One Claude Code Baseline for Three Operating Systems

01 · Share the Claude Code baseline, not a home directory

For a team running sustained coding workloads, configuration drift becomes an operations problem. Different model choices, permission rules and shell assumptions can turn the same task into three different experiences across macOS, Linux and Windows. The practical goal is not an identical machine image. It is a versioned baseline with explicit exceptions for credentials, paths and execution environments.

Treat the setup as three layers: repository policy, user configuration and machine-specific overrides. Share the first, document the second and keep the third local. That boundary makes onboarding repeatable without turning a configuration repository into a collection of secrets and personal preferences.

02 · Put portable instructions and permissions in version control

Use the project’s CLAUDE.md file for shared working instructions: build commands, test expectations, architecture boundaries and rules for handling generated files. Keep these instructions concrete. A command that only works in one developer’s shell is not a portable team standard.

Store appropriate project settings in .claude/settings.json. Review changes to this file like changes to build infrastructure, especially permission rules, hooks and external integrations. Use .claude/settings.local.json for local exceptions, and verify that version control excludes it.

• › Repository layer: Shared instructions, reviewed permissions and portable workflow conventions.
• › User layer: Personal preferences and configuration needed across multiple repositories.
• › Machine layer: Local executable locations, credentials and environment-specific adjustments.

Do not copy an entire user configuration directory between machines. It can contain authentication material, local state and assumptions that do not belong in a team baseline. Check the effective configuration after onboarding because project and user settings can interact.

03 · Choose a Windows execution boundary before rollout

On Windows, decide whether your supported environment is native Windows or Windows Subsystem for Linux. Do not leave that choice implicit. A WSL workflow follows Linux filesystem and shell conventions; a native workflow must account for Windows paths, PowerShell behavior and executable discovery.

Prefer repository-relative paths and scripts with documented runtime requirements. Avoid hard-coded home directories, drive letters and developer-specific binary locations. Standardize line endings for scripts, preserve executable permissions where needed and test filename capitalization on a case-sensitive filesystem. Hooks deserve special attention: a shell command that works on macOS may fail under native Windows.

Pin or record the Claude Code version used for validation. Roll updates through a small cross-platform pilot before making them the team default. Configuration compatibility is easier to manage when the client version is not another uncontrolled variable.

04 · Keep relay configuration portable and credentials private

A shared API relay can simplify funding and routing for heavy workloads. Quanta is a pay-as-you-go option with one balance across multiple model providers, and it works with Claude Code as well as other coding clients. Open a Quanta account, then validate the Claude Code integration on one machine before distributing the setup.

Document the approved relay base address without embedding credentials in repository settings. Claude Code supports endpoint configuration through ANTHROPIC_BASE_URL; use the authentication mechanism specified for the relay integration. Inject secrets through an approved local environment or secret-management process rather than committing them. Verify that diagnostic output does not expose credentials.

Separate the team’s model policy from provider-specific routing identifiers. For example, evaluate Claude Sonnet 5.5 for routine coding work, but confirm the exact supported identifier and endpoint compatibility before putting it into the shared baseline. A display name is not necessarily an API identifier.

05 · Validate token costs before expanding the rollout

Run the same representative repository task on all three operating systems. Check authentication, model selection, file access, test execution, hooks and permission prompts. Include one deliberately blocked operation to confirm that the intended guardrails actually apply.

Then compare usage, not just task completion. Quanta meters calls on real tokens, with cache writes and cache reads priced separately. Track those categories independently; a repeated workflow can have a different cost profile from its first run. Separate interactive developer use from unattended jobs where your credential and reporting controls allow it.

Assign an owner to the baseline, retain a rollback version and document the process for requesting exceptions. The result should be one maintained operating policy across three platforms—not three configurations that merely started from the same template.