Agents in a Stand-up Format

Your AI agents, in a stand-up call. See who is waiting on whom and who is blocked, at zero token cost. Agent Standup turns your Claude Code session into a live team meeting: the lead gives orders, sub-agents join, share their screen and report back. It is a bit silly. It is also weirdly useful.

The interface lives as a one-line indicator in the Claude Code status line or tmux panel. A single local server handles communication between the lead agent and any sub-agents that join. No cloud relay is required for basic operation—the server runs on the same machine where Claude Code is active.

How the Call Works

When a stand-up begins, the lead agent frames the session. Sub-agents join voluntarily and share their current screen state. Each agent reports back in one status line, showing what they are working on, what they have completed, and what blocks they are facing. The call records a tree of who is waiting on whom, so the team can see dependencies at a glance.

Context and token usage appear in the report, giving visibility into how much resource each agent consumed. The session is recorded: a replay, a recap, a tree view, and a video export are all available after the call ends. This means team members who missed the live stand-up can catch up later without re-running any commands.

Integration Points

Agent Standup is not tied to a single editor. The same extension works in Cursor, Windsurf, and VSCodium. It can be installed from the file marketplace, or via VSIX for environments without automatic extension hosting. In any terminal, no editor is needed—a click copies the necessary configuration.

A desktop app runs in its own window with a menu-bar indicator. For terminal-only workflows, a one-line Claude Code status line appears, followed by a tmux indicator. On macOS, the desktop app is not signed by Apple, so users must approve it manually the first time it runs.

To get started, you need to have used Claude Code on that machine at least once. After that, the stand-up server is ready to accept connections from any sub-agent in the session.

What You Actually Get

  • A live call with one status per agent.
  • Sub-agents join, share their screen, and leave when their turn is done.
  • Context gauge and token usage are reported for each participant.
  • Replay, recap, tree, and video export are saved after the call.
  • Agents can have custom voices.
  • Your own characters can be assigned to agents.
  • Eight languages are supported out of the box.

Is It Serious?

Mostly for fun. The project was made by a bored dev and his robots. No, it is not a serious toolchain integration. If you enjoy it, the creator invites you to buy him a coffee.

Despite the tongue-in-cheek origins, the tool solves a real visibility problem: in long-running Claude Code sessions, it is easy to lose track of which agent is blocked on what. The stand-up call makes that information explicit, and the recording features mean the knowledge persists beyond the call itself.