opentelemetry-instrumentation-genai-agno gaps vs openinference instrumentation

Ready to Fix

Missing Spans

None

Missing Attributes & Features

• Support streaming responses on invoke_agent and invoke_workflow (Agent.run(stream=True), Agent.arun(stream=True), Team.run(stream=True), Team.arun(stream=True), Workflow.run(stream=True), Workflow.arun(stream=True)) (OpenInference: run_stream/arun_stream, semconv: gen_ai.output.messages) (#556)
• Record error status on tool execution failure on execute_tool (FunctionCall.execute, FunctionCall.aexecute) (OpenInference: function_call.error, semconv: error.type) (#557)
• Support streaming tool results on execute_tool (FunctionCall.execute, FunctionCall.aexecute) (OpenInference: _handle_success_streaming, semconv: gen_ai.tool.call.result) (#558)
• Capture conversation ID from request arguments or instance on invoke_agent and invoke_workflow (OpenInference: session_id, semconv: gen_ai.conversation.id) (#559)

Blocked on Semantic Conventions

Missing Spans

• Add Step.execute and Step.aexecute spans (agno.workflow.step.Step.execute, agno.workflow.step.Step.aexecute) (https://github.com/open-telemetry/semantic-conventions-genai/issues/188)
• Add Parallel.execute and Parallel.aexecute spans (agno.workflow.parallel.Parallel.execute, agno.workflow.parallel.Parallel.aexecute)

Missing Attributes & Features

• Capture in-memory agent instance ID on invoke_agent (OpenInference: agno.agent.id)
• Capture user ID on invoke_agent and invoke_workflow (OpenInference: user.id)
• Capture metadata on invoke_agent and invoke_workflow (OpenInference: metadata)
• Capture knowledge base class on invoke_agent (OpenInference: agno.knowledge)
• Capture run ID on invoke_agent and invoke_workflow (OpenInference: agno.run.id)
• Capture workflow step count, names, and types on invoke_workflow (OpenInference: agno.workflow.steps_count, agno.workflow.steps, agno.workflow.step_types)
• Capture workflow description on invoke_workflow (OpenInference: agno.workflow.description)
• Capture workflow ID on invoke_workflow (OpenInference: agno.workflow.id)
• Capture parallel step count on Parallel.execute (OpenInference: agno.parallel.step_count)

Start with the Ready to Fix items and the listed entry points: Agent.run/arun, Team.run/arun, Workflow.run/arun, and FunctionCall.execute/aexecute. Review issues #556-#559 and the OpenInference mappings before choosing one gap. Done means the selected streaming, error-status, conversation-ID, or tool-result behavior is instrumented according to the referenced semantic conventions.