LLM Injection Assessment and Templates Hardening

[broomva]

Understand how the LLM Explorer could be exploited to query specific parts of the lake house. Parts of these security assessments relies on the roles and policy provided within the access token, however, injection could affect how the app behaves and return unfaithful data, wrong reports and queries, and ultimately reducing the quaility of the product. The idea is to define a set of strategies within the promoting templates and validation steps that help on ensuring the LLM’s input has a high probability of completing the requested within the bounded setup