OpenAI `images.generate()` and `images.edit()` not instrumented (wrapper and auto-instrumentation)

[braintrust-bot]

Summary

The OpenAI Node SDK provides stable images.generate() and images.edit() methods for AI-powered image generation (DALL-E 3/gpt-image-1), but neither the wrapper (wrapOpenAI) nor the auto-instrumentation plugin instruments them. Calls to these methods produce no Braintrust spans. By contrast, OpenAI embeddings.create() and moderations.create() — also "secondary" AI model APIs — are fully instrumented.

What is missing

• Wrapper (js/src/wrappers/oai.ts): No proxy for images resource. Only chat.completions, embeddings, moderations, and responses are wrapped.
• Auto-instrumentation config (js/src/auto-instrumentations/configs/openai.ts): No config entry for images.generate or images.edit.
• Channels (js/src/instrumentation/plugins/openai-channels.ts): No channel definitions for image generation.
• Plugin (js/src/instrumentation/plugins/openai-plugin.ts): No handler for image generation calls.

Upstream reference

• OpenAI Images API: https://platform.openai.com/docs/api-reference/images
• OpenAI Node SDK Images class: https://github.com/openai/openai-node/blob/master/src/resources/images.ts
• Methods: generate() (text-to-image, supports DALL-E 3 and gpt-image-1), edit() (image editing with streaming support), createVariation() (DALL-E 2 only).
• generate() supports both streaming (ImageGenStreamEvent) and non-streaming (ImagesResponse) modes.
• These are stable, production APIs — not beta or experimental.

Braintrust docs status

The Braintrust wrapper docs page broadly mentions tracing "GPT models, embeddings, and other OpenAI APIs" but does not specifically enumerate images support (unclear).

• https://www.braintrust.dev/docs/instrument/wrap-providers

Note on Responses API overlap

The OpenAI Responses API can generate images via the image_generation built-in tool, and this repo does instrument responses.create() including image_generation_call output conversion to Attachments (via processImagesInOutput in openai-plugin.ts). However, many users still use the dedicated images.generate() endpoint directly, especially for DALL-E 3 and gpt-image-1 workflows that don't use the Responses API.

Precedent in this repo

OpenAI embeddings and moderations — similarly "secondary" model APIs — are instrumented in all three layers:

• Wrapper: js/src/wrappers/oai.ts
• Auto-instrumentation config: js/src/auto-instrumentations/configs/openai.ts
• Plugin: js/src/instrumentation/plugins/openai-plugin.ts

Local files inspected

• js/src/wrappers/oai.ts
• js/src/auto-instrumentations/configs/openai.ts
• js/src/instrumentation/plugins/openai-channels.ts
• js/src/instrumentation/plugins/openai-plugin.ts
• js/src/vendor-sdk-types/openai.ts
• js/src/vendor-sdk-types/openai-common.ts