開發者平台
Claude’s Legacy Workbench Is Now Disabled, and Three Experimental Prompt APIs Have Started Returning Errors
Anthropic ended access to the legacy Workbench on August 17. Saved prompts, versions, variables, and evals cannot be carried over to the new interface. The replacement is a Messages API playground with no server-side history, while the experimental endpoints for generating, improving, and templatizing prompts have also been retired.
Anthropic’s August 17 deadline has passed: the legacy Workbench in Claude Console is no longer accessible, and the prompts, revisions, completions, variables, uploaded files, and evals previously stored there cannot be imported into the updated Workbench. This is more than an interface upgrade. Teams that relied on Console as a prompt registry, for manual regression testing, or as a shared record of experiments must now move version control and evaluation workflows to their own repositories or third-party tools.
The new Workbench has a narrower, stateless design that maps directly to the public Messages API, displays complete requests and responses, and can export code snippets. Current drafts remain in the browser; Anthropic no longer stores prompt or conversation history for users. It is therefore better suited as an API request builder than as an experiment management system. The official legacy export is provided as JSON and may include model completions and uploaded files, but the new Workbench has no import functionality. For content that was not exported before the deadline, Anthropic’s official wording is that it is “no longer accessible,” so users should not assume it can still be recovered through support channels.
Also retired on the same day were `/v1/experimental/generate_prompt`, `/v1/experimental/improve_prompt`, and `/v1/experimental/templatize_prompt`. After their removal, these endpoints return errors directly rather than transparently forwarding requests to a new service. CI pipelines, internal prompt editors, or tutorial examples that still call them should remove those dependencies or reimplement the functionality using the standard Messages API and custom templating logic. Engineering teams should next review API key traffic and error logs, place exported prompts and eval schemas under version control, and establish reproducible records of model versions, parameters, and test data. Although the new Workbench improves consistency between the API and the interface, Anthropic has not provided an official equivalent replacement for the legacy eval functionality.