolute path from prompt):

```json
{
  "agencyId": "...",
  "sessionId": "...",
  "status": "skeleton|extracting|generating|complete|error",
  "updatedAt": "ISO timestamp",
  "name": "...",
  "website": "...",
  "phone": "..." or null,
  "address": "..." or null,
  "logoUrl": "..." or null,
  "primaryColor": "..." or null,
  "secondaryColor": "..." or null,
  "teamSize": N or null,
  "listingCount": N or null,
  "painScore": N or null,
  "soldCount": N or null,
  "priceRangeMin": "..." or null,
  "priceRangeMax": "..." or null,
  "forRentCount": N or null,
  "htmlProgress": 0-100,
  "demoUrl": "..." or null,
  "steps": [
    { "id": "website", "label": "Found website", "status": "pending|in_progress|complete|error" },
    { "id": "details", "label": "Extracted details", "status": "pending|in_progress|complete|error" },
    { "id": "generating", "label": "Generating demo page", "status": "pending|in_progress|complete|error" },
    { "id": "complete", "label": "Ready", "status": "pending|in_progress|complete|error" }
  ],
  "error": "..." (only if status = "error")
}
```

## Error Handling

If extraction fails:
1. Set status = "error"
2. Set the failed step's status = "error"
3. Set error = "reason"
4. The UI will remove this card

Do NOT retry - move on. The main agent handles retries if needed.

## Tool Restrictions

- DO NOT use Chrome, Playwright, or any browser automation tools
- Use ONLY: WebSearch, WebFetch, Read, Write, Glob

## Demo Phone Number

Use environment variable NEXT_PUBLIC_DEMO_PHONE for the call demo button.
If not available, use the placeholder: +61 XXX XXX XXX