What problem does it solve?
The Claude Code cloud sandbox is an ephemeral Linux container that differs from the local environment assumed by AGENTS.md and CLAUDE.md: plugins are not installed, moon and gh may be missing, GitHub's API is unreachable from the shell, and all HTTPS traffic must pass through a local proxy. This Skill provides the detection checks, workarounds, and fallback procedures needed to keep building, testing, and committing in that environment.
Core Features & Use Cases
- Environment Detection: Identify the sandbox via CLAUDE_CODE_REMOTE and CCR_* variables, and determine whether the setup script ran by checking for proto, moon, and node_modules.
- Tooling Workarounds: Bootstrap plugins with .claude/scripts/bootstrap-plugins.sh, fall back to pnpm dlx oxfmt when the formatter is missing, and typecheck against sources with a throwaway tsconfig when moon's WASM toolchains fail to load.
- Network and Browser Configuration: Route Chromium and Playwright through the local proxy with --ssl-version-max=tls1.2, and read GitHub run status through mcp__github__* tools instead of the blocked REST API.
- Use Case: A remote session's e2e tests fail with ERR_CONNECTION_RESET and pnpm exec moon dies with a registry load error. Use this Skill to confirm the sandbox, apply the Chromium proxy flags, and validate changes with a source-level tsc check instead of the full build.
Quick Start
Ask the agent to check whether this session is running in the Claude Code cloud sandbox and configure the environment so builds and tests can run.