Know When Your Coding Agent Needs You

You send a coding agent off to work, switch to another task, and eventually wonder: is it still running, did it finish, or is it waiting for me?

Nash puts that answer where you can glance at it: in the agent's tab title and a status badge inside its pane. For Claude Code and Antigravity CLI, connecting the agent takes one hook-install command and a restart.

A Small Signal That Saves a Lot of Checking

Instead of repeatedly opening each agent session to read its output, look for its status:

• Green spinner: the agent is working.
• Orange circle: it needs your input.
• Green check-circle: work finished and the result is unread.
• Red alert-circle: work failed and needs a look.

Waiting and finished are deliberately different from working. A permission question should not look like a job that is still making progress, and a completed turn should not look like a progress bar stuck at 100 percent.

The waiting signal depends on what the agent reports. Claude Code exposes blocking permission and MCP input events; Antigravity's ask_question tool reports a question, but its permission prompts do not expose a distinct event Nash can recognize.

Tabs Tell You Where; Panes Tell You Which Agent

A tab aggregates activity from its splits, including panes that are not selected or are hidden by a zoomed pane. You can work in one split without losing track of an agent running in another.

The pane badge makes that signal more specific: this is the worker that needs you. When a turn finishes or fails, the badge also shows when it ended.

An unread tab outcome is acknowledged when you return to the worker pane. The pane's outcome badge stays until you type in it, so simply looking at a result does not immediately erase its timestamp. If you were already watching a successful turn finish, Nash does not mark that tab's completion as unread.

Connect Your Agent

Install Nash 0.5.8 or newer, set up your agent separately, and open a shell inside Nash.

For Claude Code:

For Antigravity CLI (agy):

Run the command for the agent you use, then exit and restart that agent in a Nash pane. Hooks are loaded at agent startup. Submit a prompt and watch its tab and pane badge.

To check the registration:

There is no hook JSON to transcribe and no separate Python adapter to maintain. Nash installs hooks into your user-wide agent configuration, merges with the settings already there, and safely handles repeated installation. The hook command ships with Nash.

Prefer to Try It in One Project?

Preview the change first, then install from your project's root:

User-wide setup remains the simplest choice across projects. For Antigravity print-mode runs (agy -p), user-wide hooks are required.

The setup guide has the configuration paths, scope options, verification steps, troubleshooting, and uninstall commands. The FAQ provides the short version.

---

Your agent can keep working while you do something else. Nash makes it easier to know when to come back.

-Nash Team