Claude Code 2.1.145 Update

Claude Code version 2.1.145 adds Managed Agents self-hosted sandbox functionality. It supports outbound worker polling, environment key management, SDK and CLI worker paths, webhook-based activation, orchestration, and monitoring in self-hosted environments, and reference documentation is provided covering differences from the cloud environment and the separation of security responsibilities.

A new Run skill has been added that lets you run and test apps in the project's actual runtime environment. It prioritizes using a project-specific custom run skill, and if one isn't present, it falls back on built-in patterns by project type, including CLI, server, browser app, Electron app, TUI, and library.

Added skill templates and examples:

• Run skill generator: provides guidance for writing project-specific custom run skills along with verified build/run steps
• Run skill template: a reusable template including prerequisites, build, run, test, and troubleshooting sections
• Browser web app pattern: starting the dev server, checking readiness, the chromium-cli driver, and capturing screenshots
• CLI tool pattern: installation, running representative commands, and verifying output/exit codes/stdin behavior
• Electron desktop GUI pattern: running in an xvfb environment, Playwright REPL, screenshot capture, and documenting automation caveats
• Library/SDK pattern: build/test steps and smoke tests at the public API boundary