Install an agentic coding CLI from the publisher's current documentation, authenticate through its supported flow, and run the first task in a small Git repository with a clean working tree. The quickest safe setup is the one you can verify and undo.
Use this sequence:
- Confirm the supported operating system, runtime, and shell in the official documentation.
- Install the documented package or signed binary; avoid copied one-line installers from third-party posts.
- Authenticate without pasting long-lived secrets into the prompt, repository, or shell history.
- Open a disposable project or a new branch with no unrelated changes.
- Ask the CLI to inspect the repository and propose a read-only plan.
- Approve one narrow edit, review the diff, and run the project's own tests.
The exact command varies by product and can change, so this page does not present one generic command as universal. A sandbox run reveals whether the CLI can find files, respect ignore rules, request permissions, run commands, and stop when blocked. If it supports isolated branches or worktrees, use them for experimental changes.
After the first successful run, record the project's build, test, formatting, and permission rules in the repository's agent-instruction file. That small investment makes later runs repeatable instead of turning every session into a new trust exercise.
Related reading: How to install an AI coding agent and AI agents in VS Code and JetBrains.
