[codex] Support pretty rendering for Responses API inputs

[alexanderkreidich]

What changed

This teaches the OpenAI ChatML adapter to recognize and normalize OpenAI Responses API request payloads that use input instead of messages, including requests with tool definitions.

It also excludes the raw input field from additionalInput so the conversation is not duplicated in the pretty view, and adds coverage for Responses-style requests plus an embedding payload guard.

Why

GPT subscription traffic is being logged in Responses API request shape, while the existing pretty renderer only recognized chat-completions style messages and Responses output payloads. As a result, request-side pretty rendering showed nothing useful for those rows.

Impact

Responses API requests with input should render as ChatML in the Langfuse pretty view instead of falling back to opaque JSON. Embedding-style payloads are intentionally not reclassified as chat.

Root cause

The adapter detected messages and output, but not request payloads shaped like { input: ... }, which is what the GPT subscription path emits.

Validation

• Added adapter tests for Responses API request detection and normalization
• Added integration coverage to ensure additionalInput excludes input
• Ran git diff --check

Notes

I did not run the full TypeScript/Vitest suite in this environment because the required package manager and dependencies were not available locally.

[github-actions]

@claude review

[alexanderkreidich]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 48993f613c

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".