Claude Platform 正式上线 Computer Use、Skills API 与 Files API，新增浏览器操作工具
Key Highlights
Anthropic announced today that three major capabilities are now generally available on the Claude Platform: Computer Use, the Skills API, and the Files API. The company also launched a brand-new browser automation tool alongside them. The significance of this release is considerable and should not be underestimated by teams that are building AI products right now. It means that agents are no longer confined to a chat window where they only produce text in response to prompts from a human operator. Instead, they can now directly operate software interfaces, invoke skills that a team has carefully built up over time, and deliver finished files that can be used right away without further manual assembly or copy-paste work. In short, the boundary between a language model and a working operator is starting to dissolve, and that shift has real consequences for how software gets built and run inside organizations of every size. The launch of this combination marks a key step for Claude as it moves from a chatty assistant into a platform that can actually get work done, and for enterprise buyers the appeal is that one vendor now covers the full loop from intention to delivered artifact.
What the capabilities do
Computer Use lets an agent read the screen, move the cursor, click buttons, and type text just like a person would, which allows it to drive traditional desktop software that was never given a programmatic interface in the first place. The Skills API turns a company's internal workflows, knowledge bases, and specialized tools into reusable "skills" that an agent can automatically discover and call the moment it receives a task from a user, without a human having to specify which one to use. The Files API lets an agent read and write files inside a conversation, returning both intermediate results and final deliverables as files that the user can open directly without leaving the platform, removing the need to copy and paste back and forth. The new browser tool fills a clear gap for web scenarios: an agent can open pages, fill out forms, and extract content to complete end-to-end online tasks without human supervision at every single step, so a single prompt can span a spreadsheet, a legacy app, and a web form without the user wiring those systems together.
Technical details
All three capabilities are exposed through the unified Claude Platform interface, which keeps the developer experience consistent across every feature. A developer only needs to declare the relevant tool permissions in the API request to enable them, with no separate SDK to install or separate service to host. Computer Use combines visual understanding with action planning: it first captures a screenshot to understand the current interface state, then decides the next step before it acts on the screen. The Skills API supports version management and permission control so that every skill call is traceable and auditable by the owning team. The Files API handles streaming reads and writes as well as format conversion behind the scenes. The browser tool reuses the action engine of Computer Use but optimizes specifically for the DOM structure, making it noticeably more stable than pure visual clicking alone, and because the same permission model governs all three features, security reviews stay simple and consistent.
Comparison with competitors
Compared with similar web agents such as OpenAI's Operator and Google's Project Mariner, Anthropic's differentiation is that it connects three things — operating software, calling skills, and handling files — inside a single platform rather than shipping them as isolated demos that do not talk to each other. Developers can obtain all three capabilities through one API, which lowers integration cost and reduces the number of vendors they must manage and monitor on a daily basis. Competitors tend to show one impressive trick at a time, whereas Anthropic is assembling a coherent toolkit that fits together from the start and is meant for production use, and the bet is that breadth beats a single headline feature once companies move from demos to daily work.
Industry impact and use cases
For scenarios that rely heavily on graphical interfaces, such as software automation, support ticket handling, data entry, and report generation, this suite means teams can finally retire fragile scripts and rigid RPA rules that break whenever the user interface changes by even a few pixels. Enterprises can also codify their own processes into Skills, turning agents into digital workers that can actually be put on the payroll, so to speak, and that can be governed like any other internal system. Simply put, Claude is evolving from a chat assistant into a platform that can roll up its sleeves and get things done, and that change may matter more to buyers than any single model benchmark published in a leaderboard, because the larger story is the slow merger of chatbot and robot.