School Action Extractor

Uses Claude AI to intelligently extract parent actions from school emails, classify urgency, and identify deadlines.

Invocation

school:extract-actions --email-data <json>

Process

1. •Receive email data (body + attachment text)
2. •Build prompt with context (child, school, date)
3. •Call Claude Sonnet to extract actions
4. •Parse JSON response
5. •Classify urgency (HIGH/MEDIUM/LOW)
6. •Return structured action list

Action Structure

Each extracted action contains:

{
  "description": "What needs to be done",
  "child": "Elodie",
  "deadline": "2026-02-08",
  "urgency": "MEDIUM",
  "type": "PERMISSION",
  "source_text": "Quote from email"
}

Urgency Classification

• •HIGH: Deadline within 48 hours OR explicit urgent language
• •MEDIUM: Deadline 3-7 days away OR requires action (forms, payments)
• •LOW: Deadline > 7 days OR informational

Action Types

• •FORM - Form to fill out
• •PAYMENT - Payment required
• •PERMISSION - Permission slip
• •REPLY - Response needed
• •EVENT - Calendar event
• •INFO - Informational only
• •OTHER - Other action type

AI Prompt

The AI is given:

• •Email metadata (from, subject, date, child)
• •Email body text
• •Attachment text (from PDFs)
• •Today's date for deadline context
• •List of urgent keywords

It returns ONLY valid JSON array of actions.

Usage

This skill is typically called automatically by the email processor. For testing:

cd ~/clawd/projects/school-email-automation
python -m school_automation.action_extractor.extractor --test

Requirements

• •ANTHROPIC_API_KEY environment variable
• •anthropic library installed