All shelves / Mods / Workflow
whats-agent-doing
By tzafrir
whats-agent-doing is a Claude Code workflow mod by tzafrir.
Draws a box above the prompt that names the step Claude is on, how long it has lasted, and each background agent.
Draws
Install
The marketplace name is tzafrir.
/plugin marketplace add tzafrir/whats-agent-doing /plugin install whats-agent-doing@tzafrir
It runs inside the Claude Code process. The side panel is only what the author's README names. Read the repo before you install, or run claude plugin validate on the plugin folder.
- Version
- 0.4.3
- License
- MIT
- Repo stars
- 7
- Last change
- 2026-10-04
What it can reach
- Draws in the session
- Sees tool calls
Needs Claude Code in a terminal, or in the desktop app's Code tab. Start a new session, or run /reload-plugins. The README says function hooks are early access, they load only where function hooks are enabled, and their API may change between releases. The box is not drawn in Claude chat or Cowork. A box above the prompt names the step Claude is on and how long it has been on it. The states it names are reading your prompt, thinking, writing a tool call, running a tool, a subagent at work, background agents, waiting for your approval or your answer, reviewing tool results, writing the reply, compacting, and idle between turns. A command is labeled with that command's own description, or with the command itself when it has none. Parallel calls show the newest, with a count of how many more. Click the triangle, or focus the band and press Enter, to expand the history: each prompt, stretch of thinking, tool call, reply, and turn ending, with durations. Marks distinguish done, failed, denied, and interrupted. Background agents get a row each and keep it after your turn ends, while they are still working. The box steps aside while a survey holds the band. Other mods that draw above the prompt keep their place under this box. The README says it only displays: no network, no files, no telemetry. The headline and the history stay in the session's own state and nothing else. The README says every hook is display only and changes nothing, and that settings, instructions, tool descriptions, other hooks' results, tool calls and their results, and the model's stream all reach Claude exactly as they would without the plugin. It shows a permission wait and never answers it. The box and its history display your prompts, file paths, URLs, command descriptions, MCP tool names, and, where the model streams it, a line of its thinking. The README says anyone who can see your screen can read them. Update later with claude plugin marketplace update tzafrir.
The marketplace name is tzafrir. The step box, with a row for each background agent, is why this page files it under workflow. agent-watch draws each subagent's tokens and a stall toast, and the last tool calls, with a summary of the heaviest agent when they finish. plain-view hides successful tool rows and draws the agent's plan as a card above the prompt, with the current step and a summary when the turn ends. This box names the step in plain English and keeps the history of the turn. It does not draw tokens, and it does not hide tool rows.
Demo

What it hooks into
session.startWhen a session startsturn.startWhen Claude starts a turnturn.stepAt each step of a turntool.callAround a tool call, where it can hold oneclassic.PermissionRequestA Claude Code eventui.renderDraws into the sessionToolProgressAbovePromptagent.spawnWhen a subagent startsclassic.SubagentStopA Claude Code eventclassic.PreCompactA Claude Code eventclassic.PostCompactA Claude Code eventturn.completeWhen a turn ends
Read from the on() calls in its hook modules. Open its source under Files to check what each one does.
Files
The author's README and source, 4 files. They load when you open them.
Facts
- Author
- tzafrir
- Category
- Workflow
- Marketplace
- tzafrir
- Version
- 0.4.3
- License
- MIT
- Repo stars
- 7
- Last change
- 2026-10-04
- Language
- TypeScript
- Files
- 18