A volunteer maintaining a community events site has a new computer and wants Claude Code to read the project, make a small copy change and run the existing checks. Verify the setup in the terminal and project the volunteer will actually use. Keep a record of the executable, connection method and observed result. Use a practice directory without real registration data for the first task.
Establish which installation the terminal will run
Record the operating system, shell, project path and intended installation channel. A native Windows terminal and a WSL shell have separate executable paths and environments. If the command already runs, inspect its location and version before installing another copy. In PowerShell, Get-Command claude identifies the resolved command. On macOS or Linux, command -v claude shows the executable selected by the current shell. Save that path so an update can be checked against the same installation.
The official repository currently lists WinGet installation on Windows with winget install Anthropic.ClaudeCode and Homebrew installation with brew install --cask claude-code on macOS. Choose the current platform instructions from the official setup page when using a different channel. The repository marks npm installation as deprecated, so older advice that treats Node.js as a universal prerequisite needs to be checked against the selected method. After installation, reopen the terminal and run claude --version. Compare the resolved path with the channel you selected.
If the command is missing, check whether installation finished, whether the terminal has loaded its new path and whether the binary is in Windows or WSL. If several installations exist, record their paths and versions, then make the intended terminal resolve the intended copy. Follow the documented uninstall process for any confirmed obsolete installation after you know which one works. Repeatedly adding installers can make the next update or removal act on a different copy from the one you are testing.
Enter a small practice project containing a page, a README and an existing check command. Launch claude from that directory. Ask it to identify the entry point and explain one file. Record the working directory and result. This distinguishes an executable that starts from an environment that can read the intended project.
Choose the connection before configuring credentials
Ask the account owner which service the project will use. An official account, an organization-managed deployment and another provider's compatible endpoint have their own access and capability requirements. Follow the current instructions for the chosen service. For a compatible endpoint, record its exact host, model identifier and authentication type. Test a small text request and then a project task that reads a file and uses a permitted tool. Compatibility for one request does not establish every development workflow.
Inventory the configuration scopes that can affect the session. Current-process environment variables, user settings and project settings may contribute values. Record variable names, file locations and whether a fresh terminal was opened after a change. Let the credential holder inspect the secret values. Diagnostic notes can state whether a value exists and which scope supplies it without including the key. If an old provider configuration is still present, establish where the current request is going before changing settings for the intended connection.
For a network failure, identify the stage. Downloading an installer, completing a browser login and sending a model request involve different destinations. Record the hostname, time, response category and redacted error. Give that information to the account or network administrator. Use the organization's supported proxy and certificate configuration, and preserve the starting state before making an approved change. A single status message such as connection failed gives the maintainer little basis to distinguish a transport problem from missing account access.
After configuring the service, ask for an explanation of a file in the practice project. Then try a bounded task that requires reading the page and running the documented check. If text answers work but the local check fails, inspect the project runtime, working directory and execution permissions. Save the actual output. The setup record should make clear which connection worked and which local dependency still needs attention.
Complete the authentication flow for the active session
Start the selected authentication method using the current service documentation. For a browser flow, check the destination domain, account and named authorization target. Have the account holder enter credentials and complete any additional identity verification. Return to the waiting terminal after the browser finishes, then run a minimal request. Keep tokens, verification codes and sensitive callback content out of screenshots and shared troubleshooting notes.
If the browser reports success while the terminal remains unauthenticated, check that the completed flow belongs to that terminal session. An old browser tab, a separate operating-system user or another waiting terminal may account for the mismatch. Record the environment and time, close confirmed obsolete tabs and resume the current flow according to the documentation. If the callback cannot reach the local session, have the administrator inspect that path. Preserve the exact error and account method before removing configuration or reinstalling.
If authentication succeeds but the service rejects a request, classify the response. Account permissions, organization policy, service availability and budget can affect access. Send the redacted category and stage to the owner who can resolve it. A setup skill can organize checks and explain prerequisites, while the account holder performs identity verification and makes any subscription or payment decision.
Validate one reviewable change in a practice project
Create a recoverable starting state for the page. Ask the assistant to read the README and propose a change to the registration button text, naming the file and the check it expects to run. Review the plan against the task, then inspect the resulting diff and actual check output. If a required runtime is missing, record the command and dependency and follow the project instructions to supply it. Repeat the same check after the environment is ready.
Give the assistant a task with deliberately missing information, such as editing a form that is absent from the practice directory. Confirm that it identifies the missing input and offers a concrete next step. Start the same project in a fresh terminal and verify the executable path, working directory and authentication again. Keep these observations in a short handoff so the next volunteer can distinguish a machine setup problem from a project failure.
For an installation or diagnostic skill, inspect supported platforms, installation channels and side effects. It may add an editor, change a shell profile, install runtimes or switch a provider. Scope its work to the current gap and ask for the intended changes and verification before executing them. Test its guidance on the practice project and retain the package revision with the results. Add components when the project's actual workflow requires them.
Questions that occur after setup
Why did an update leave the old version running?
Compare the current executable path with the one saved before the update. Another installation channel may have a separate copy earlier in the shell's path. Confirm the intended channel, reopen the terminal and verify again before following the official cleanup steps for an obsolete copy.
Why does an API key appear to be ignored?
Check whether the active terminal sees the expected variable and whether its name matches the chosen provider's instructions. Confirm the configuration scope and selected service. The credential holder can verify the secret value while the troubleshooting record keeps only the variable name and redacted response.
Is an all-in-one setup script a good fit?
Compare the planned environment changes with the task. A machine that needs one missing executable has a different requirement from a fresh development workstation. Review the components, paths and connection changes the script proposes, and verify the resulting command and behavior after each relevant step.