feat: partially implements optimize_from_config

[andrewklatzke]

Requirements

• I have added test coverage for new or changed functionality
• I have followed the repository's pull request submission guidelines
• I have validated my changes against all supported platform versions

Describe the solution you've provided

Implements an LDApiClient that can talk to the Launchdarkly API to fetch configs and post results back to the optimization process. Runs is not yet implemented, so the post back will silently fail for now. Properly pulls the config from the LD and maps it to the correct options to run the optimization from the provided config key (with some options for handling the call, etc.).

This also adds validation on the response/input for the POST call

Additional context

There are two paths for running an optimization:

• Passing options directly to the SDK
• Pulling a pre-configured set of options from the LaunchDarkly API

This implements the second path. None of the "core" logic for the optimization has changed.

Call to this method looks like:

    result = await client.optimize_from_config("test-optimization", OptimizationFromConfigOptions(
        project_key="default", 
        context_choices=[
            context_builder("user-123"),
        ],
        handle_agent_call=handle_agent_call,
        handle_judge_call=handle_judge_call,
        base_url="https://ld-stg.launchdarkly.com/", # optional, defaults to app.
    ))

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

[cursor]

Ground truth validation bypasses check but fails downstream

Medium Severity

The early validation in _build_options_from_config treats groundTruthResponses as a valid evaluation criterion (has_ground_truth bypasses the "no criteria" check), but ground truth is never passed to OptimizationOptions and that dataclass's __post_init__ only checks for judges or on_turn. When a config has ground truth but no judges and no on_turn, the early check passes, then OptimizationOptions(judges=judges or None, ...) converts the empty dict to None, and __post_init__ raises with the less helpful message "Either judges or on_turn must be provided" instead of the descriptive one mentioning ground truth.

Additional Locations (1)

• packages/optimization/src/ldai_optimization/client.py#L1058-L1059

 

[andrewklatzke]

This is not yet implemented