Claude Code

Lasso can hand the coding task to the Claude Code CLI while keeping the same selection → context → diff → accept workflow. Claude Code runs in read-only plan mode, so it proposes changes and never writes to your project.

Install the CLI

npm install -g @anthropic-ai/claude-code
claude --version

Lasso detects the binary with which claude. If Claude Code is installed after lasso dev has already started, restart the dev server so detection runs again.

Sign in

claude
# approve the login prompt, then exit

Run claude once in a terminal and complete sign-in. Alternatively export ANTHROPIC_API_KEY in your shell or project .env before starting lasso dev.

Select it in Lasso

Open the model picker in the overlay toolbar. Claude Code appears as a group only when the binary is detected, and offers Sonnet, Opus, and Haiku.

How a request runs

claude -p "<lasso context>" \
  --output-format stream-json \
  --verbose \
  --permission-mode plan \
  --max-turns 3 \
  --model sonnet

Lasso supplies the selected component context and expects a JSON patch proposal back, which becomes a reviewable diff. Tool activity streams into the task log while the agent runs. A request that has not finished within five minutes is stopped.

Permission requests

If the agent asks for approval or a value, Lasso pauses the task and shows the request in the overlay with Allow and Deny. Your answer is written to the standard input of the agent and the task continues.

Troubleshooting

• Not listed in the picker: the binary is not on the PATH that lasso dev sees. Check with which claude, then restart lasso dev.
• Sign-in errors: finish authentication in a real terminal first, then retry.
• Permission denied writing state: see the shared note on the OpenCode page; the same fix applies to any agent with a root-owned state folder.

Next steps

Compare with Codex or OpenCode, or review the CLI reference.