unhardcoded-engine

Extend the unhardcoded-engine (Σ_pol) policy algebra — add a filter predicate, a scorer, a selector, an Xform, a FailPlan action, or a catalog field; and do it without breaking the invariants that make the core trustworthy (determinism, append-only signature, canonical normal form, golden conformance). Load this skill when changing anything under llm_policy/, adding an operator, or touching term admission/normalization/encoding. To merely use or embed the engine, use unhardcoded-engine-use instead.

Use the unhardcoded-engine (Σ_pol) policy algebra — either to EMBED the pure Lua core in a host, or to AUTHOR a policy_ir term that decides which (provider, model) serves an LLM call. Load this skill when you need to integrate llm_policy into a host (lupa/mlua/luerl), write or dry-run a Σ_pol policy, understand the request/trace contract, or debug why a candidate was rejected. For extending the algebra itself (new operators/fields) use unhardcoded-engine-contribute instead.