💡 New script: gmail-action-extractor (Gemini-powered actionable item extraction from emails to Google Tasks)

[github-actions[bot]]

Summary

A new script that uses Gemini to extract actionable items — deadlines, commitments, questions requiring responses, and follow-up requests — from labeled email threads, presenting them in a daily digest email with one-click Google Tasks creation links. Unlike classification or summarization, this bridges "reading" to "doing" by converting email content into structured task items.

Market Signal

AI email tools in 2026 are moving beyond classification and summarization toward action extraction. Tools like Sintra and Shortwave offer task extraction, but they're SaaS with monthly subscriptions and data leaving the user's account. Gmail's native Gemini summarizes threads but does not extract actionable items into Tasks. Multiple AI email organizers now treat smart categorization as table stakes — the competitive frontier has shifted to intelligent action extraction.

This project's privacy advantage (data stays entirely within the user's Google account, open-source auditable code) is especially relevant for action extraction, which processes sensitive commitments, deadlines, and business obligations.

User Signal

• The project already has proven Gemini integration in gmail-ai-classifier — PR feat(classifier): add tag-based retention and 2-way rules sync engine #530 merged a tag-based retention and 2-way rules sync engine, demonstrating mature AI email processing capabilities
• Discussion 💡 New script: gmail-to-calendar-events (Gemini-powered event extraction from emails to Calendar) #563 covers email-to-Calendar-events extraction; Discussion 💡 New script: meet-recap-to-tasks (Meeting Smart Notes to Google Tasks pipeline) #485 covers Meet-recap-to-Tasks — neither addresses email-to-Tasks action extraction
• Issue Feature Proposal: AI-Powered Gemini Semantic Email Classifier & Auto-Filter Generator #491 proposes a semantic email classifier, showing maintainer demand for AI-powered email intelligence
• getCleanBody() in src/gas-utils.js provides the text normalization pipeline needed for clean Gemini input

Technical Opportunity

gmail-ai-classifier already proves the Gemini-in-Apps-Script pattern works via UrlFetchApp.fetch() to the Gemini REST API. The new script reuses:

1. Gemini API integration pattern — API key via Script Properties, structured prompt/response handling
2. getCleanBody() — email text normalization (quoted-reply stripping, line break normalization)
3. Gmail label-based processing loop — the established pattern for processing labeled threads
4. Existing test infrastructure — GAS service injection, Jest test helpers in test-utils/

Google Tasks API is available as an Apps Script Advanced Service, making task creation straightforward. The script follows the established architecture: code.gs → src/index.js → tests/.

Assessment

Dimension	Score	Rationale
Feasibility	med	Requires Gemini API key and careful prompt engineering for reliable action extraction (more nuanced than classification). Pattern is proven in the codebase but action extraction precision needs validation.
Impact	high	Addresses a major pain point — important action items buried in email threads — with a privacy-first approach that SaaS competitors cannot match.
Urgency	med	AI email action tools are proliferating; competitive window exists but is not immediate. Gemini being free in Workspace plans lowers the adoption barrier.

Adversarial Review

Strongest objection: False positives (creating wrong tasks from misinterpreted email content) are worse than false negatives. Users will lose trust quickly if the system creates nonsensical tasks.
Rebuttal: Start with a "review before commit" model: the script extracts and presents actions in a daily digest email, each with a one-click link to create the Task — never auto-creates. This puts the user in control while removing the cognitive load of parsing emails for action items. The Gemini prompt can be tuned to prefer precision over recall (miss some actions rather than hallucinate). The gmail-ai-classifier's experience with Gemini prompt engineering directly transfers.

Suggested Next Step

Prototype the Gemini prompt for action extraction against a sample of real email threads. Measure precision/recall across action types (deadlines, commitments, questions, follow-ups). Design the daily digest email template with one-click Google Tasks creation links. Reuse gmail-ai-classifier's API integration pattern and test infrastructure.