Loop when failing on simple tasks

[trueliarx]

What happened?

1. loops (for hours) when is not able to solve simple tasks as editing a file, when the searched string can't be found.
2. often goes over the max context and it's unrecoverable, except asking to save the context, clean, reload the context.
3. Consumes a lot of tokens without solving even a simple task, pretending that everything was ok.

What did you expect to happen?

1. edit a string in a file.
2. compress before exceeding the max context. (it can't be done manually when already at the limits)
3. avoid extreme misuse of tokens.

Client information

Qwen Code: 0.13.2 (1b1a029)
Runtime: Node.js v25.8.2 / npm 11.6.2
OS: win32 x64 (10.0.19045)
Auth: Qwen OAuth
Model: coder-model
Session ID: 84152b08-dd5a-4a74-ac49-6856a7e56ef6
Sandbox: no sandbox
Proxy: no proxy
Memory Usage: 526.5 MB

Login information

qwen code oauth

Anything else we need to know?

No response

[tanzhenxin]

Thanks for reporting this. This is a known issue where the model can get stuck in a loop on certain tasks — we're tracking it in #3184.

Regarding context overflow: we're working on improvements to automatic context management. For now, you can use /compact to manually compress the conversation before it reaches the limit.

Closing as duplicate of #3184.