cmr care ported to akd extension

[sanzog03]

Summary 📝
Porting a CMR agent created through CARE process using the openai agent platform into the AKD as an extension.

Details
A new CMR agent is created going through the CARE process in openai agent creation platform. The provided agent code is ported to AKD as extension for its easy usability on the AKD backend. To easily port the generated code with minimal intervention to it, the code is copied to get_response_async method that is extended from the base agent.

Usage

            from akd_ext.agents.cmr_care_agent import (
                CMRCareAgent,
                CMRCareAgentConfig,
                CMRCareInputSchema,
                CMRCareOutputSchema
            )

            # Initiate Agent
            config = CMRCareAgentConfig()
            agent = CMRCareAgent(config=config, debug=True)

            # Curate input
            test_input = CMRCareInputSchema(input_as_text='Find datasets related to sea surface temperature in the Pacific Ocean')

            # Run agent
            result = await agent.arun(test_input)

            # Display results
            print(f"\nFound {len(result.dataset_concept_ids)} dataset(s):\n")

Tested Changes

• Test cases to find the dataset concept ids using the search query. test with usage

[NISH1001]

@sanzog03 Thanks for working on this. Couple of things:

• Could you put the PR description in a way it's way more like documentation/elaborate. Eg: What are the major changes. What are the minor changes. IF there's any Usage example for the component, provide that usage section as well.
• The naming conventions: for all the base, i'd encourage to use _base.py pattern we use and some of the open source projects follow that convention as well.

Example PR

I'd create a PR template that will be easier to fill in next time.

[NISH1001]

Thank you for quickly having this PR. I have added a detail feedback, particularly how we want to design the openaisdk porting.

[NISH1001]

If this doesn't add anything new fields, maybe just removing the config will be better. Or
Add tracing_params: dict[str, Any]? I'd also add reasoning_effort as part of the OpenAIBaseAgentConfig and probably mcp as well.

class OpenAIBaseAgentConfig(BaseAgentConfig):
  """Configuration for OpenAI agents based AKD Agents.
    This configuration extends BaseConfig to provide OpenAI-specific
    settings for agents built with OpenAI's platform agent builder.
  """
reasoning_effort: Literal["low", "medium", "high",...] = Field(....)
tracing_params: dict[str, Any] | None = Field(default_factory=dict, description="Paramters that can be passed to OpenAI Agents SDK for traceability in the platform API)

[NISH1001]

We can remove this boilerplate as akd.agents._base.BaseAgent already does it. We also don't need to do self.config=config because the base meta class already handles this.

What we can do is:

class OpenAIBaseAgent[
  InSchema: InputSchema,
  OutSchema: OutputSchema,
](BaseAgent):
  """Base class for OpenAI agents.
  Any agent generated from the openai's platform agent builder should inherit from this class.
  """

  config = OpenAIBaseAgentConfig
  
  def __init__(self,
      config: OpenAIBaseAgentConfig | None = None,
      debug: bool = False
    ) -> None:
    super().__init__(config=config, debug=debug)

[NISH1001]

Please change the base filename to akd_ext.agents._base.py. This will make it consistent in naming convention.

[NISH1001]

Do we need this now? IF it's just a dummy, then we can in fact just remove the config.

Normally, I'd inherit the config if there's extra things that are being added. But in this case OpenAIBaseAgentConfig or BaseAgentConfig handles it. Or when I want to override some field values. Eg: reasoning_effort could be set to medium by default for CMR while the base will default to low.

Base on the comment in another thread, I think this can take advantage of just overriding system_prompt with CARE prompt we have.

This way, we can make sure Openai agents SDK is properly ported for configurability.

[NISH1001]

Rename input_as_text to just maybe query to have consistent input field across everything. And reflect the changes accordingly. This will make consistent for any agents like we have in akd core.

[NISH1001]

Move this prompt to CMRAgentConfig.system_prompt.

Since akd.agents._base.BaseAgentConfig.system_prompt already exists. So CMRCareAgentConfig.system_prompt could be overridden from this. It will make it proper in a way that if we have newer care version for CMR which would be just change in system prompt and instructions TBH. So, CMRCareAgentConfig.system_prompt is a proper place.

[NISH1001]

the name could come from tracing_params

[NISH1001]

also if we want to properly generalize theopenaibase agent then...the Agent object could in fact be abstracted out within the base class as self.agent or something.

[NISH1001]

again, this can come from self.config.tracing_params

[NISH1001]

I think we should make use of conversation tracking we have inBaseAgent. Since memory (see) is abstracted in base...we can have OpenAIBaseAgent.memory as list of dict. Looking a tit. IT does seems like it's a list.

[NISH1001]

no need to install deepeval for akd-ext.

[NISH1001]

Please remove deepeval now