Skip to main content
These examples assume Claude Code is connected to Yelu as described in the Claude Code setup guide. Run commands from the repository root so Claude Code receives the right project context.

Confirm the active gateway

macOS WSL
Inside the session, run /status. Confirm that the base URL is https://api.yelu.ai before sharing repository context.

Explore without editing

Plan mode lets Claude inspect the repository and propose work without immediately modifying files:
Terminal
Use this first when joining an unfamiliar codebase or scoping a migration.

Implement a focused change

Terminal
Claude Code asks for permission when a command exceeds the active permission policy. Review commands and diffs before approving them.

Review uncommitted changes

Pipe the diff into a non-interactive review. The command works in macOS, PowerShell, Git Bash, and WSL when git and claude are on PATH:
Terminal

Produce machine-readable output

Use print mode for scripts and CI:
Terminal
Treat model output as untrusted input even when the command returns JSON. Validate required fields before another tool consumes it.

Continue a session

Terminal
To select an older conversation instead, use:
Terminal

Add project instructions

Create a reviewed CLAUDE.md in the repository root when the team wants consistent behavior:
CLAUDE.md
Project instructions are code: review them, keep them specific, and update them when commands or architecture change.

Windows notes

  • In native PowerShell, launch Claude from the terminal where ANTHROPIC_AUTH_TOKEN is set.
  • In WSL 2, run Claude and Git inside WSL and keep the repository in the Linux filesystem.
  • If shell commands behave differently across systems, tell Claude which shell and runtime the project supports.
Never approve destructive commands, credential access, deployments, or production data changes solely because the model requested them. Verify scope, target, and rollback first.
Last modified on July 13, 2026