Blog
2 min read

Claude API: Files, Agent Skills and Computer Use Leave Beta, and a New Browser Use Tool

On 19 August 2026 Anthropic took the Files API, Agent Skills and the computer use toolset out of beta, and launched a browser use tool that drives a browser your app hosts. What no longer needs a beta header, what changed in responses, and what the browser tool adds over screenshot-and-click.

On 19 August 2026, Anthropic moved several Claude API features out of beta and added a new one for agents that work in a browser.

Out of beta

Files API. Uploading files once and referencing them in later requests no longer needs the files-api-2025-04-14 header. Requests without the header get the current response format, which adds file expiry (expires_in_seconds on upload, expires_at on file objects) and page/next_page pagination with an ids[] filter when listing. Requests that still send the old header keep the old format, so migrate deliberately.

Agent Skills and the Skills API (/v1/skills). No more skills-2025-10-02 header, including when loading skills through the container parameter. (Claude Code Skills)

Computer use — as the computer_toolset_20260801 toolset: batch actions (several in one turn), zoom on by default, and per-member configuration. Moving from computer_20251124 changes the request shape, so follow the migration notes. Newer models require the toolset. (Claude Haiku 5.5 Released)

New: the browser use tool

browser_toolset_20260801 is a client toolset for driving a browser your application hosts. Rather than treating the browser as pixels on a desktop, it works within the viewport and reads the page itself — its accessibility tree, elements, forms and tabs — adding element references, form input, tab management, download reporting and opt-in file upload on top of screenshot-and-click.

That tends to be more reliable and cheaper than full-desktop computer use for web tasks. Declaring the toolset adds roughly 6,600 input tokens per request, so prompt caching matters. (Prompt Caching Explained)

Python and TypeScript SDK classes for both toolsets arrived in beta on 7 October.

Also on 19 August

  • Managed Agents can restrict web_search and web_fetch to allowed or blocked domains. (Agent Egress Control)
  • Managed Agents in self-hosted sandboxes can attach memory stores.

What it means for you

  • Using Files or Skills in production: remove the beta headers on your schedule, and handle file expiry.
  • Building agents that use websites: try the browser toolset before full computer use. Run the browser somewhere isolated — an agent that browses can meet prompt injection on any page. (Prompt Injection in Coding Agents, Playwright MCP)

Source: Claude Platform release notes

EasySpawn gives your agent features a server-side home where browsers, files and API keys stay off users' devices, with Claude Code to build them. See how it works or join the waitlist.

Related: Build an AI Agent · What Is Function Calling? · Claude Agent SDK · Run Untrusted AI-Generated Code Safely

Keep reading