sglang token exedeeded count is not being recognized for auto compact

[koush]

Description

Bad Request: Requested token count exceeds the model's maximum context length of 202752 tokens. You requested a total of 202941 tokens: 170941 tokens from the input messages and 32000 tokens for the completion.
Please reduce the number of tokens in the input messages or the completion to fit within the limit.

That's what sglang is returning

Plugins

No response

OpenCode version

1.14.30

Steps to reproduce

1. exceed context length in open code on an sglang (openai compatible). open code does not recover.

Screenshot and/or share link

No response

Operating System

No response

Terminal

No response

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• GLM / Vertex overflow errors with INVALID_ARGUMENT are not classified as context overflow #23713: Same symptom — the exact same error message pattern ("Requested token count exceeds the model's maximum context length") is not being recognized as a context overflow, so auto-compaction doesn't trigger. A comment on that issue even reproduces the same error text from sglang/Vertex.
• Auto-compaction enabled by default, but context_length_exceeded errors still occur in agent workflows #8089: Broader issue where auto-compaction is enabled but context_length_exceeded errors still occur and compaction does not trigger.

If your issue is specific to sglang's OpenAI-compatible endpoint returning this error shape (as opposed to a standard context_length_exceeded code), feel free to clarify that distinction so it can be tracked separately.

[Arjxm]

I love to work on this