fix: keep responses reasoning adjacent to messages

[yzlu0917]

Summary

• emit assistant message items before function_call items when a Responses API assistant turn contains both text and tool calls
• preserve the existing function-call behavior for tool-only turns while keeping the message and reasoning items adjacent
• add a regression test covering a reasoning turn that emits both commentary text and a tool call in the same assistant response

Why

OpenAI Responses reasoning models require the reasoning item to stay adjacent to its assistant output item. Continue currently emits function_call items before the assistant message item when a turn contains both text and tool calls, which can separate the reasoning item from the message it belongs to and trigger:

Item 'msg_...' was provided without its required 'reasoning' item: 'rs_...'.

Reordering these items keeps the reasoning/message pair intact without changing the tool-call payloads themselves.

Validation

• ran npm test -- --runInBand openaiTypeConverters.test.ts in core

Closes #11994

---

Summary by cubic

Reorders Responses output so assistant messages are emitted before function_call items when a turn has both text and tool calls. This keeps reasoning next to its message and fixes the “Item 'msg_...' was provided without its required 'reasoning' item” error (addresses #11994).

• Bug Fixes
  • In toResponsesInput, emit the assistant message first and then tool calls; keeps reasoning/message adjacent and leaves tool-only turns unchanged.
  • Added a regression test for a mixed reasoning + tool-call turn to lock the new ordering.

^{Written for commit feff4df. Summary will update on new commits.}

[cubic-dev-ai]

No issues found across 2 files

[chatgpt-codex-connector]

💡 Codex Review

continue/core/llm/llms/Ollama.ts

Line 514 in feff4df

if (options.tools?.length && ollamaMessages.at(-1)?.role === "user") {

Skip tools when Ollama template lacks .Tools support

This change removes the guard that previously suppressed chatOptions.tools when /api/show indicated the model template does not support tools. As a result, any tool-enabled request now sends tool definitions to models that explicitly advertise no tool support, which can cause Ollama chat calls to fail for those models instead of degrading gracefully to normal text responses. This is user-impacting whenever a non-tool-capable Ollama model is configured with Continue features that attach tools.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".