opentelemetry-instrumentation-genai-anthropic gaps vs openinference instrumentation

Ready to Fix

Missing Spans

• Add beta.messages.create span (anthropic.resources.beta.messages.Messages.create / AsyncMessages.create) - maps to GenAI gen_ai.inference.client (chat) #550
• Add beta.messages.stream span (anthropic.resources.beta.messages.Messages.stream / AsyncMessages.stream) - maps to GenAI gen_ai.inference.client (chat) #551
• Add beta.messages.parse span (anthropic.resources.beta.messages.Messages.parse / AsyncMessages.parse) - maps to GenAI gen_ai.inference.client (chat) #552

Missing Attributes & Features

• Capture tool definitions on chat from request tools in messages_extractors.py (OpenInference: llm.tools.*.tool.json_schema, semconv: gen_ai.tool.definitions) #553
• Capture cache write tokens on chat from usage.cache_creation_input_tokens in messages_extractors.py using semconv attribute name (OpenInference: llm.token_count.prompt_details.cache_write, semconv: gen_ai.usage.cache_write.input_tokens) #554

Gaps for Legacy APIs

• Add legacy completions.create span (anthropic.resources.completions.Completions.create / AsyncCompletions.create) - maps to GenAI gen_ai.inference.client (text_completion)
• Support legacy prompt parameter on text_completion (OpenInference: llm.prompts)

Read messages_extractors.py and the listed Anthropic Messages.create, stream, and parse entry points first. Choose one numbered gap, such as issue #550, #551, or #552, and trace its corresponding GenAI mapping. Done means the selected span or attribute is captured according to the requested semantic-convention mapping.