fix(server): sanitize reasoning content in v1 compact endpoint#248
Merged
lidge-jun merged 1 commit intoJul 22, 2026
Merged
Conversation
The /v1/responses/compact handler forwarded the request body verbatim to
the ChatGPT backend, bypassing the adapter's buildRequest which applies
sanitizeReasoningInputContent. Routed-model reasoning items with
content: [{type: 'reasoning_text', ...}] caused 400 array_above_max_length
errors during remote compaction, permanently bricking threads that mixed
routed and native models.
Export sanitizeReasoningInputContent from the openai-responses adapter
and apply it to the compact body before forwarding, matching the regular
/v1/responses path behavior.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
Issue #234: Threads mixing routed models (Kimi) with native OpenAI models permanently brick when remote compaction triggers. The error is
400 array_above_max_lengthoninput[201].content.Root cause
The bridge emits reasoning items with
content: [{ type: "reasoning_text", text: ... }]for routed models. The regular/v1/responsespath appliessanitizeReasoningInputContentvia the adapter'sbuildRequest, which strips this non-standard content before forwarding to OpenAI.But the
/v1/responses/compactendpoint (v1 remote compaction) forwarded the request body verbatim, bypassing the sanitizer entirely. When Codex tried to compact a thread containing routed-model reasoning items, OpenAI rejected it because it expects reasoning items to have emptycontentarrays.Fix
Export
sanitizeReasoningInputContentfromsrc/adapters/openai-responses.tsand apply it to the compact body insrc/server/responses.tsbefore forwarding, matching the regular/v1/responsespath behavior.Impact
This fixes both new threads going forward and already-bricked threads, since the sanitizer runs on every compact request. No bridge changes needed.
Verification
bun x tsc --noEmitpasses