DOC-1892: Fix Semantic Kernel notebook for semantic-kernel 1.x

[snova-gilvanm]

Summary

The Semantic Kernel example notebook does not run. It fails on the first line of main() with:

TypeError: ChatCompletionAgent.__init__() got an unexpected keyword argument 'service_id'

This is the sambanova/integrations half of DOC-1892. The docs half is sambanova/docs#984.

The notebook appears to have been committed without ever completing a run — cells 2-4 carried execution counts 3/4/5 while the imports cell and await main() had none and no outputs.

Blocking fixes

1. service_id removed from ChatCompletionAgent

Removed in semantic-kernel 1.0; the agent takes a configured kernel instead. Matches the current upstream sample, step06_chat_completion_agent_group_chat.py. OpenAIChatCompletion still accepts service_id, so _create_kernel_with_chat_completion is unchanged.

2. metadata.kernelspec was semantic_kernel_venv

A machine-local kernel name. Anyone else opening the notebook gets:

jupyter_client.kernelspec.NoSuchKernel: No such kernel named semantic_kernel_venv

Now python3, which any ipykernel install provides.

Correctness

• Restored the "not approved" guard in ApprovalTerminationStrategy. The check was "approved" in history[-1].content.lower(), so a reviewer replying "this is not approved" terminated the chat as if it had passed. Also guards a None content.
• Subclass TerminationStrategy rather than DefaultTerminationStrategy, matching upstream.

Dependencies

• Added python-dotenv. The notebook does from dotenv import load_dotenv, but nothing declared it — it resolved only transitively via semantic-kernel → pydantic-settings.
• Bounded semantic-kernel>=1.44,<2. On Python 3.9, pip does not error; it backsolves to 0.9.6b1 (2024) which fails with ImportError: cannot import name 'Url' from 'pydantic.networks' — a traceback that names pydantic and never mentions the Python version. The floor turns that silent downgrade into a resolver error.

Housekeeping

• Fixed the logo anchor: stray backslash in <a href="..."\> and crossed <a>/<picture> nesting.
• Fixed the dead upstream link (step3_chat.py → 404) and a levaraging typo.
• Removed the unused asyncio import and sys.path.append('../../'); cleared stale execution counts and outputs.

Verification

Fresh Python 3.11 venv built only from the updated requirements.txt, then nbconvert --execute with no override flags:

EXIT CODE: 0
ERROR CELLS: 0
exec counts: [1, 2, 3, 4, 5]
# Agent - CopyWriter: '"Charge Ahead" - simple, direct, ...'
# Agent - ArtDirector: '... Approved, but with room for refinement.'
# IS COMPLETE: True

Notes for reviewers

• Opened from a fork — I have read-only access to this repo.
• The approval heuristic is still substring-based and terminates on "Approved, but with room for refinement", as the run above shows. Upstream has the same weakness, so I matched it rather than diverging. Happy to harden it if preferred.
• AgentGroupChat is deprecated upstream in favour of GroupChatOrchestration. Migrating the sample is a larger change and out of scope here.
• Eight other notebooks in this repo carry the same machine-local kernelspec defect (agentops_venv, autogen_venv, instructor_venv, venv, haystack_test, langgraph_env, base_env, petro_test_2). Worth a sweep in a separate ticket.

🤖 Generated with Claude Code

[snova-gilvanm]

Closing this cross-repo PR. Reopening from a branch on the official repo now that I have admin access.