Proposal: add optional trust-boundary example for untrusted retrieved content

anviren

Hi Giovanni, I found this repo while looking for practical Agentic RAG examples built with LangGraph. The modular structure makes it a good place to show one production concern that often appears once RAG agents move beyond demo data: retrieved content can start behaving like control flow. In real deployments, the agent may read PDFs, web pages, docs, tickets, or other external content. That content is supposed to be data, but it can still contain instruction-like text that influences the agent. Would you be open to an optional example showing how to add a runtime trust boundary around the LangGraph workflow? I’m working on Omega Walls, an open-source Python package for this exact layer. It already has a LangGraph adapter: ```python from omega.integrations import OmegaLangGraphGuard guard = OmegaLangGraphGuard(profile="quickstart") safe_graph = guard.wrap_graph(compiled_graph) safe_tool = guard.wrap_tool("network_post", network_post_fn) guard_node = guard.build_guard_node() ``` Install: `pip install "omega-walls[integrations]"` Possible contribution: - add a small examples/security/omega_walls_guard/ example - show the same Agentic RAG flow with and without a guard - demonstrate where a trust boundary should sit: graph wrapper, guard node, and tool wrapper - keep it optional and non-invasive GitHub: https://github.com/synqratech/omega-walls PyPI: [https://pypi.org/project/omega-walls/](https://pypi.org/project/omega-walls/) If this fits the repo, I’d be happy to open a PR with a minimal example.

GiovanniPasq

Hi, Thank you for reaching out and for the thoughtful suggestion—your point about retrieved content behaving like control flow in real-world RAG systems is very relevant, and Omega Walls looks like an interesting approach to addressing that concern. That said, I’d prefer not to include this example in the repository at the moment. I’m aiming to keep the scope focused and avoid introducing additional dependencies or layers that might complicate the core learning objectives. I really appreciate you taking the time to share this and to propose a contribution. Wishing you the best with Omega Walls—it looks like a valuable project. Thanks again!