agent-pulse

Repository removed. As of 19 October 2026, bradtraversy/claude-mods no longer exists on GitHub. The author’s new repository, bradtraversy/traversy-claude-mods, does not include agent-pulse, so the install commands on this page no longer work. We keep the page as a record of what the mod did and will update it if the author publishes it again.

agent-pulse gives every subagent in the session a card in an Agents pane, with its status, how long it has been running and the tool it is using or used last. The pane opens by itself the first time Claude starts a subagent, without taking focus, and stays out of the way in sessions that have none.

What it does

• Cards. The status (PENDING, RUNNING, WAITING, IDLE for a teammate free for more work, COMPLETED, FAILED, KILLED) inside a border of the matching colour, the agent’s name or task, the elapsed time and a line such as Using Grep or Last tool: Read. Active agents sit on top, under a count of active and finished ones.
• Detail view. View swaps the list for one agent: type, status, elapsed time, parent agent when there is one, and its last eight transcript messages, each marked Prompt or Agent with the tools it called and cut at 600 characters. It reloads when that agent finishes a turn; Back returns to the list.
• History. Finished agents stay in the pane for the rest of the session, even after Claude Code drops them from its live list. A reload or a new session clears it.

Install

Needs Claude Code 2.1.287 or later. The author also runs it in local sessions of the Claude Desktop Code tab, from a desktop build with Claude Code 2.1.286 or later.

claude plugin marketplace add bradtraversy/claude-mods
claude plugin install agent-pulse@traversy-mods

After /reload-plugins or in a new session, ask Claude to use a subagent and the pane appears. /agents opens it by hand, with focus, or brings it back after you close it.

It comes from Brad Traversy’s traversy-mods marketplace, which has eight mods: agent-pulse, session-hud, verification-ledger, impact-map, tool-cards, safe-shell, quality-console and work-queue, the last of which shows the session’s tasks where this one shows who is working.

How it works

The mod hooks every tool call but reads only the tool name and which agent made it. Every two seconds, while the pane is on screen, it asks Claude Code for the live agent list. Transcripts come from Claude Code’s session API, only when you press View, and stay in memory. It runs no processes, reads no project files, makes no network requests and calls no model, so it costs no tokens.

Limitations

• Claude Code can refuse the transcript of an agent running in another process, a remote workflow agent, or one whose saved transcript is gone. The card still shows status and activity, and the detail view says why the transcript is missing.
• Elapsed time counts from when the mod first saw the agent, so an agent already running when the mod loaded shows less than its real time.