Chat is a poor display for multi-step agent work. It does not clearly show how much context remains, which tool is running, whether child agents are active, or whether the task is moving.
Claude HUD puts those states under the input line: context usage, active tools, running agents, and task progress.
What the operator needs to see
A useful instrument panel should not show everything. It should show states that help the human decide:
- context health shows when it is time to compact or hand off;
- tool activity shows what the agent is doing right now;
- running agents helps the operator avoid interrupting child work;
- todo progress separates real movement from polished silence.
This is not security and it is not a full audit trail. A HUD does not decide whether an agent may touch production. But it gives the operator early sensory data: the session is healthy, stuck on a tool, losing pace, or waiting for a human decision.
New Runtime Read
Agent observability does not have to start with a large dashboard. A small status line is useful when it shows the states that drive human decisions.
For agent UX this is an important shift. Operators need quick access to work state: what is running, what blocks, what is done, and when to intervene.