agentop for VS Code

Watch Claude Code multi-agent workflows without leaving the editor.

The Runs panel updating as a workflow runs: eight agents across three phases, agents reporting back one by one, the attention badge changing, with one agent's transcript open as an ordinary editor tab

Install

Click Install above. The extension also needs the agentop binary on the machine your editor's workspace lives on:

brew install pimlabs/tap/agentop
npm install -g @pimlabs/agentop
curl -fsSL https://agentop.pimlabs.id/install.sh | sh

If it lands somewhere off PATH, point agentop.binaryPath at it. Run agentop: Report Diagnostics from the Command Palette when in doubt; it prints what the extension can and cannot see.

Nothing to try it on yet? Turn on agentop.demo and the panel fills with a sample workflow. It is generated rather than read, so nothing from your machine is in it, and it is what the recording and the screenshot here show.

What you get

A status bar item for the newest run: how many agents reported, how many are running, and how many need your eyes.
A Runs panel in the Activity Bar, three levels deep: run, phase, agent. Its badge counts the agents that need attention.
Each agent's transcript as an ordinary read-only editor tab, so search, split view and copy all work the way they already do.

One agent's transcript: its state, model, duration and token counts, then its attention, its prompt and its timeline of tool calls

A notification when a run stops moving with agents still running.
Mission Control, one page with the totals across every run.

Everything is about the folder you have open. The panel, the badge, the status bar and the notification move together, and a worktree counts as part of the repository it was cut from. Open in Terminal: Every Project drops the scope when you want the whole machine.

What it reads, and what that means

It reads the journal and the transcripts Claude Code already writes to disk, through the agentop binary. That is the whole of it:

No API key, no account, no sign-in, no service anyone runs.
Nothing is sent anywhere. The extension makes no network request at all.
It never writes to a Claude Code directory, and it can neither start, stop nor alter an agent.
Running it beside a live workflow is safe. There is no race against Claude Code, because nothing here writes what Claude Code reads.

Commands

Every command is under the agentop category in the Command Palette.

Command	What it does
Mission Control	Open the run and agent overview page
Open in Terminal	Start the TUI in a new terminal tab, scoped to this workspace
Open in Terminal: Every Project	The same, with no scope, so every project on the machine is listed
Open Transcript	Open the selected agent's transcript as a document
Open Transcript to the Side	The same, in a split editor
Copy Agent ID	Copy the selected agent's id to the clipboard
Refresh	Re-read now, without waiting for the interval
Report Diagnostics	Print what the extension sees, and why it might not

The transcript and copy commands act on the selection in the Runs panel, so they are hidden from the Command Palette and appear in that panel's context menu instead.

Settings

Setting	Default	What it does
`agentop.binaryPath`	`""`	Full path to the agentop binary. Empty means: use one bundled with the extension, then whatever is on PATH.
`agentop.home`	`""`	Directory to search for `.claude` and `.claude-work`. Empty means the home directory of the machine the extension host runs on.
`agentop.refreshInterval`	`5`	Seconds between disk checks.
`agentop.notifications`	`true`	Show a message when a run stops moving with agents still running. A run already stuck when the window opened is never announced on its own; instead, opening a workspace whose own runs went quiet in the last day says so once, as a count.
`agentop.demo`	`false`	Show a built-in sample workflow instead of reading the disk.

The status colours are themeable too, under agentop.busy, agentop.ok, agentop.fail and agentop.dim, with their own high-contrast values.

Remote workspaces, and WSL

The extension runs where your workspace does, never in the local UI. Under Remote SSH, a dev container, or WSL, Claude Code writes its journal on the remote machine, and reading the local disk instead would honestly report nothing at all.

That leaves one case it cannot solve by itself: an editor on Windows driving a workspace in WSL, where the workflow was started on the Windows side. Point agentop.home at the Windows home directory, which WSL mounts:

/mnt/c/Users/<you>

Report Diagnostics names the remote kind, the platform, the home directory it resolved, which route found the binary, and the schema version the binary answered with. Read it before filing anything.

Elsewhere

The same runs, read the same way, in a terminal and in a browser: pimlabs/agentop.

For contributors

The source lives in a private repository; the public one carries the README, the installer and the releases. Two things to know before changing anything here.

src/contract.ts is a generated, untracked copy of the TypeScript view of the wire format. Edit clients/contract/src/contract.ts instead; an edit to the copy is destroyed by the next build with no warning.

Extension releases are tagged vscode-v* and move independently of the binary's v* tags. The two do not have to be on the same number, and the extension refuses to speak to a binary whose schema it does not know.

agentop - Claude Code workflow monitor

pimlabs