Update mapper for open ai

[nikhilNava]

Task
Support A365 message format for openAi and SK

Result

The latest version of SK no longer reports input and output messages on span attributes.

 Semantic Kernel 1.41.0+ no longer emits gen_ai.input.messages / gen_ai.output.messages as span attributes or span
  events.

  Instead, SK moved message content to Python logging.info() calls with extras like event.name: gen_ai.user.message.
  This is in SK's decorators.py — the _set_completion_input() (line 367-395) and _set_completion_response() (line
  398-447) functions use logger.info() exclusively, not span.add_event() or span.set_attribute().

  The old behavior (documented on Microsoft Learn as of Sep 2024) used span events like gen_ai.content.prompt / 
  gen_ai.content.completion. That's gone in the current version.

openAI

Invoke agent

 {
     "name": "invoke_agent WeatherAgent",
     "context": {
         "trace_id": "0xe9fcb489e54cbb78b6efd7736f96947b",
         "span_id": "0xf1a4f830aafb58ba"
     },
     "kind": "SpanKind.INTERNAL",
     "parent_id": null,
     "status": { "status_code": "OK" },
     "attributes": {
         "gen_ai.operation.name": "invoke_agent",
         "gen_ai.provider.name": "openai",
         "gen_ai.input.messages": "{\"messages\": [{\"role\": \"user\", \"parts\": [{\"content\": \"What's the 
weather in Seattle?\", \"type\": \"text\"}]}], \"version\": \"0.1.0\"}",
         "gen_ai.output.messages": "{\"messages\": [{\"role\": \"assistant\", \"parts\": [{\"content\": \"The 
weather in Seattle is sunny with a temperature of 22°C.\", \"type\": \"text\"}]}], \"version\": \"0.1.0\"}"
     }
 }

inference span

 {
     "name": "chat gpt-4o-mini",
     "context": {
         "trace_id": "0xe9fcb489e54cbb78b6efd7736f96947b",
         "span_id": "0xe328c0caf7f4cb4d"
     },
     "kind": "SpanKind.INTERNAL",
     "parent_id": "0xf1a4f830aafb58ba",
     "status": { "status_code": "OK" },
     "attributes": {
         "gen_ai.operation.name": "chat",
         "gen_ai.provider.name": "openai",
         "gen_ai.request.model": "gpt-4o-mini",
         "gen_ai.usage.input_tokens": 114,
         "gen_ai.usage.output_tokens": 16,
         "gen_ai.input.messages": "{\"messages\": [{\"role\": \"system\", \"parts\": [{\"content\": \"You are a 
weather assistant. Always use the get_weather function.\", \"type\": \"text\"}]}, {\"role\": \"user\", \"parts\": 
[{\"content\": \"What's the weather in Seattle?\", \"type\": \"text\"}]}, {\"role\": \"assistant\", \"parts\": 
[{\"name\": \"get_weather\", \"id\": \"call_3fhF3lmlJ5kaDk7lVJVNmoIv\", \"arguments\": 
\"{\\\"city\\\":\\\"Seattle\\\"}\", \"type\": \"tool_call\"}]}, {\"role\": \"tool\", \"parts\": [{\"id\": 
\"call_3fhF3lmlJ5kaDk7lVJVNmoIv\", \"response\": \"The weather in Seattle is sunny, 22°C.\", \"type\": 
\"tool_call_response\"}]}], \"version\": \"0.1.0\"}",
         "gen_ai.output.messages": "{\"messages\": [{\"role\": \"assistant\", \"parts\": [{\"content\": \"The 
weather in Seattle is sunny with a temperature of 22°C.\", \"type\": \"text\"}]}], \"version\": \"0.1.0\"}",
         "gen_ai.input.messages.0.message_role": "system",
         "gen_ai.input.messages.0.message_content": "You are a weather assistant...",
         "gen_ai.input.messages.1.message_role": "user",
         "gen_ai.input.messages.1.message_content": "What's the weather in Seattle?",
         "gen_ai.input.messages.2.message_role": "assistant",
         "gen_ai.input.messages.2.message_tool_calls.0.gen_ai.tool.name": "get_weather",
         "gen_ai.input.messages.3.message_role": "tool",
         "gen_ai.input.messages.3.message_content": "The weather in Seattle is sunny, 22°C."
     }
 }

[github-actions]

⚠️ Deprecation Warning: The deny-licenses option is deprecated for possible removal in the next major release. For more information, see issue 997.

Dependency Review

✅ No vulnerabilities or license issues or OpenSSF Scorecard issues found.

Scanned Files

None

[Copilot]

Pull request overview

Adds OpenAI-specific message mapping so gen_ai.input.messages / gen_ai.output.messages span attributes are converted into the Agent365 (A365) versioned message schema (v0.1.0) before spans are ended/exported, with unit + integration coverage.

Changes:

• Introduce message_mapper.py to normalize multiple OpenAI message shapes (chat completions, Response API, plain strings) into A365 typed-parts format.
• Apply the mapper in the OpenAI trace processor on span end for invoke_agent and chat operations.
• Add unit tests for the mapper and integration tests validating exported spans contain versioned message payloads.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
libraries/.../openai/message_mapper.py	New mapping layer converting OpenAI message payloads into A365 v0.1.0 wrapper format.
libraries/.../openai/trace_processor.py	Invokes the new mapper during on_span_end to rewrite message attributes before span end.
libraries/.../openai/trace_instrumentor.py	Clears OpenAI Agents SDK trace processors on uninstrument.
tests/observability/extensions/openai/test_message_mapper.py	Unit tests covering supported input/output JSON shapes and fallback behavior.
tests/observability/extensions/openai/integration/test_message_format.py	Integration tests that capture enriched spans from real Azure OpenAI calls and validate versioned messages.