fix(responses): preserve image tool_choice without pass-through

[llc1123]

Summary

• preserve Responses type-only tool choices such as {"type":"image_generation"} during inbound conversion
• omit tool_choice.name on outbound unless a function name is present
• add regression coverage for gpt-5.5 + image_generation using gpt-image-2 when body pass-through is disabled

Scope

This is intentionally separate from #2217. This PR only fixes the non-pass-through transformer path where tool_choice became {} and Codex returned Missing required parameter: 'tool_choice.type'.

Test

• cd llm && go test ./transformer/openai/responses -run 'Test(OutboundTransformer_TransformRequest_ImageGenerationToolChoice|ConvertToolChoiceToLLM)' -count=1

Summary by CodeRabbit

• Bug Fixes

  • Improved handling of image-generation tool choices when no function name is provided.
  • Requests now correctly omit an unnecessary tool name while preserving the selected tool type.
  • Ensured image-generation requests retain the appropriate model and tool configuration.

• Tests

  • Added coverage for inbound and outbound image-generation tool choice conversion.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 64f27b6d-2efa-406c-93d2-39769bf3a583

📥 Commits

Reviewing files that changed from the base of the PR and between fae797d and ddd73e4.

📒 Files selected for processing (4)

• llm/transformer/openai/responses/inbound.go
• llm/transformer/openai/responses/inbound_test.go
• llm/transformer/openai/responses/outbound_convert.go
• llm/transformer/openai/responses/outbound_test.go

---

📝 Walkthrough

Walkthrough

The OpenAI Responses transformers now preserve non-function tool choice types without requiring a name. Outbound conversion omits empty names. Tests cover inbound conversion and outbound image-generation serialization.

Changes

Image-generation tool choice handling

Layer / File(s)	Summary
Inbound tool choice normalization llm/transformer/openai/responses/inbound.go, llm/transformer/openai/responses/inbound_test.go	Inbound conversion creates a named tool choice when Type is set. It assigns the nested function name only when present. Tests cover a name-less image_generation choice.
Outbound tool choice serialization llm/transformer/openai/responses/outbound_convert.go, llm/transformer/openai/responses/outbound_test.go	Outbound conversion omits the name field when the function name is empty. Tests verify image_generation type serialization and the gpt-image-2 tool model.

Estimated code review effort: 2 (Simple) | ~10 minutes

Mergeability Score: ⚪ Minimal · up to ddd73

This localized change preserves image tool-choice behavior during request conversion and adds regression coverage; no actionable merge-blocking risk remains after normal checks and review.

Possibly related PRs

• looplj/axonhub#707: Adds related image-generation tool-choice handling.
• looplj/axonhub#882: Modifies related provider-to-LLM tool-choice conversion.
• looplj/axonhub#2106: Updates the same Responses API tool-choice conversion paths.

Suggested reviewers: looplj

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly identifies the main change: preserving image tool choices in the Responses transformer without pass-through.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}