[BUG] Roo Code is prone to HTTP 400 errors after multiple rounds of communication.

[aloneSingingStar]

Problem (one or two sentences)

model: Qwen3-Coder-30B-A3B
Symptom: Reopening a conversation usually works normally, but occasionally returns an HTTP 400 error. Multiple rounds of conversations are more prone to HTTP 400 errors.

Context (who is affected and when)

model: Qwen3-Coder-30B-A3B
Symptom: Reopening a conversation usually works normally, but occasionally returns an HTTP 400 error. Multiple rounds of conversations are more prone to HTTP 400 errors.

Reproduction steps

model: Qwen3-Coder-30B-A3B
Symptom: Reopening a conversation usually works normally, but occasionally returns an HTTP 400 error. Multiple rounds of conversations are more prone to HTTP 400 errors.

Expected result

normal

Actual result

http 400

Variations tried (optional)

No response

App Version

Version: 3.31.0 (3ee7ee1)

API Provider (optional)

OpenAI Compatible

Model Used (optional)

Qwen3-Coder-30B-A3B

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote]

I'm exploring a potential fix for the HTTP 400 errors occurring with the Qwen3-Coder-30B-A3B model after multiple conversation rounds. I'll investigate the OpenAI Compatible API handling and open a PR to share progress.

[roomote]

Opened PR #9189 with my attempt at addressing this issue. The fix implements automatic retry logic with progressive conversation history truncation when HTTP 400 errors occur. This should help when conversation history becomes too long for the model's context window. Feedback welcome!

[aloneSingingStar]

A round of dialogue also occurred; was the content too long?

[aloneSingingStar]

Is it because referencing multiple files caused the token to reach the upper limit set by Qwen3-Coder-30B-A3B?