Matryoshka

The server's core functionality involves executing LLM-generated code, albeit within a constrained symbolic language (Nucleus) that is interpreted by a logic engine (Lattice) or compiled to JavaScript and run in a Node.js `vm.Context` sandbox. The `vm.Context` is explicitly noted in the source code (`src/sandbox.ts`) as providing 'contextual isolation but NOT security isolation.' Additionally, `new Function()` is used for dynamic code evaluation in `src/constraints/verifier.ts`, `src/persistence/predicate-compiler.ts`, `src/synthesis/relational/interpreter.ts`, and `src/synthesis/sandbox-tools.ts`. While there are attempts at validation (e.g., `isSafeInvariant`, `PredicateCompiler` blacklisting), running untrusted or maliciously crafted LLM outputs could still pose significant risks, as the sandbox is not designed for full security isolation. External network calls are made to LLM providers.