Catch blocked coding agents in your 15-minute stand-up

The stand-up already has a blocker slot

Your daily stand-up already has a slot for blockers. The room can still be talking about yesterday while a coding agent is running, waiting, or stuck. A local board makes that state visible in the same ritual, so the team can name the blocker and assign the next action without leaving the meeting.

Agent Standup is free and open source, and it works wherever Claude Code runs. It takes the local session files already written by Claude Code and Codex and turns them into a live call with per-agent status plus a board. It avoids Claude calls unless you switch on French translation or jokes. For a small team, that keeps the monitoring from adding model spend. The same set also offers playback, summaries, hierarchical views, video output, audio, character customization, shared screens, a context meter, token counts, and support for eight languages.

The board turns session files into a meeting

On Node 18 or newer, run the npx command @double_geste/agent-standup --open to open it in 30 seconds, with no install and no editor. The setup should be boring. If the board is up before the human updates start, it is not the bottleneck. The team can keep the existing meeting shape and add one shared view.

The live call shows the state you would otherwise ask for by hand. A tile pulses amber after a command runs for more than one minute, a sleep icon appears when a working agent is quiet for three minutes, and a red counter shows failed tool calls. The per-agent token view counts input, output, and cache writes once per message from Claude Code's records, while cache reads appear separately. Those signals separate motion from waiting before a slow run becomes a wasted one. When several agents are working in parallel, the board becomes the shared view of who is moving and who is not.

If the next day needs a record, the export records the call at 1280×720 with voices and subtitles, saving MP4 when the browser supports it and WebM otherwise. For a stand-up, the video matters as a short artifact you can replay when a blocker is disputed.

The five-line check keeps the meeting short

Use this check in the room. Each line is a question you can answer quickly. It names blocked or waiting sessions and assigns the next human action before the room moves on.

• Running: Is the agent actively producing work, or is it idle?
• Waiting on: What human input, file, permission, or decision is it waiting for?
• Blocked: Is the blocker technical, environmental, or ambiguous?
• Context/token risk: Is token usage or context pressure high enough to restart or split the task?
• Next human action: What one person does next, and by when?

Motion and waiting get named early. The token line protects budget. The final line makes the meeting actionable. If a line cannot be answered quickly, the task is too large or the context is too thin.

Keep the ritual boring

Open the board before the human part of the stand-up. Let it sit on the screen while people give updates. Then take one round on the agent board. Do not narrate every tile. Call out only the states that need a decision: amber, asleep, failed calls, or high token pressure. The ritual stays short enough to fit the daily stand-up, and the next action stays visible.

Amber asks for a command check. Asleep needs a human to read the last output. High token pressure calls for a restart, split, or stop. A rising failed-call counter signals a wrong approach, not just a slow one.

Use replay, recap, tree views, and eight-language support when they save a conversation, not when they add ceremony. Skip a separate agent meeting. Keep the board from becoming another stand-up. If a blocker needs more than a quick answer, take it offline and write the owner and next step in the same ticket before the call closes.