Is Your Claude Code Session Stuck? How to Tell — and What to Do

A stuck Claude Code session is dangerous precisely because it doesn't look stuck. It's typing, running commands, sounding confident — and going nowhere. By the time you're sure, you've fed it twenty minutes. Here are the tells, and the fastest way out.

Two kinds of "stuck"

• Genuinely hung (rare): no output for minutes, a spinner that never resolves, the process wedged. That's a process problem, not a reasoning one.
• Spinning (common): the agent is active but not progressing — looping over the same ground, looking busy the whole time. This is the one that costs you.

The tells it's spinning

You're stuck if you see two or more of these:

• It re-reads files it read a few turns ago, as if fresh.
• It re-runs the same command and gets the same error.
• It re-applies an edit that already failed — sometimes verbatim.
• It announces a "different approach" that is, on inspection, the same approach.
• It oscillates — fixes A by breaking B, then B by breaking A.
• The "you're absolutely right" ping-pong: it agrees with your fix, changes one thing, quietly reintroduces the original problem.

If you're nodding, it's spinning. The turn count is going up and the progress isn't.

What to do

If it's genuinely hung: interrupt it (Esc, or Ctrl+C), and if the process itself is wedged, end it and start fresh.

If it's spinning — and it usually is — don't nurse it. A spinning session is a poisoned context, and explaining harder just adds to the pile that's confusing it. Break the context instead:

1. /clear (or a fresh session) to drop the failed attempts.
2. Re-prompt with the one fact it kept missing — the loop was almost always circling a single thing.
3. Hand it the evidence up front — paste the failing test or exact error so it starts from ground truth.

That's the whole move. Deeper version: Claude Code stuck in a loop? Restart it, don't debug it.

The hard part: telling stuck from working

In one session, you'll eventually notice. But when you're running five sessions in parallel, a quietly-spinning agent hides in plain sight — it looks as busy as the productive ones.

That's the gap Mwah closes: it reads each session's live activity and flags it ALIVE or STUCK on your desktop, so the spinning one lights up instead of blending in. See it, and one click ends it — no hunting for the right terminal.

Don't debug the agent. When it's stuck, break the context and go again.

---

Related: Run multiple Claude Code sessions · Shoot your Claude Code in the head