Linked Knowledge Base, Runtime, and Webapp

[JOrjales-gds]

Summary

Adds Knowledge Base integration to the AgentCore runtime and enables cross-stack wiring between AgentCore, KnowledgeBase, and WebApp constructs. The agent template now conditionally loads a KB retrieval tool when a Knowledge Base is attached, and the AgentCore stack exposes helpers (grant_invoke, environment_variables, runtime_role) that make it straightforward to compose these stacks together in a CDK app.

Changes

• Knowledge Base retrieval in the agent template — the built-in agent conditionally imports the strands_tools.retrieve tool when KB_ID is present, appends a system prompt hint to use it, and adds strands-tools as a dependency.
• Cross-stack integration on AgentCore — exposed runtime_role and runtime_arn attributes, with grant_invoke() method for granting InvokeAgentRuntime permission to other principals (e.g. a WebApp task role), and an environment_variables property that provides the runtime ARN for downstream containers.
• Updated KnowledgeBase docstring — added an example showing how to wire a Knowledge Base to an AgentCore runtime.
• Example: agent_with_kbase.py — standalone CDK app that deploys an AgentCore runtime backed by a Knowledge Base, ready for quick testing.
• Example: webapp_with_agent/ — full end-to-end example with a Streamlit app that invokes the agent, a Dockerfile, devcontainer config, and a README covering both CDK deployment and local smoke testing via idea-app.

Testing

Follow the Readme file in examples/webapp_with_agent/. That walks through the process of creating firstly a linked agentcore runtime with a knowledge base present, but then also linking it back into a local webapp runnign streamlit, using the same structure as other apps (and compatible with smoke-test) such that it should run with other frameworks and as part of the CI/CD pipeline.

185 tests across all constructs, all pass