What's Agent Doing

What it does

A small box above the prompt says what Claude is doing, in plain words, and how long it has been doing it: "Reading your prompt", "Reading greet.js", "Try the greeting with a name", "Writing the reply · 120 words". A shell command shows as its description, not as raw bash. The border is cyan while Claude works and turns yellow when it's waiting for your approval or your answer, so you can tell at a glance whether it's stuck, busy or waiting on you.

Background agents each get a row of their own with their task, their current step and a timer, and the rows stay after the turn ends while the agents keep working.

How to use it

Install it and start a session. The box is there from the start.

• Click the ▸ at the left of the box to open the history: each prompt, stretch of thinking, tool call (✓ done, ✗ failed, ⊘ denied, ■ interrupted), reply and turn ending, with how long each took.
• From the keyboard: press ctrl+x then Tab to focus the band, then Enter to open or close the history.
• ctrl+x ctrl+a collapses the whole band.
• Between turns the box shows how the last one ended, like "Idle · last turn done in 7.3s, 2 actions".

Good to know

• It only displays. It reads Claude Code's events and passes them on unchanged, and never answers a permission prompt for you.
• Everything it shows lives in the session's own state. It writes no files, runs nothing and keeps nothing between sessions.
• The box shows your prompts, file paths, command descriptions and a line of Claude's thinking where the model streams it, so anyone who can see your screen can read them.
• It draws the whole band above the prompt, so it can hide another mod that draws there. It steps aside while a survey uses the band.
• While Claude or an agent is working, it redraws once a second to keep the timers moving.