feat: refactor agent skills and support OpenAI Agents SDK

[supreme-gg-gg]

Completes #1068

• Integrates OpenAI SDK (fixed event converters, session service, tools, tracing) and added e2e test for OpenAI SDK
• Refactor skills prompt and tool logic into kagent-skills, update ADK and OpenAI SDK to wrap them with their functions tool implementations

[Copilot]

Pull request overview

This pull request refactors agent skills architecture and adds OpenAI Agents SDK support by creating a centralized kagent-skills package that both ADK and OpenAI integrations can leverage. The refactoring improves code reusability and maintainability while enabling OpenAI agents to utilize the same skills infrastructure as ADK agents.

Key Changes:

• Created kagent-skills package with core skills logic (discovery, session management, shell operations, prompts)
• Added kagent-openai package with OpenAI Agents SDK integration including tools, session service, event conversion, and agent executor
• Refactored kagent-adk to use shared skills package, removing duplicated code

Reviewed changes

Copilot reviewed 50 out of 54 changed files in this pull request and generated 12 comments.

Show a summary per file

File	Description
python/packages/kagent-skills/	New centralized package for skills discovery, file operations, shell execution, and session management
python/packages/kagent-openai/	New OpenAI SDK integration with A2A protocol support, session persistence, and tools
python/samples/openai/basic_agent/	Sample OpenAI agent demonstrating calculator and weather tools
python/packages/kagent-adk/src/kagent/adk/tools/	Refactored to use kagent-skills package instead of local implementations
python/packages/kagent-crewai/	Import order fixes and CrewAI version update to >= 1.2.0
python/Dockerfile	Added kagent-skills package, updated PYTHONPATH and venv configuration
go/test/e2e/	Added e2e tests for OpenAI agent invocation with mocked LLM responses
python/pyproject.toml	Added samples/openai to workspace members

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Missing error handling for the case when BASH_VENV_PATH environment variable is set but the path doesn't exist or is invalid. If bash_venv_path is set but the directory doesn't exist, this could lead to the command failing with unclear errors.

Consider adding validation:

if bash_venv_path:
    if not os.path.exists(bash_venv_path):
        logger.warning(f"BASH_VENV_PATH {bash_venv_path} does not exist, using system python")
    else:
        bash_venv_bin = os.path.join(bash_venv_path, "bin")
        env["PATH"] = f"{bash_venv_bin}:{env.get('PATH', '')}"
        env["VIRTUAL_ENV"] = bash_venv_path

[Copilot]

The TimeoutError exception should be asyncio.TimeoutError for clarity and consistency. While they may be the same in Python 3.13+, being explicit about using asyncio's TimeoutError improves code clarity, especially since asyncio.wait_for is being used above.

Change line 284 from:

except TimeoutError:

to:

except asyncio.TimeoutError:

Suggested change

	except TimeoutError:
	except asyncio.TimeoutError:

[Copilot]

The name 'generate_shell_skills_system_prompt' is exported by all but is not defined.

[supreme-gg-gg]

I used this stub as a temporary solution to fix opentelemetry-instrumentation-openai-agents due to some dependency hell. Ideally we would update it to >0.50.0 since the traceloop import is made optional there, but it conflicts with the opentelemtry api pinned by ADK.

If needed I can look for a more elegant method to fix the issue, but I tried different versions of traceloop and it has some weird version pinning with the instrumentation library and some versions have bugs that just crashed... 😭