On August 20, the official ClaudeDevs account announced on X (formerly Twitter) that Computer use, the browser tool, Skills API, and Files API are now fully available on the Claude Platform. The official blog published a corresponding post titled Build production agents with computer use, the Skills API, and the Files API. According to Anthropic, developers can now build Claude Managed Agents with versioned skills and reusable files.

Official teaser video:

Let's start with Computer use. This capability enables agents to operate software that does not have an available API: given a screenshot, it can click, type, and scroll just like a human sitting in front of a keyboard. The key update in this generally available release is batch actions. Claude can now return multiple actions in a single response, for example returning `left_click`, `type`, and `screenshot` together, and the application executes them in sequence without requiring a round trip for every single step. The new toolset is `computer_toolset_20260801`, and beta headers are no longer required. Based on official documentation, one developer estimates that this reduces the number of round trips by 20-40%.

The newly added browser use tool complements Computer use. Pure pixel-based positioning is prone to misclicks when page layouts change. The browser tool reads the page structure and operates based on element references plus coordinates, making targeting much more reliable. Official documentation notes that for tasks that only occur within a web page, the browser use tool is the more suitable option.

Skills API and Files API equip agents with structured capability and persistent memory. Skills API allows teams to upload their internal workflows as a versioned skill: you write it once, manage it with version control, and pin it to a `version_id` or `latest` when making requests. Skills use a folder structure that includes SKILL.md, scripts, and templates. Claude loads them on demand: metadata stays in memory, content is only read after an instruction trigger, scripts run via bash, and only their outputs are added to the context window. This means loading dozens of skills won't overflow the context window.

Files API solves the problem of repeated file uploads. You upload a file once to get a `file_id`, and reference it directly in subsequent requests. This update adds the `expires_in_seconds` parameter, raises the rate limit to 500 RPM, and provides 1TB of storage per organization. Agents can read PDFs, edit spreadsheets, and download results all within this system.

The official team shared a concrete use case: a claims agent reads intake documents from Files API, follows the team workflow defined in the skill, uses the browser use tool to fill out and submit the form on the insurance company's website, then saves the confirmation page back to the file system. This end-to-end workflow previously required a lot of glue code, but now all functionality is provided at the platform level. According to a research engineer quoted in the official blog, their longest claims workflow went from 32 minutes to 13 minutes, costs dropped by roughly 30%, and the completion rate hit 100%.

On the security front, the official guidance is straightforward: use isolated virtual machines or containers, never give login credentials to the model, restrict accessible domains, and prevent malicious instruction injection. All screenshots, mouse movements, and keyboard inputs are stored in your own environment—Anthropic does not retain this data. Computer use now also qualifies for HIPAA BAA coverage.

Developer reactions are quite interesting. Some said this API is increasingly looking like an actual employee; others joked: You didn't build a chatbot, you hired a ghost with admin permissions. That's a bit of an exaggeration, but it points to the same trend: with screen viewing, browser operation, custom skills, and file storage, the core building blocks of a practical agent are now all in place.

All of these features are now generally available on the Claude API. Microsoft Foundry already supports Skills API and Files API, and computer use and browser use are coming soon to Vertex AI. Existing beta integrations will continue to run to facilitate a smooth migration.

More details:

- [Computer use Documentation](https://platform.claude.com/docs/en/agents-and-tools/tool-use/computer-use-tool)

- [Agent Skills Documentation](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview)

- [Official Blog Post](https://claude.com/blog/computer-use-skills-api-files-api)