[bot] OpenAI Videos API (`videos.create()` / Sora) not instrumented

首先将 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 和 js/src/vendor-sdk-types/openai.ts 与现有的 provider instrumentation 进行比较。确定应如何表示 create、polling 和 download 的生命周期，然后在 e2e/scenarios/openai-instrumentation/ 下添加覆盖。完成的标准是：videos.create 已按照约定的生命周期完成 instrument，并且 e2e 场景验证生成的 spans。

Summary

The OpenAI Node SDK (openai npm package, >=6.2.0) exposes a GA client.videos.create() API for text-to-video and image-to-video generation using the Sora 2 / Sora 2 Pro models. This is a distinct generative execution surface — separate from chat completions, the Responses API, and images.generate() — and is completely uninstrumented in this repo. Users generating video with Sora through the JS SDK get no Braintrust spans.

This repo already treats the analogous gaps for other providers/modalities as in-scope and trackable:

• OpenAI images.generate() / images.edit() — open gap braintrustdata/braintrust-sdk-javascript#1628
• Google GenAI models.generateVideos() (Veo) — open gaps braintrustdata/braintrust-sdk-javascript#2030 / braintrustdata/braintrust-sdk-javascript#2176

The OpenAI videos.* resource is the same category of gap (a dedicated generative media API, distinct from the text/chat path) but for OpenAI specifically, and it is not covered by any of those issues or any other open issue in this repo.

What instrumentation is missing

• Wrapper (js/src/wrappers/oai.ts): No proxy for the videos resource. Only chat.completions, embeddings, moderations, and responses are wrapped (confirmed by reading the file — no video references at all).
• Auto-instrumentation config (js/src/auto-instrumentations/configs/openai.ts): No config entry for videos.create (or videos.remix/videos.retrieve). The file only instruments chat.completions.create, embeddings.create, chat.completions.parse, moderations.create, chat.completions.stream, and the responses.* family.
• Channels (js/src/instrumentation/plugins/openai-channels.ts): No channel definitions for video generation (confirmed — no video references).
• Plugin (js/src/instrumentation/plugins/openai-plugin.ts): No handler for video generation calls.
• Vendor types (js/src/vendor-sdk-types/openai.ts): No Videos/VideoCreateParams types.
• e2e: No test scenario under e2e/scenarios/openai-instrumentation/ exercises video generation.

Upstream reference

• OpenAI Videos API reference: https://developers.openai.com/api/reference/resources/videos/methods/create
• Video generation guide: https://developers.openai.com/api/docs/guides/video-generation
• client.videos.create({ model, prompt, size, seconds, input_reference }) — model accepts sora-2 / sora-2-pro; returns a video job object (object: "video", with id, status, created_at, model) that is polled via videos.retrieve(id) until the render completes, then downloaded via videos.downloadContent(id).
• This is a stable, GA, production API available in openai Node SDK >=6.2.0 — well within this repo's already-supported openai version range (>=4.0.0, and the repo's own dev-pin is [email protected], which already includes this resource).
• Similar to models.generateVideos() in Google GenAI, this uses a long-running-operation/polling pattern rather than a single synchronous response — instrumentation will likely need to decide whether to span just the initial create() call or track the full poll-to-completion lifecycle (the existing google-genai-generate-videos issues note the same design question).

Braintrust docs status

not_found — Fetched https://www.braintrust.dev/docs/integrations/ai-providers/openai directly: it documents chat completions, embeddings (via gateway), audio transcriptions, images, structured outputs, and function calling as traced via wrapOpenAI, but does not mention videos.create, Sora, or any video generation capability anywhere.

Local repo files inspected

• js/src/wrappers/oai.ts — no video references
• js/src/auto-instrumentations/configs/openai.ts — full method list confirmed (chat completions, embeddings, parse, moderations, stream, responses create/stream/parse/compact); no videos entry
• js/src/instrumentation/plugins/openai-channels.ts — no video references
• js/src/vendor-sdk-types/openai.ts — no video/Video references
• e2e/scenarios/openai-instrumentation/ — no video generation scenario
• Full repo grep for video/Video under js/src/ — only unrelated matches in attachment-utils.ts, google-genai.ts vendor types, and strands-agent-sdk files; none reference OpenAI's videos resource