Claude's mid-August API GA wave: browser use, computer use, Files, Skills, and write-capable Workspace connectors
On August 18–19, 2026 Anthropic graduated browser use and computer use toolsets, the Files API, and Agent Skills off beta headers on the Claude API. The same window finished Cowork on web and mobile for paid plans and added Gmail/Drive write actions with approval defaults. Here is the builder sequence I use so GA toolsets do not become an unsupervised outbound email machine.
In this post (6 sections)
Introduction
Early August gave enterprises Cowork Chrome, Compliance API coverage, and Auto mode defaults. Mid-August finished the API surface those products sit on: browser and computer toolsets you can call without beta flags, Files and Skills as stable endpoints, and Workspace connectors that can send mail and move Drive files. That is enough power to automate real work. It is also enough power to spam a customer list if approvals are wrong.
Read this next to Claude August 2026 enterprise stack and Cowork mobile/web playbook.
API GA checklist (August 19)
Computer use and browser use toolsets
- Computer use: `computer_toolset_20260801`, batch actions, zoom on by default, per-member configs.
- Browser use: client-hosted viewport with accessibility-tree reads, element refs, forms, tabs, opt-in uploads.
- Available on Fable 5, Mythos 5, Opus 5, Sonnet 5, and Opus 4.8 on the Claude API.
- Upgrading from older computer beta versions changes request shape. Migrate with a golden path, not a global find-replace.
Files API and Agent Skills
Files and Skills no longer require `files-api-2025-04-14` or `skills-2025-10-02` headers. GA responses add file expiration and pagination. Requests that still send beta headers keep legacy shapes, so plan an explicit header-removal pass. If you package skills for multiple clients, keep them portable with Agent Plugins 1.0 and MCP GA day-one notes.
Primary source: Claude Platform release notes.
Product surfaces that landed the day before
- Cowork on web and mobile for Pro, Max, and Team (Enterprise admin-gated).
- Gmail send/reply/forward and Drive share/move/trash/upload with approval defaults.
- Console Playground (formerly Workbench) covering every Messages API parameter with SDK request/response views.
Migration sequence I actually run
- 01Inventory beta headersList every Messages and Files/Skills call that still sends computer, files, or skills beta headers.
- 02Shadow one browser or computer pathMove a single non-production agent to the 20260801 toolsets. Re-run sandbox egress probes from agent eval sandbox escape checklist.
- 03Header removal PRRemove beta headers only after response-shape tests pass. Keep a rollback branch that restores headers for 48 hours.
- 04Connector allowlistsEnable Gmail/Drive write in a pilot workspace. Log every send and trash for a week before org-wide allow.
Common mistakes
- Assuming GA means the old beta request body still works.
- Turning on Gmail auto-allow for all Team members on day one.
- Skipping egress proofs on browser use because "it is only a viewport."
- Leaving skills beta headers forever and wondering why deletes behave oddly.
- Treating Playground prototypes as production configs without effort and tool allowlists.
Conclusion
Mid-August closed the gap between Claude's enterprise apps and the API builders use. Browser and computer toolsets, Files, and Skills are now stable enough to standardize on. Workspace write actions make that power consequential. Ship the migration behind evals, approvals, and connector policy, not behind a changelog screenshot.
Sources: Claude Platform docs ; Claude blog ; Claude blog
Agentic AI patterns, delivered Thursdays
What I am shipping, watching, and pruning out of client stacks each week. One email. No fluff.