extract_todo_screenpipe.md

Extract Todos from Screenpipe Data

Query your local Screenpipe API for the last 48 hours (all modalities):

# All content (OCR + audio + UI events)
curl -s "http://localhost:3030/search?limit=500&start_time=$(date -v-48H -u +%Y-%m-%dT%H:%M:%SZ)"

# Or query specific modalities:
# OCR (screen text): &content_type=ocr
# Audio transcripts: &content_type=audio
# UI events: &content_type=ui

Then ask the AI to analyze with this prompt:

---

Prompt:

Analyze this screenpipe data and extract all todos, reminders, and commitments. Look for:

Explicit signals:

• - [ ] unchecked markdown
• "TODO:", "FIXME:", "need to", "have to", "should", "must"
• "don't forget", "remember to", "by tomorrow", "deadline"

Commitment language (especially in audio transcripts):

• "I'll", "I will", "I promised", "I told [person] I would"
• "Follow up with", "Reply to", "Send [person]"
• Verbal commitments in meetings/calls

Meeting action items:

• Any bullet points from meeting notes
• Names + context suggesting follow-up needed
• Things said out loud in transcripts

Output format:

Extracted Reminders - [DATE]

🔴 High Priority (Time-Sensitive)

• [Task] — Source: [app/audio], [timestamp]

🟡 Medium Priority (Commitments)

• [Task] — Source: [app/audio]

👥 People to Follow Up With

• [Person]: [context]

📧 Pending Replies

• [Subject/thread]

---

Be exhaustive. Better to surface a false positive than miss a real todo.