feat: introduce ManagedResult, RunnerResult, and LDAIMetricSummary

[jsonbailey]

Summary

• Adds RunnerResult interface (provider-level result: content, metrics, raw?, parsed? — no evaluations)
• Adds ManagedResult interface (managed-layer result with evaluations: Promise<JudgeResult[]>)
• Adds LDAIMetricSummary (flat summary: success, usage?, toolCalls?, durationMs?, resumptionToken?)
• Adds toolCalls? and durationMs? fields to LDAIMetrics
• TrackedChat.run() replaces/supplements invoke(), returning ManagedResult with metric summary built from tracker
• Adds createModel() to LDAIClient and LDAIClientImpl as the preferred replacement for createChat()
• Updates chat-judge example to use createModel() and run()

Test plan

• All 188 tests pass
• chat-judge example updated to use new API

🤖 Generated with Claude Code

---

Note

Medium Risk
Medium risk due to API shape changes (createModel/createChat, TrackedChat removal) and new provider abstraction (Runner) that can affect downstream integrations and prompt/message handling.

Overview
Introduces a new managed invocation layer via ManagedModel.run() that returns a ManagedResult with flattened LDAIMetricSummary and an asynchronous evaluations promise, alongside provider-level Runner/RunnerResult types.

Replaces the stateful TrackedChat API with stateless model execution: LDAIClient.createModel() is added as the preferred entry point, while createChat/initChat are deprecated shims; examples are updated to use createModel + run. Judge execution is split into a standalone Evaluator (parallel, best-effort) and configs now carry an internal evaluator reference for the managed layer.

Extends metrics to include optional toolCalls and durationMs, and adds initial graph runner result/metrics types to support the new runner protocol.

^{Reviewed by Cursor Bugbot for commit e163f7d. Bugbot is set up for automated code reviews on this repo. Configure here.}

[jsonbailey]

@cursor review

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

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

^{Reviewed by Cursor Bugbot for commit e163f7d. Configure here.}

[cursor]

Evaluator built but never called in run()

High Severity

createModel() builds an Evaluator (initializing judges via async network calls), attaches it to configWithEvaluator, and passes that config to ManagedModel. However, ManagedModel.run() ignores this.aiConfig.evaluator entirely and hardcodes evaluations to Promise.resolve([]). Since the old TrackedChat with working judge evaluations is deleted in this same PR, judge evaluations are silently non-functional. The chat-judge example will always print empty results despite judges being configured.

Additional Locations (1)

• packages/sdk/server-ai/src/LDAIClientImpl.ts#L427-L438

 

^{Reviewed by Cursor Bugbot for commit e163f7d. Configure here.}

[cursor]

Conflicting LDAIMetricSummary interfaces with same name

Medium Severity

Two incompatible interfaces named LDAIMetricSummary now exist. The original in config/LDAIConfigTracker.ts has tokens?: LDTokenUsage and success?: boolean; the new one in model/types.ts has usage?: LDTokenUsage and success: boolean (required). The new one is publicly exported, but LDAIConfigTracker.getSummary() returns the old one — users importing LDAIMetricSummary to type the return of getSummary() will get a type mismatch.

 

^{Reviewed by Cursor Bugbot for commit e163f7d. Configure here.}