Anthropic has extended Claude Cowork, its multi-step task feature, into the Claude for Chrome extension, and paired it with support for skills and plugins inside the browser itself. According to The Decoder, the update turns the extension from a page-reading assistant into an agent that can carry reusable instructions and third-party integrations into whatever tab it happens to be working in.

The move closes a gap between Anthropic's two agent surfaces. Claude Code has run on a skills and plugin architecture for months, letting developers package prompts, scripts, and permissions into portable units that Claude loads on demand. The Chrome extension, by contrast, has mostly stayed a research-preview tool for clicking buttons and filling forms on a user's behalf. Bringing skills and plugins to it means the same reusable building blocks that power terminal-based agent work can now run against live web pages instead of a command line.

For teams that have already written Claude Skills — folders of instructions, examples, and helper scripts — for use in Claude Code, the extension update means that work no longer stops at the terminal. In our estimation, that reuse is the more consequential part of this release for builders, since it lowers the cost of maintaining separate automation logic for command-line agents and browser agents.

From a research-preview clicker to a pluggable agent

Claude for Chrome started as a way for Claude to observe and act inside a browser tab — reading a page, filling a form, clicking through a multi-step flow — kept behind a research-preview label and permission gates for good reason: an agent that can click anything on a page can also be tricked into clicking the wrong thing. Cowork pushed that base capability from single-shot actions toward longer, more collaborative task runs. Layering skills and plugins on top is the logical next step: instead of describing a workflow from scratch in every session, a user can hand the extension a packaged skill and expect it to execute the same steps the same way each time.

What skills and plugins change in the browser

The distinction between the two pieces matters for anyone building on top of Claude.

That combination is what separates a Claude that can read a page for you from a Claude that repeats an exact multi-step browser task the same way every time, using whatever tools a team has configured for it.

Practical implications for builders

For teams already automating parts of their stack with Claude, the update has a few concrete consequences worth planning around:

AiiN's takeaway

The headline detail here isn't that Claude can now click around inside Chrome — that's been true for a while. It's that Anthropic is standardizing on one skills-and-plugins model across its agent surfaces instead of building a separate extension mechanism for each one. For teams that already treat Claude Skills as shared infrastructure, that consistency is the part worth tracking, since it determines how much of an existing automation library survives the move from terminal to browser.