New serverless pattern - strands-agentcore-lambda

[humem001]

Serverless AI agent pattern using Strands Agents SDK, AWS Bedrock, and AgentCore Gateway with Lambda interceptors for multi-tenant JWT authentication and tool execution via MCP protocol.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[marcojahn]

Hello @humem001,
Thanks for the submission — strong pattern overall. Well-architected, good security posture (per-function IAM, least-privilege), and thorough documentation.

Please fix

1. AWS naming conventions in README.md:
   • "AWS Bedrock" → "Amazon Bedrock" (Amazon-branded, not AWS-branded)
   • "Cognito" → "Amazon Cognito" on first reference

Suggested improvements

• Folder structure — ~10 loose scripts at root level alongside organized src/, tests/, infrastructure/. Recommend moving to a scripts/ directory.
• IaC tooling — Raw CFN + 6-script packaging pipeline. Consider SAM (sam build && sam deploy replaces the whole pipeline). If AgentCore resources lack SAM support, a one-liner in the README explaining the choice would help.
• Security — CloudWatchLogsFullAccess on the Gateway role is broader than needed. Scope to an inline policy with logs:CreateLogGroup/CreateLogStream/PutLogEvents.
• Portability — Hardcoded us-east-1 in create_cognito_user.py and test_e2e_flow.py. Read from AWS_REGION env var or stack outputs instead.
• Performance — Gateway URL is cached at instance level in AgentProcessor, but the class is re-instantiated per invocation. Move cache to module level for cross-invocation benefit.