Claude Code 2.1.162 shows why agents are waiting

When an AI agent stops, the stop itself is rarely the expensive part. The annoying part comes after: is it waiting for approval, missing a tool, or just stuck? Claude Code 2.1.162 makes that question a little less murky.

Source: Claude Code changelog

Source: GitHub release v2.1.162

Source: latest npm package metadata

New in Claude Code 2.1.162: waitingFor in the agent queue

The most useful change is small: claude agents --json now includes waitingFor, showing what a waiting session is blocked on, for example a permission prompt. Agent view, claude agents, is Claude Code's overview for running, blocked, and completed sessions. JSON output makes the same queue readable by scripts, dashboards, or a simple weekly report.

The same release smooths a few everyday edges. If you list Grep or Glob with --tools, native builds now provide the dedicated search tools. /effort confirms when the chosen level is saved as the default for new sessions. Clicking a slash command in autocomplete now fills it into the prompt instead of running it immediately. Small changes, yes. But they reduce the chance that the wrong thing happens when someone is moving fast.

Why this matters in a real work queue

If Claude Code runs background jobs for a team, you do not only want to know that something is waiting. You want to know what the human should do next. A session waiting on a permission prompt needs a decision. A session blocked by the wrong tool setup may need a changed tool list. A session that is done needs review, not more chat.

This is a good Tool Forge pattern: do not just create more AI runs, create small routines around them. A simple queue with session, working area, waitingFor, and the next human decision goes a long way before you need a heavy agent-ops system.

Try this prompt this week

First, do the human setup: pick one repo or internal routine where Claude Code already does background work. Update to 2.1.162, run your normal agent queue, and inspect claude agents --json. Do not put secrets in status fields or logs.

Then ask Claude Code to help make the queue readable:

Read our Claude Code settings, permission rules, and runbook for [workflow].
Suggest a small "blocked agent" note that uses `waitingFor` from `claude agents --json`.
Show which states a human should approve, retry, or stop.
Do not edit files yet. First give me a short proposal I can approve.

A good result is not a perfect dashboard. A good result is that the next person can see why the agent is waiting and make the right call without reading the whole chat history.