opentelemetry-instrumentation-genai-bedrock gaps vs openinference instrumentation

Ready to Fix

Missing Spans

• Add invoke_model span (botocore.client.BaseClient.invoke_model) - maps to GenAI chat or text_completion (in progress in #514)
• Add invoke_model_with_response_stream span (botocore.client.BaseClient.invoke_model_with_response_stream) - maps to GenAI chat or text_completion (in progress in #514)
• Add invoke_agent span (botocore.client.BaseClient.invoke_agent on bedrock-agent-runtime) - maps to GenAI invoke_agent (#569)
• Add invoke_inline_agent span (botocore.client.BaseClient.invoke_inline_agent on bedrock-agent-runtime) - maps to GenAI invoke_agent (#570)
• Add retrieve span (botocore.client.BaseClient.retrieve on bedrock-agent-runtime) - maps to GenAI retrieval (#571)

Missing Attributes & Features

• Capture guardrailConfig on chat (OpenInference: metadata.guardrailIdentifier, semconv: aws.bedrock.guardrail.id) (in progress in #514)
• Support aiobotocore async client instrumentation for chat (Converse, ConverseStream) (#572)
• Capture buffered completion stream content on invoke_agent (OpenInference: output.value, semconv: gen_ai.output.messages) (#573)

Blocked on Semantic Conventions

Missing Spans

• Add apply_guardrail span (botocore.client.BaseClient.apply_guardrail) (https://github.com/open-telemetry/semantic-conventions-genai/pull/427)
• Add retrieve_and_generate span (botocore.client.BaseClient.retrieve_and_generate on bedrock-agent-runtime)
• Add retrieve_and_generate_stream span (botocore.client.BaseClient.retrieve_and_generate_stream on bedrock-agent-runtime)
• Add child spans for Bedrock Agent internal server traces (orchestrationTrace, preProcessingTrace, postProcessingTrace, routingClassifierTrace, guardrailTrace)

Missing Attributes & Features

• Capture guardrail assessment results and actions on apply_guardrail (OpenInference: metadata.action, metadata.actionReason) (https://github.com/open-telemetry/semantic-conventions-genai/pull/427)
• Capture citations and references on retrieve_and_generate (OpenInference: retrieval_documents)

Gaps for Legacy APIs

• Support legacy text completion model formats on invoke_model (AI21 completions, Anthropic V1/V2 completions, Cohere generations, Meta generations, Titan output text) (OpenInference: llm.invocation_parameters, output.value) (in progress in #514)

Start by choosing one unblocked Ready to Fix item and inspect its named botocore entry point, such as invoke_agent, invoke_inline_agent, or retrieve. Check the related issues (#569–#573) and the work in progress in #514 before starting. Done means the selected gap is implemented and its checklist item can be marked complete without taking on the convention-blocked work.