[BUG] Workflow progress indicators (elapsed timer, token counter, current-tool) freeze during final structured-output generation — an active task looks hung

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

During a multi-agent Workflow run, the live progress indicators (elapsed timer, token counter, and "current tool") freeze during the final part of an agent's work — specifically while it is generating a long, schema-constrained structured output with no intervening tool calls. Because the indicators stop updating, an agent that is actively generating tokens looks stalled/hung.

The task itself completes fine and returns a correct, complete result — so this is not a functional bug. It's a "visibility of system status" / observability problem: during pure output generation the progress UI shows no sign of activity, so a healthy run looks frozen. The effect is worst on the longest synthesis/verification stages, where the tool-call-free window is largest.

What Should Happen?

While an agent is generating its final output (i.e., between its last tool call and result completion), the UI should show some sign of ongoing activity — e.g., a running spinner, an incrementing token counter, an "elapsed" timer that keeps ticking off wall-clock, or an explicit "generating output…" state. A live run should never look frozen.

Error Messages/Logs

No error is produced — the workflow completes successfully.

Evidence that the agent was NOT actually hung during the "frozen" window: the agent's transcript file on disk kept growing (observed ~154 KB → ~203 KB over the interval) and its log showed ongoing token generation, all while the timer / token counter / "current tool" fields in the UI stayed frozen. In other words: absence of tool-call events is being treated as absence of work.

Steps to Reproduce

1. Launch a multi-agent Workflow whose final stage is an agent that returns a large, schema-constrained (structured) result after doing tool calls — e.g., a verification/synthesis agent that first runs several web searches, then emits a big structured output.
2. Open the progress view (/workflows) and watch that final agent.
3. After it issues its last tool call and begins composing the structured output, watch the elapsed timer, token counter, and "current tool" fields.
4. Observe that they freeze at their last values for the rest of the generation, making the still-working agent look hung.

Observed on a 6-agent workflow (5 parallel research agents + 1 verification agent, ~8 min total); the freeze was most visible on the final verification agent composing a large structured result after its searches finished.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

1.22209.3

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

_No response_