claude-code - 💡(How to fix) Fix Model stuck in extended thinking loops — consumed 38% weekly token budget with no usable output

Reporting on behalf of user (MrNobodySMC). Session: 2026-05-07, project directory /home/mrnobodytrsm.

What happened

Session started with a straightforward Pine Script debugging task. After one failed fix attempt, the model entered repeated thinking loops and never recovered:

• 11:52–11:57 — 5 min thinking loop, no output. User interrupted.
• 11:57–12:15 — 18 min thinking loop, no output. User interrupted.
• 12:15–12:46 — 31 min thinking loop, produced a response that crashed with API Error: Claude's response exceeded the 32000 output token maximum.
• 12:46–13:34 — 48 min thinking loop after the crash, still no output. User could not interrupt it into responding.

User tried multiple interventions: manually interrupting, asking what I was thinking, /autofix-pr, /superpowers:systematic-debugging, /effort max. None broke the loop. Session had to be manually cleared.

Result

~38% of user's weekly token limit consumed. Zero working code produced.

Request

Token reimbursement for the session. The failure was on the model side — infinite internal reasoning with no output is not a user error.