claude-code

Use print mode with `claude -p "your task"` which runs a one-shot task and exits without interactive prompts. Add `--max-turns` to limit agentic loops and `--allowedTools` to restrict which tools Claude can use.

Pass `--output-format json` in print mode to receive a result object containing the response text, session_id, token usage, and cost. You can also supply `--json-schema` to force the output to match a specific structure.

Send Enter with tmux send-keys to accept the default workspace trust dialog. For the bypass-permissions warning, send Down then Enter because the safe default is 'No, exit' and you must select 'Yes, I accept'.

Yes, use `claude -c` to continue the most recent conversation in the current directory, or `claude -r <id>` to resume a specific session. In print mode, capture the session_id from JSON output and pass it with `--resume`.

Print mode (`-p`) runs a single task and exits, skipping all interactive dialogs, making it ideal for automation and CI. Interactive mode provides a full REPL with slash commands and multi-turn conversation but requires tmux orchestration to script.

Set `--max-turns` to cap agentic loop iterations and `--max-budget-usd` to cap dollar spend per run, both available in print mode. You can also use `--fallback-model haiku` to switch to a cheaper model when the default is overloaded.