fix: omit malformed reasoning items to prevent OpenAI 400 errors #9362

pallaprolus · 2025-12-31T05:36:50Z

Description

This PR addresses an issue where OpenAI models (specifically newer ones like o1-preview) return a 400 error when a reasoning item is provided without its required encrypted_content (malformed/missing).

Changes

Omit Malformed Reasoning: In core/llm/openaiTypeConverters.ts, we now check if encrypted content is present for a reasoning item. If not, the item is skipped.
Handle Dangling References: When a reasoning item is skipped, we flag this state and strip the responsesOutputItemId from the subsequent assistant message. This prevents the API from rejecting the request due to a message ID validating against a missing reasoning item.

Verification

Verified logic ensures that message and reasoning streams remain consistent with OpenAI's requirements for paired/encrypted content.

Summary by cubic

Prevents OpenAI 400 errors by skipping reasoning without encrypted_content only when the next assistant message references it. Strips the assistant responsesOutputItemId when paired reasoning is dropped and preserves tool call IDs.

Bug Fixes
- Use look-ahead to skip reasoning only when the next assistant has responsesOutputItemId.
- Preserve tool call IDs; strip assistant ID when its paired reasoning was dropped.

^{Written for commit c20d4c5. Summary will update on new commits.}

continue-development-app · 2025-12-31T05:36:52Z

Learn more

All Green is an AI agent that automatically:

✅ Addresses code review comments

✅ Fixes failing CI checks

✅ Resolves merge conflicts

Unsubscribe from All Green comments

continue · 2025-12-31T05:36:56Z

Learn more

All Green is an AI agent that automatically:

✅ Addresses code review comments

✅ Fixes failing CI checks

✅ Resolves merge conflicts

Unsubscribe from All Green comments

github-actions · 2025-12-31T05:37:01Z

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

cubic-dev-ai

No issues found across 1 file

pallaprolus · 2025-12-31T05:39:04Z

I have read the CLA Document and I hereby sign the CLA

pallaprolus · 2025-12-31T05:39:11Z

recheck

cubic-dev-ai

1 issue found across 1 file (changes from recent commits).

Prompt for AI agents (all issues)


Check if these issues are valid — if so, understand the root cause of each and fix them.


<file name="core/llm/openaiTypeConverters.ts">

<violation number="1" location="core/llm/openaiTypeConverters.ts:818">
P1: Using `&quot;placeholder&quot;` as `encrypted_content` is likely incorrect. The PR description says malformed reasoning items should be &quot;skipped,&quot; but this code includes them with an invalid placeholder string. OpenAI&#39;s API expects valid encrypted content, so this could still cause 400 errors or unexpected behavior. Consider actually omitting the reasoning item when `encrypted` is falsy, similar to the original logic that returned early.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

core/llm/openaiTypeConverters.ts

pallaprolus · 2025-12-31T14:42:11Z

recheck

RomneyDa

@pallaprolus if this only happens for some newer models, would it cause stripping of all thinking for older models/APIs which do not provide encrypted content?

Added some nitpick comments too

RomneyDa · 2026-01-06T17:09:52Z

core/llm/openaiTypeConverters.ts

+function serializeAssistantMessage(
+  msg: ChatMessage,
+  dropNextAssistantId: boolean,
+  pushMessage: (role: "assistant", content: string) => void,


refactor to avoid passing this once-used callback

Refactored callback → discriminated union { type: 'item' | 'skip', ... }

RomneyDa · 2026-01-06T17:10:33Z

core/llm/openaiTypeConverters.ts

+        name: name || "",
+        arguments: typeof args === "string" ? args : "{}",
+        call_id: call_id,
+      } as any;


remove any type cast

Removed as any on empty item → { type: 'skip' }

RomneyDa · 2026-01-06T17:10:44Z

core/llm/openaiTypeConverters.ts

+  if (id) {
+    if (!encrypted) {
+      // Return empty item signal and flag to drop next ID to prevent 400 error
+      return { item: {} as any, dropNextAssistantId: true };


remove any type cast

Removed as any on function call → ResponseFunctionToolCall has optional id

RomneyDa · 2026-01-06T17:13:04Z

core/llm/openaiTypeConverters.ts

+      // BUT `ResponseInputItem` variants are specific.
+
+      // Alternative: If we are forced to drop the ID, maybe we just don't send the ID field?
+      // Let's try to return the object but cast to any to suppress TS if ID is mandatory but we want to test behavior.


Let's try to...
This comment feels too conversational, could you make it super consice to just describe behavior and why?

Done, Condensed verbose comments → concise JSDoc

Fixes issue continuedev#9359 where OpenAI returns a 400 error when a reasoning item is provided without its required content (malformed). This change: 1. Skips sending reasoning items if they lack encrypted content. 2. Strips the ID from the subsequent assistant message when its corresponding reasoning item is skipped, preventing 'dangling reference' errors.

…ping items

- Use look-ahead logic to check if next assistant message has responsesOutputItemId - Skip reasoning only when: no encrypted_content AND next message has reference - Keep reasoning when no reference (prevents breaking older models/APIs) - Refactor to use discriminated unions instead of callbacks - Remove as any casts - Add unit tests for toResponsesInput reasoning handling

pallaprolus · 2026-01-07T01:47:16Z

@pallaprolus if this only happens for some newer models, would it cause stripping of all thinking for older models/APIs which do not provide encrypted content?

Added some nitpick comments too

Hi RomneyDa, thank you for the review and detailed comments!

You're right that the original fix would strip all reasoning for older models/APIs that don't provide encrypted_content.

I've updated the approach to use look-ahead logic:

Skip reasoning only when: no encrypted_content AND the next assistant message has a responsesOutputItemId reference
Keep reasoning when: encrypted_content is present, OR when the next assistant message has no responsesOutputItemId reference
The key insight is that the 400 error only occurs when:

We send a reasoning item without encrypted_content
AND the following assistant message references that reasoning via its output item ID
If there's no reference (e.g., older conversation formats), there's no risk of a 400 error, so we preserve the reasoning.

Also added unit tests for toResponsesInput covering these scenarios.

pallaprolus requested a review from a team as a code owner December 31, 2025 05:36

pallaprolus requested review from sestinj and removed request for a team December 31, 2025 05:36

github-project-automation bot added this to Issues and PRs Dec 31, 2025

github-project-automation bot moved this to Todo in Issues and PRs Dec 31, 2025

dosubot bot added the size:S This PR changes 10-29 lines, ignoring generated files. label Dec 31, 2025

cubic-dev-ai bot reviewed Dec 31, 2025

View reviewed changes

dosubot bot added size:L This PR changes 100-499 lines, ignoring generated files. and removed size:S This PR changes 10-29 lines, ignoring generated files. labels Dec 31, 2025

cubic-dev-ai bot reviewed Dec 31, 2025

View reviewed changes

core/llm/openaiTypeConverters.ts Outdated Show resolved Hide resolved

RomneyDa requested changes Jan 6, 2026

View reviewed changes

github-project-automation bot moved this from Todo to In Progress in Issues and PRs Jan 6, 2026

pallaprolus added 8 commits January 6, 2026 19:11

refactor: reduce complexity of toResponsesInput helper

0f777ae

fix: preserve tool call IDs even if reasoning is dropped

cdcb2f6

fix: inject placeholder for missing encrypted content instead of drop…

4760893

…ping items

fix(core): resolve TS error 'toolCalls does not exist on ChatMessage'

6ed43c4

fix: revert placeholder; omit malformed items but preserve tool calls

f17632e

ci: re-trigger checks to resolve flakes

a70532b

pallaprolus force-pushed the fix/openai-reasoning-400 branch from cb00ecc to c20d4c5 Compare January 7, 2026 01:40

fix: omit malformed reasoning items to prevent OpenAI 400 errors #9362

Are you sure you want to change the base?

fix: omit malformed reasoning items to prevent OpenAI 400 errors #9362

Conversation

pallaprolus commented Dec 31, 2025 • edited by cubic-dev-ai bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Changes

Verification

Summary by cubic

Uh oh!

continue-development-app bot commented Dec 31, 2025

Uh oh!

continue bot commented Dec 31, 2025

Uh oh!

github-actions bot commented Dec 31, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cubic-dev-ai bot left a comment

Choose a reason for hiding this comment

Uh oh!

pallaprolus commented Dec 31, 2025

Uh oh!

pallaprolus commented Dec 31, 2025

Uh oh!

cubic-dev-ai bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

pallaprolus commented Dec 31, 2025

Uh oh!

RomneyDa left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

pallaprolus commented Jan 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

pallaprolus commented Dec 31, 2025 •

edited by cubic-dev-ai bot

Loading

github-actions bot commented Dec 31, 2025 •

edited

Loading