nova-act_skill

Overview

This skill lets you write and execute Python scripts that use Amazon Nova Act to perform AI-powered browser automation for tasks like flight searches, data extraction, and form filling. It bundles a runner script for quick tasks and supports custom Python scripts for complex workflows with structured extraction using schemas.

How this skill works

The skill launches a real Chrome browser (headless optional) and drives it via Nova Act's act() and act_get() APIs to click, type, navigate, and extract data. It reads the NOVA_ACT_API_KEY from environment or a local config and writes trace files (screenshots, HTML, session recordings) to the working directory for verification and debugging.

When to use it

• Automating multi-step web interactions (searches, navigation, form filling).
• Structured data extraction from web pages into Pydantic models or Python types.
• Rapid prototyping of browser workflows that benefit from natural-language steps.
• Taking reproducible screenshots or session traces for debugging or audits.
• Headless or headed runs where you need programmatic control of a real browser.

Best practices

• Combine related browser actions into a single nova.act() call to preserve context.
• Provide explicit dates and precise selectors or descriptions to avoid ambiguity.
• Always use schemas with nova.act_get() for reliable, structured extraction.
• Review and remove trace files after runs if they contain PII or sensitive data.
• Take screenshots (nova.page.screenshot) to verify results and include MEDIA lines for attachments.

Example use cases

• Search and scrape the top 3 cheapest flights for specified departure/return dates into a Flight model.
• Fill and test signup or checkout forms up to the confirmation step without completing transactions.
• Extract top news headlines and metadata from a news site into a list of dictionaries.
• Run recurring checks on a dashboard and save screenshots and session traces for audit.
• Prototype a complex, multi-click workflow by combining steps into one natural-language prompt.