A new Concise output style is now available in Claude Code. Claude leads with the result, keeps responses short by default, and still provides full detail on request. Enable it via /config, Output style, or by setting "outputStyle": "Concise" in settings.json.
The new Concise output style is a quick band-aid, not the final solution. A longer-term fix for verbosity is still in progress.
The /config command only modifies the current project's settings, not the global config. Setting outputStyle in ~/.claude/settings.json makes it persist across all projects. lydiahallie acknowledged the UI should make this distinction clearer.
Claude was forgetting custom output styles after a few turns. A fix has shipped that reminds Claude about custom output styles more frequently during a session.
Three updates shipped to Claude Managed Agents. The first: work done inside a Self-Hosted Sandbox can now be saved to memory, making it easier to persist context across agent runs.
The Console session viewer now features a minimap with one lane per agent, a streaming transcript grouped by agent iteration, and a cost inspector showing dollar amounts per thread and per session. The update is live now.
The web_search and web_fetch tools in Claude Managed Agents now accept allowed_domains or blocked_domains parameters, giving developers fine-grained control over what an agent can search and read on the web.
A new cookbook pairs Claude Managed Agents with AG-UI, CopilotKit's open Agent-User Interaction Protocol, letting managed agents power a wider range of UI surfaces.
The AG-UI adapter for Claude Managed Agents maps each chat thread to a managed session and streams text, tool calls, and model thinking directly to your UI. A live demo is available to try.
Users not seeing the new Concise output option need to run 'claude update' to get the latest version.
The Concise output style can't be set directly in the desktop app's settings, but setting Transcript View to Summary gives a similar result.
A complaint about Claude writing far more code than a human would was redirected: lydiahallie asked whether the user is on high or medium effort, suggesting the verbosity issue is related to effort level rather than output style.
A user on v2.1.224 couldn't find a recently discussed option; the answer is that it requires v2.1.237.
