Email Agent Skill

Overview

The Email Agent is responsible for:

1. Processing unprocessed emails from a workspace
2. Extracting sender information and linking to existing contacts
3. Analyzing email content for intents and sentiment
4. Updating CRM contacts with interaction data
5. Creating audit logs for all actions

How to Use This Agent

Trigger

User says: "Process emails for Duncan's workspace" or "Analyze unprocessed emails"

What the Agent Does

1. Fetch Unprocessed Emails

Call: convex query emails.getUnprocessed({
  orgId: "k57akqzf14r07d9q3pbf9kebvn7v7929",
  workspaceId: "kh72b1cng9h88691sx4x7krt2h7v7dehh",
  limit: 50
})

Returns array of emails not yet processed (isProcessed: false)

2. For Each Email

Step A: Extract Sender Email

From: "john@techstartup.com"
Extract: sender_email = "john@techstartup.com"

Step B: Link to Contact

Call: convex query contacts.getByEmail({
  orgId: "k57akqzf14r07d9q3pbf9kebvn7v7929",
  workspaceId: "kh72b1cng9h88691sx4x7krt2h7v7dehh",
  email: "john@techstartup.com"
})

If exists → contactId = found_contact._id If NOT exists → Create new contact with:

• email: sender_email
• name: extracted from email or "Unknown"
• source: "email"
• status: "lead"

Step C: Analyze Email Content

Extract these intent keywords:

• "interested" / "partnership" / "collaboration" → intent: inquiry
• "proposal" / "quote" / "pricing" → intent: proposal
• "issue" / "problem" / "help" → intent: complaint
• "?" / "how" / "what" / "when" → intent: question
• "follow up" / "re:" → intent: followup
• "meeting" / "call" / "sync" / "schedule" → intent: meeting

Multiple intents can apply to one email.

Step D: Analyze Sentiment

Read email tone:

• Positive indicators: "excited", "love", "great", "thank you", "appreciate"
• Negative indicators: "problem", "issue", "concerned", "unhappy", "urgent"
• Neutral: Standard business tone

Classify as: positive, neutral, or negative

Step E: Generate Summary

Create 1-2 sentence summary of email intent:

Example: "John from TechStartup is inquiring about Q4 marketing services and partnership opportunities."

Step F: Mark as Processed

Call: convex mutation emails.markProcessed({ orgId: "k57akqzf14r07d9q3pbf9kebvn7v7929", emailId: "email_id_from_step_1", contactId: "contact_id_from_step_b", intents: ["inquiry", "partnership"], sentiment: "positive", summary: "John inquiring about Q4 partnership" })

Step G: Update Contact

If this is a NEW interaction, update:

Call: convex mutation contacts.updateAiScore({
  orgId: "k57akqzf14r07d9q3pbf9kebvn7v7929",
  contactId: "contact_id",
  score: 75  // Increase score based on engagement
})

Call: convex mutation contacts.addNote({
  orgId: "k57akqzf14r07d9q3pbf9kebvn7v7929",
  contactId: "contact_id",
  note: "Email from John: Inquiring about Q4 partnership. Sentiment: positive. Intents: inquiry, partnership"
})

Step H: Log Audit Event

Call: convex mutation system.logAudit({ orgId: "k57akqzf14r07d9q3pbf9kebvn7v7929", action: "email_processed", resource: "email", resourceId: "email_id", agent: "email-agent", details: JSON.stringify({ from: "john@techstartup.com", intents: ["inquiry", "partnership"], sentiment: "positive", contactLinked: true }), status: "success" })

Error Handling

If something fails:

Call: convex mutation system.logAudit({
  orgId: "k57akqzf14r07d9q3pbf9kebvn7v7929",
  action: "email_processing_error",
  resource: "email",
  resourceId: "email_id",
  agent: "email-agent",
  details: JSON.stringify({ error: "error message" }),
  status: "error",
  errorMessage: "description"
})

Then continue to next email (don't stop).

Summary Output

After processing all emails, provide:

✅ Email Processing Complete

Total processed: X
Successfully linked: X
New contacts created: X
Intents extracted: X
Average sentiment: X

Contacts engaged:
- John Smith (TechStartup) - positive, inquiry
- Lisa Johnson (eCommerce) - positive, proposal
- Carlos Rodriguez (Agency) - positive, collaboration

Next steps:
1. Review high-priority contacts (positive sentiment + inquiry)
2. Generate followup emails for warm leads
3. Schedule meetings with decision-makers

Key Points

• Org isolation: All operations scoped to orgId
• Workspace scope: Process only emails from target workspace
• Contact linking: Always try to link email to existing contact
• AI scoring: Increase contact score when they engage (email received)
• Audit trail: Log every action for compliance

Example: Processing One Email

Input Email:

From: john@techstartup.com
Subject: Interested in your services
Body: Hi Duncan, we're looking to revamp our marketing strategy for Q4. Would love to chat about partnership opportunities.

Agent Process:

1. ✅ Extract sender: john@techstartup.com
2. ✅ Query contact: Found "John Smith" in contacts
3. ✅ Extract intents: ["inquiry", "partnership"]
4. ✅ Analyze sentiment: "positive" (enthusiastic tone)
5. ✅ Generate summary: "John inquiring about Q4 marketing strategy and partnership"
6. ✅ Mark email processed with contact link
7. ✅ Increase contact AI score from 68 → 78
8. ✅ Add note with timestamp and details
9. ✅ Log audit event with full context

Result:

• Contact updated with fresh interaction data
• Audit trail shows agent processed email
• Contact now appears in "high-value prospects" due to increased score