[bot] OpenAI and Anthropic tracers do not set `braintrust.span_attributes` with `{"type": "llm"}`

[braintrust-bot]

Summary

The OpenAI chat completions tracer, OpenAI Responses API tracer, and Anthropic messages tracer do not set braintrust.span_attributes with {"type": "llm"} on their spans. Every other integration in the repository correctly sets this attribute. This causes spans from the three highest-traffic integrations to lack proper LLM span categorization in the Braintrust UI.

What is missing

1. These tracers do NOT set braintrust.span_attributes:

• trace/contrib/openai/chatcompletions.go — chatCompletionsTracer.StartSpan() (line 38): no span_attributes set
• trace/contrib/openai/responses.go — responsesTracer.StartSpan() (line 39): no span_attributes set
• trace/contrib/anthropic/messages.go — messagesTracer.StartSpan() (line 38): no span_attributes set

2. These tracers DO correctly set it:

Integration	File	Line
OpenAI embeddings	trace/contrib/openai/embeddings.go	40
Gemini generateContent	trace/contrib/genai/generatecontent.go	121
Gemini embedContent	trace/contrib/genai/embedcontent.go	41
Genkit generate	trace/contrib/genkit/tracegenkit.go	77
Genkit embedder	trace/contrib/genkit/embedder.go	116
LangChainGo LLM	trace/contrib/langchaingo/tracelangchaingo.go	207
LangChainGo embedder	trace/contrib/langchaingo/embedder.go	134
Eino model	trace/contrib/cloudwego/eino/traceeino.go	124
Eino embedding	trace/contrib/cloudwego/eino/traceeino.go	179
ADK	trace/contrib/adk/traceadk.go	326, 430
Bedrock Converse	trace/contrib/bedrockruntime/converse.go	170
Bedrock InvokeModel	trace/contrib/bedrockruntime/invokemodel.go	46, 102

3. Impact

Per Braintrust's advanced tracing documentation, span_attributes.type = "llm" enables:

• LLM-specific span icons in the UI for easy identification
• The "Try prompt" button in the Braintrust UI
• Proper categorization for LLM-specific monitoring and duration metrics

Without this attribute, OpenAI and Anthropic spans appear as generic spans rather than recognized LLM calls, despite being the most commonly used integrations.

4. The fix is straightforward

Add this call in each tracer's StartSpan method, matching the pattern used by all other integrations:

if err := internal.SetJSONAttr(span, "braintrust.span_attributes", map[string]string{"type": "llm"}); err != nil {
    return ctx, span, err
}

Braintrust docs status

Braintrust's advanced tracing docs explicitly document span_attributes.type as a recognized attribute with "llm" as a valid value that enables LLM-specific UI features. Status: supported.

Upstream sources

• This is an internal consistency issue, not an upstream API gap. The braintrust.span_attributes attribute is a Braintrust convention used across all other integrations in this repository.

Braintrust docs sources

• https://www.braintrust.dev/docs/instrument/advanced-tracing — documents span_attributes.type values including "llm", "tool", "score", "function", "eval", "task", "review"

Local repo files inspected

• trace/contrib/openai/chatcompletions.go — StartSpan() method: no span_attributes set
• trace/contrib/openai/responses.go — StartSpan() method: no span_attributes set
• trace/contrib/anthropic/messages.go — StartSpan() method: no span_attributes set
• trace/contrib/openai/embeddings.go — correctly sets span_attributes at line 40
• All other integration files listed in the table above — all correctly set span_attributes