[FEATURE] LlamaFirewall support in Strands SDK

[theagenticguy]

Problem Statement

https://github.com/meta-llama/PurpleLlama/tree/main/LlamaFirewall is a comprehensive suite for guarding inputs and outputs at every step of an Agent loop.

Proposed Solution

Allow customers to use LlamaFirewall, similar to how we support Guardrails.

Use Case

Minimize the risk of prompt injection, insecure code generation (e.g. for python repl tools), etc.

Alternatives Solutions

Continue support only Bedrock Guardrails

Additional Context

No response

[akingscote]

NemoGuardrails and Guardrails AI support would also be great. I assume the expectation is to manually add these checks in ourselves for now. It would be good if there were some examples for doing that

[dbschmigelski]

Hi, marked this with ready-for-contribution. This would be a great task to get started with as the blueprint is already there for Bedrock.

@akingscote, regarding your comment. We are going to work to produce a sample for these model agnostic guardrail systems. The mechanism for doing this would be through the new Hooks system. We will update here in a sample is uploaded to strandsagents.com