Summary
Amazon Bedrock AgentCore Evaluations recently extended support beyond Strands and LangGraph to additional agent frameworks, including OpenAI Agents SDK and LlamaIndex (see Supported agent frameworks).
Today 01-features/06-observe-evaluate-optimize-your-agent/02-evaluate/ only demonstrates evaluation for the built-in Strands HR Assistant. There is no runnable sample showing how to make an agent built with one of the newly-supported frameworks evaluable end-to-end.
Proposal
Add a supported-frameworks/ subfolder under 02-evaluate/ with one self-contained code sample per newly-supported framework, starting with:
openai-agents/ — the shared HR Assistant re-implemented with the OpenAI Agents SDK, using OpenAI GPT-5.5 on Amazon Bedrock (OpenAI-compatible Responses API + Bedrock API key).llamaindex/ — the same HR Assistant as a LlamaIndex FunctionAgent workflow, using Amazon Nova Lite via BedrockConverse.
Each sample deploys to AgentCore Runtime with OpenTelemetry instrumentation (auto-discovered by ADOT), persists conversation history in AgentCore Memory, and evaluates the agent with built-in + custom LLM-as-a-judge evaluators (on-demand and online), plus an AgentCore CLI re-evaluation example. Reusing the same HR Assistant domain keeps ground truth consistent so results are comparable across frameworks.
Value
Gives users a copy-runnable, framework-specific starting point for AgentCore Evaluations, and a generalized recipe ("make any framework agent evaluable") they can apply to the other supported frameworks.
This supports the AgentCore Evaluations launch content for the newly-supported frameworks.
Summary
Amazon Bedrock AgentCore Evaluations recently extended support beyond Strands and LangGraph to additional agent frameworks, including OpenAI Agents SDK and LlamaIndex (see Supported agent frameworks).
Today
01-features/06-observe-evaluate-optimize-your-agent/02-evaluate/only demonstrates evaluation for the built-in Strands HR Assistant. There is no runnable sample showing how to make an agent built with one of the newly-supported frameworks evaluable end-to-end.Proposal
Add a
supported-frameworks/subfolder under02-evaluate/with one self-contained code sample per newly-supported framework, starting with:openai-agents/— the shared HR Assistant re-implemented with the OpenAI Agents SDK, using OpenAI GPT-5.5 on Amazon Bedrock (OpenAI-compatible Responses API + Bedrock API key).llamaindex/— the same HR Assistant as a LlamaIndexFunctionAgentworkflow, using Amazon Nova Lite viaBedrockConverse.Each sample deploys to AgentCore Runtime with OpenTelemetry instrumentation (auto-discovered by ADOT), persists conversation history in AgentCore Memory, and evaluates the agent with built-in + custom LLM-as-a-judge evaluators (on-demand and online), plus an AgentCore CLI re-evaluation example. Reusing the same HR Assistant domain keeps ground truth consistent so results are comparable across frameworks.
Value
Gives users a copy-runnable, framework-specific starting point for AgentCore Evaluations, and a generalized recipe ("make any framework agent evaluable") they can apply to the other supported frameworks.
This supports the AgentCore Evaluations launch content for the newly-supported frameworks.