Property Email Agent

Prerequisites

Required Skills

Install and configure these skills first:

1. google-workspace - Gmail API access

   • Follow setup in /skills/google-workspace/SKILL.md
   • Must have OAuth2 credentials configured
   • Required scopes: gmail.readonly, gmail.send, gmail.modify

2. ragie - Property knowledge base RAG

   • Follow setup in /skills/ragie/SKILL.md
   • Must have RAGIE_API_KEY in environment
   • Must have property documents indexed in Ragie

3. supabase - Conversation history database

   • Follow setup in /skills/supabase/SKILL.md
   • Must have SUPABASE_URL and SUPABASE_SERVICE_KEY in environment
   • Must have database schema created (see supabase skill docs)

LLM API Key

For email classification and draft generation:

# Option 1: OpenAI
export OPENAI_API_KEY="sk-..."

# Option 2: Anthropic (Claude)
export ANTHROPIC_API_KEY="sk-ant-..."

OpenClaw Gateway

This workflow runs as a cron job, so the Gateway must be running 24/7:

# Start Gateway (if not already running)
openclaw gateway run

Setup

Step 1: Test Individual Skills

Before orchestrating, verify each skill works:

# Test Gmail access
python3 ../google-workspace/scripts/gmail.py list --max-results 5

# Test Ragie search
python3 ../ragie/scripts/search.py \
  --query "property cooling measures Singapore" \
  --max-results 3

# Test Supabase connection
python3 ../supabase/scripts/test_connection.py

# Test contextgathering
python3 ../supabase/scripts/get_client_context.py \
  --email [email protected]

Step 2: Configure Prospect Filter

Edit config/prospect_domains.txt to list email domains/addresses to monitor:

# Prospect email patterns (one per line)
*@gmail.com
*@hotmail.com
[email protected]
[email protected]
# Add your client/prospect domains here

Or use the default: monitor all emails (filter by subject/content instead).

Step 3: Deploy the Cron Job

# Add hourly cron job
openclaw cron add \
  --name "Property Email Agent" \
  --cron "0 * * * *" \
  --tz "Asia/Singapore" \
  --session isolated \
  --message "Check Gmail for property questions and draft replies" \
  --announce \
  --channel slack \
  --to "channel:C1234567890"

# Or webhook delivery:
openclaw cron add \
  --name "Property Email Agent" \
  --cron "0 * * * *" \
  --tz "Asia/Singapore" \
  --session isolated \
  --message "Check Gmail for property questions and draft replies" \
  --webhook \
  --to "https://your-webhook.example.com/drafts"

Customization:

• Change --cron "0 * * * *" to adjust frequency (hourly by default)
• Use --announce to post drafts to chat (Slack/WhatsApp/Telegram)
• Use --webhook to POST drafts to your backend
• Set --tz to your local timezone

Step 4: Monitor the Workflow

# List cron jobs
openclaw cron list

# See job runs
openclaw cron runs --id <job-id> --limit 10

# Manually trigger (for testing)
openclaw cron run <job-id> --force

Usage Examples

Manual Workflow Run (Testing)

Before deploying the cron job, test the workflow manually:

# Process a specific email
python3 scripts/process_email.py \
  --email-id "msg_abc123xyz" \
  --debug

# Dry run (no database writes, no email sends)
python3 scripts/process_email.py \
  --email-id "msg_abc123xyz" \
  --dry-run

Process Unread Emails

# Check last hour of unread emails
python3 scripts/check_gmail.py \
  --since "1h" \
  --unread-only

# Check specific time range
python3 scripts/check_gmail.py \
  --since "2026-03-03T10:00:00Z" \
  --until "2026-03-03T11:00:00Z"

Classify Email as Property Question

# Test classifier on an email
python3 scripts/classify_email.py \
  --subject "Property inquiry" \
  --body "Is it a good time to buy property in District 9?" \
  --debug

Generate Email Draft

# Full workflow: context gathering + draft generation
python3 scripts/draft_reply.py \
  --client-email "[email protected]" \
  --original-subject "Property inquiry" \
  --original-body "Is it a good time to buy in District 9?" \
  --query "District 9 Singapore property investment timing" \
  --debug

Script Reference

check_gmail.py - Gmail Monitoring

Lists unread emails within a time range.

Parameters:

• --since (default: "1h"): Time range start (duration or ISO timestamp)
• --until (optional): Time range end (ISO timestamp)
• --unread-only (flag): Only fetch unread messages
• --debug: Enable debug output

Output (JSON):

{
  "emails": [
    {
      "id": "msg_abc123",
      "thread_id": "thread_xyz789",
      "from": "[email protected]",
      "subject": "Property inquiry",
      "snippet": "Is it a good time to buy...",
      "date": "2026-03-03T10:30:00Z"
    }
  ],
  "count": 1
}

classify_email.py - Property Question Detection

Uses LLM to classify if email contains property-related questions.

Parameters:

• --subject (required): Email subject
• --body (required): Email body text
• --from-email (optional): Sender email (for context)
• --debug: Enable debug output

Output (JSON):

{
  "is_property_question": true,
  "confidence": 0.95,
  "detected_topics": ["buying timing", "district analysis"],
  "reasoning": "Email asks about property investment timing in a specific district"
}

draft_reply.py - Email Draft Generation

Orchestrates context gathering + draft generation.

Parameters:

• --client-email (required): Prospect/client email address
• --original-subject (required): Original email subject
• --original-body (required): Original email body
• --query (required): Search query for Ragie (extracted from email)
• --thread-id (optional): Gmail thread ID (for threading)
• --dry-run: Don't save draft or write to database
• --debug: Enable debug output

Output (JSON):

{
  "draft": {
    "subject": "Re: Property inquiry",
    "body": "Dear John,\n\nThank you for your inquiry...",
    "word_count": 387
  },
  "context_used": {
    "ragie_chunks": 3,
    "supabase_conversations": 2,
    "supabase_memory": true
  },
  "saved_to": "gmail_draft"
}

process_email.py - End-to-End Workflow

Full workflow: classify → gather context → draft reply → save.

Parameters:

• --email-id (required): Gmail message ID
• --dry-run: Don't save draft or write to database
• --debug: Enable debug output

Email Draft Format

Generated email drafts follow this structure:

Subject: Re: [Original Subject]

Dear [Client Name],

Thank you for your inquiry about [topic]. Here's what you need to know:

## [Main Heading 1]

[2-3 paragraphs with data-driven insights from Ragie knowledge base]

Key points:
• [Bullet point 1]
• [Bullet point 2]
• [Bullet point 3]

## [Main Heading 2]

[Additional context based on client history from Supabase]

## Next Steps

[Clear call-to-action or offer for follow-up meeting]

Best regards,
[Your Name]
[Your Title]

---
📚 Sources:
• [Document 1 name] (via Ragie)
• [Meeting notes from [date]] (internal)

Constraints:

• 300-450 words (enforced by LLM prompt)
• 2-3 main headings (H2 level)
• Professional tone, data-driven
• Always include sources/attribution

Cron Job Configuration

Recommended Setup

# Production: Hourly monitoring with Slack delivery
openclaw cron add \
  --name "Property Email Agent" \
  --cron "0 * * * *" \
  --tz "Asia/Singapore" \
  --session isolated \
  --message "Run property email workflow (check Gmail, draft replies)" \
  --announce \
  --channel slack \
  --to "channel:C_PROPERTY_DRAFTS"

Alternative: Business Hours Only

# Run only during business hours (9 AM - 6 PM, Mon-Fri)
openclaw cron add \
  --name "Property Email Agent (Business Hours)" \
  --cron "0 9-18 * * 1-5" \
  --tz "Asia/Singapore" \
  --session isolated \
  --message "Run property email workflow" \
  --announce \
  --channel WhatsApp \
  --to "+6591234567"

High Frequency (Every 15 Minutes)

# For high-volume agents: check every 15 minutes
openclaw cron add \
  --name "Property Email Agent (15min)" \
  --cron "*/15 * * * *" \
  --tz "Asia/Singapore" \
  --session isolated \
  --message "Run property email workflow" \
  --announce

Customization

Adjust Email Classification

Edit scripts/classify_email.py to customize the LLM prompt:

# Add domain-specific keywords
PROPERTY_KEYWORDS = [
    "buy", "sell", "invest", "property", "condo", "HDB",
    "district", "cooling measures", "ABSD", "resale", "BTO",
    "rental yield", "capital appreciation", "mortgage",
]

# Adjust confidence threshold
CLASSIFICATION_THRESHOLD = 0.7  # Default: 0.7 (70% confidence)

Customize Draft Template

Edit scripts/draft_reply.py to adjust the LLM prompt for email generation:

DRAFT_PROMPT_TEMPLATE = """
You are a professional Singapore property agent...
[Customize tone, structure, constraints here]
"""

Add Custom Delivery Channels

The cron job can deliver to multiple channels:

# Multiple announcements (Slack + WhatsApp)
# (Not directly supported - use webhook + custom fanout)

# Webhook to your backend for custom processing
openclaw cron add \
  --name "Property Email Agent" \
  --cron "0 * * * *" \
  --session isolated \
  --message "Run workflow" \
  --webhook \
  --to "https://your-backend.com/api/email-drafts"

Error Handling

Common Issues

"Gmail API quota exceeded"

• Gmail API: 1 billion quota units/day (free tier)
• Reading emails: 5 units each
• Sending emails: 100 units each
• Solution: Reduce cron frequency or upgrade to Google Workspace

"Ragie API rate limit"

• Free tier: 1000 requests/month
• Solution: Cache frequent queries or upgrade Ragie plan

"Supabase connection timeout"

• Free tier pauses after 1 week inactivity
• Solution: Wake project in Supabase dashboard or upgrade

"LLM classification errors"

• Increase temperature for less strict matching
• Adjust confidence threshold in classify_email.py

Debug Mode

# Enable full debug logging
python3 scripts/check_gmail.py --debug
python3 scripts/classify_email.py --debug
python3 scripts/draft_reply.py --debug

Dry Run Mode

Test workflow without side effects:

# No writes to database, no email sends, no drafts saved
python3 scripts/process_email.py \
  --email-id "msg_abc123" \
  --dry-run

Security & Privacy

Data Handling

• Gmail: Only accesses your Gmail (OAuth scoped access)
• Ragie: Searches your property documents (stored on Ragie servers)
• Supabase: Stores conversation history in your database
• LLM providers: Email content sent to OpenAI/Anthropic for classification & drafting

⚠️ GDPR/PDPA Compliance:

• Client emails may contain personal data (PII)
• Ensure your LLM provider terms allow processing customer data
• Consider data retention policies in Supabase
• Inform clients about automated email processing (transparency)

OAuth Token Security

• Gmail tokens stored at ~/.openclaw/credentials/ (chmod 600)
• Never commit tokens to git
• Revoke tokens via Google Account settings if compromised

API Key Security

All API keys must be in environment variables, never hardcoded:

# ✅ GOOD
export RAGIE_API_KEY="..."
export SUPABASE_SERVICE_KEY="..."

# ❌ BAD (never do this)
# API_KEY = "sk-12345"  # in source code

See SECURITY.md for comprehensive threat model and incident response procedures.

Cost Estimation

Monthly cost for moderate usage (100 emails/day processed):

Cost optimization:

• Use GPT-3.5-turbo for classification (10x cheaper)
• Cache Ragie results for frequent queries
• Batch process emails (reduce API calls)

Monitoring & Analytics

Cron Job Logs

# View last 20 runs
openclaw cron runs --id <job-id> --limit 20

# Filter by status
openclaw cron runs --id <job-id> --status success
openclaw cron runs --id <job-id> --status error

Database Analytics

# Count total conversations
python3 ../supabase/scripts/query.py \
  --table conversations \
  --select "count"

# Get daily email volume
python3 scripts/analytics/daily_volume.py --days 30

Success Metrics

Track these in Supabase or your analytics platform:

• Emails processed per hour
• Property questions detected (vs false positives)
• Drafts generated vs sent (conversion rate)
• Average response time (email received → draft ready)

Troubleshooting

Workflow Not Running

# Check Gateway is running
openclaw gateway status

# Verify cron job is enabled
openclaw cron list

# Check job's last run
openclaw cron runs --id <job-id> --limit 1

# Manually trigger
openclaw cron run <job-id> --force

No Emails Detected

# Test Gmail access
python3 ../google-workspace/scripts/gmail.py list

# Check time range
python3 scripts/check_gmail.py --since "24h" --debug

# Verify OAuth scopes
python3 ../google-workspace/scripts/google_auth.py check-status

Classification Not Working

# Test classifier directly
python3 scripts/classify_email.py \
  --subject "Test" \
  --body "Is it a good time to buy property?" \
  --debug

# Check LLM API key
env | grep -E "OPENAI|ANTHROPIC"

Draft Quality Issues

• Adjust LLM temperature (higher = more creative, lower = more focused)
• Improve Ragie document quality (better source data = better drafts)
• Add more client context in Supabase (preferences, past conversations)
• Customize the draft template prompt

Related Skills

• google-workspace: Gmail, Calendar, Docs integration
• ragie: RAG search for property knowledge base
• supabase: Conversation history and client memory
• scrapling: Web scraping for property listings (complementary)

Roadmap

Future enhancements (not yet implemented):

• Multi-language support (English, Mandarin, Malay)
• Calendar integration (auto-schedule viewings)
• WhatsApp Business API integration (move beyond email)
• A/B testing for draft templates
• Sentiment analysis for client priority scoring
• Auto-send for high-confidence drafts (with approval workflow)

References

• OpenClaw Cron Jobs
• Gmail API Quotas
• Ragie.ai Best Practices
• Supabase Row Level Security

Changelog

v1.0.0 (2026-03-03)

• Initial release
• Hourly Gmail monitoring via cron
• Property question classification
• Ragie + Supabase context gathering
• Email draft generation (300-450 words)
• Multi-channel delivery (Slack, WhatsApp, webhook)