Get it for macOS

Run the commands an agent suggests right under its reply, and have the agent read the result and answer on its own. Every bash, sh, zsh or console code block in a reply gets a Run button that opens a live terminal beneath it, where you can answer prompts, type passwords and press Ctrl-C. When the block finishes, its exit code and output reach the agent as hidden context, so it replies without you pasting anything and the chat gets no extra message. The shell starts in the thread's workspace with your startup files loaded and keeps cd, virtualenvs and variables for follow-up commands. Needs a macOS or Linux machine; no account or external service.


Overview

View source

Agents often answer with a command for you to run. This plugin turns that code block into a terminal you can use on the spot, and hands the outcome back to the agent the way !command does in Claude Code.

What you get

  • A Run button under every bash, sh, shell, zsh, console, shell-session and terminal code block in agent replies. It becomes active once the block has finished streaming.
  • A live, interactive terminal right under the block that grows with its output: answer prompts, type passwords, press Ctrl-C, run follow-ups in the same session. Type exit to end it, then Run again to start fresh.
  • When the block finishes, its exit code and output go to the agent as hidden context and the agent replies right away. You see the output in the terminal; nothing extra is posted to the chat. If the agent is still busy, the result waits until its current reply ends.
  • The terminal icon in the terminal's top-right corner opens the same session in bb's terminal panel.
  • To keep results to yourself, turn off Send command results to the agent in the plugin settings.

How it works

The shell starts in the thread's workspace on the thread's machine, with your own shell startup files loaded, and runs the block inside that interactive shell. cd, activated virtualenvs, aliases and variables set by the block stay in the session, and the block lands in your shell history. Ctrl-C reports exit code 130 and keeps the shell open; exit and set -e still report their status.

In console and shell-session transcripts, only lines that start with $ or % run; the output lines are skipped.

Only the block's own run is reported to the agent. Commands you type afterwards in the same terminal stay between you and the shell.

Requirements

  • A macOS or Linux machine for the thread. zsh and bash get full support; other shells fall back to /bin/sh.
  • No account, API key or external service.

Privacy

Nothing runs until you click Run. The output of a run you start, up to the last 20,000 characters, is sent to the thread's agent. Leave auto-send off for commands that print secrets.

More from Dmitrii Maksimov

More in Command Line 7

Work with shells and command-line programs inside bb.

View all