[BUG] Severe productivity regression: long stalls (90s–minutes) with nonsensical reasoning labels, no token usage displayed (v2.1.145)

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?

Claude Code regularly stalls for extended periods — from 90+ seconds up to several minutes (potentially indefinitely) — both on trivial queries AND in the middle of complex coding sessions. During these stalls:

Nonsensical reasoning status labels are displayed, e.g.:

"consulting the rubber duck..."
"doublechecking the doublechecks..."
"Unfurling… (1m 31s · almost done thinking)"

These do not correspond to valid reasoning steps and look like filler/hallucinated status strings.
No token usage is displayed after these turns. Token counts normally appear after each turn; during this degraded behavior they are absent entirely. This makes it impossible to verify what was actually consumed.
Workaround that consistently revives the session: Pressing Esc twice to interrupt, then asking Claude something like "are you done?" — after that, the session usually continues normally. This is reproducible and suggests the model is stuck in some kind of loop or waiting state rather than genuinely working.
Earlier in the same session I also hit repeated API Error: terminated (streaming connection drops).

Productivity impact
This is not a minor annoyance. My productivity has dropped by an estimated factor of 3–4x over the past days. As a paying customer I am actively evaluating switching to a different provider, because the service no longer delivers what I have come to expect and what I am paying for. I would much rather see this fixed than switch — hence this report.
Environment

Claude Code version: 2.1.145 (latest, freshly updated)
OS: WSL2 + DevContainer (Podman), user vscode
Model: Opus 4,7

Fresh session: yes (cleared after update)
Plan: Max

Screenshot shows the stall: waiting for 9m 48s for just a simple commit...and would probably still wait.
<img width="602" height="169" alt="Image" src="https://github.com/user-attachments/assets/7a41972f-bcaf-46d3-ba93-ea318c1927b7" />

What Should Happen?

Expected behavior

Thinking phases proportional to query complexity.
Valid reasoning status labels.
Token usage consistently displayed after each turn.
No need for Esc Esc + nudge workaround to unstick the session.

Error Messages/Logs

Steps to Reproduce

Steps to Reproduce
Not always immediate, but occurs reliably within a working session:

Start fresh session, normal coding work.
At some point — sometimes on a trivial query, sometimes mid complex task — the thinking phase stalls for 90s to several minutes.
Status labels show nonsensical strings like "Unfurling…", "consulting the rubber duck...", "doublechecking the doublechecks...".
No token usage is displayed when (or if) the turn completes.
Workaround: Esc Esc to interrupt, then ask "are you done?" — session resumes.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

dunno

Claude Code Version

latest

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

WSL (Windows Subsystem for Linux)

Additional Information

Additional notes

Same behavior observed across both fresh sessions and ongoing ones.
Update from an earlier version (2.1.112 → 2.1.145) did not resolve it.
This appears similar in pattern to the March/April 2026 reasoning regression that was acknowledged in Anthropic's April 23 postmortem, though that was reported as resolved.