Automate Your Web Tasks with Chrome AI Skills

Google has launched Skills in Chrome to let users save and reuse their favorite AI prompts as one-click tools. This feature enables the automation of repetitive tasks like product comparisons and data extraction through a simple command interface. It is currently rolling out to desktop users in the U.S. with integrated security protections for sensitive actions.

Key Points

• Skills in Chrome allows users to transform repetitive AI prompts into one-click, reusable workflows.
• Users can access their saved Skills by typing a forward slash or clicking the plus button within the Gemini interface in Chrome.
• A new Skills library provides pre-made tools for common tasks such as shopping comparisons, recipe adjustments, and document scanning.
• The feature includes built-in security safeguards that require explicit user permission before the AI performs actions like sending emails.
• The rollout is initially available for English-US users on desktop platforms including Mac, Windows, and ChromeOS.

Sentiment

The overall sentiment is cautious-to-negative. While a vocal minority sees practical utility and potential in Chrome Skills, the dominant voices express concern about security implications, distrust of Google's intentions, and skepticism about the feature's maturity. The Manifest V3 hypocrisy argument resonates strongly, and the security-focused commenters are particularly emphatic that prompt injection risks make the feature dangerous.

In Agreement

• Skills solve a real pain point for users who repeatedly copy-paste the same prompts for tasks like generating alt text, adding travel bookings, or filling forms
• Having browser automation built into a mainstream browser could help break down data silos on a web that has become increasingly hostile to programmatic access
• The feature could be a good accessibility win, giving users a simple way to interact with web content through AI
• WebMCP as a complementary standard could make browser AI agents far more reliable by providing structured data interfaces instead of raw HTML parsing

Opposed

• Prompt injection remains an unsolved problem, and giving an AI agent access to page content turns every comment section into a potential attack vector
• Google restricted extension capabilities via Manifest V3 citing security, then introduced a more powerful AI agent with broad page access, which is hypocritical
• AI features that extract and summarize web content undermine content creators who depend on page visits and ad revenue for monetization
• Google has a track record of deprecating or changing pricing on products, making it risky to invest in their ecosystem
• The feature is premature and risky to release to the general public given the current state of LLM security