GLM 5.3 Flash not receiving arguments or stream ended

Summary

I was trying to find alternative for dsv4f peak hours and I think GLM 5.3 flash is a good model

I was on Go model before, using 5.3 flash with no hassle, both in cmcd, opencode, pi, routed with 9router

Since my Go plan almost ended, I upgraded my plan to GOAT and started using GLM 5.3 Flash.

Weirdly, whenever the model almost reached 100K, It keep getting something like

Validation failed for tool "bash":
   - command: must have required properties command
Received arguments:
 {}

or

Stream ended without finish_reason

I've tried to ask dsv4f to fix this issue, most of the solution says to change the context window to somewhere between 80-90K. The solution sometimes working sometimes not that I reached annoyed state (and moved to dsv4f on peak hours 😭). I believe that this is neither the harness nor 9router's problem (also this is what ds4vf stated too). Just wondering why this happened when I switch plan and it was working fine before

Expected Behavior

GLM 5.3 Flash working fine like a few weeks ago

Actual Behavior

Weirdly, whenever the model almost reached 100K, It keep getting something like

Validation failed for tool "bash":
   - command: must have required properties command
Received arguments:
 {}

or

Stream ended without finish_reason

Steps to reproduce the issue

1. Choose GLM 5.3 Flash as model
2. Produce something with it that it almost reached 80-90K context
3. The Received arguments error will continuously occurred if not auto compact

Command Code Version

1.44.0

Operating System

Linux

Terminal/IDE

Unknown

Shell

cmd.exe

Session file (optional)

No response

Fix prompt (optional)

No response

Additional context

No response

Comece reproduzindo o problema com GLM 5.3 Flash próximo ao intervalo de contexto relatado de 80–100K usando Command Code 1.44.0 no Linux. Compare os argumentos bash vazios repetidos e a ausência de finish_reason na mudança de plano relatada, e considere a issue concluída quando as chamadas de ferramentas receberem argumentos válidos e os streams terminarem normalmente sem exigir auto-compact.