rigor: Skill for Claude Code

Rigor Built-in Type Import

Use this skill to fold a new core / stdlib class (or a new family of refinement names) into Rigor's catalog-driven inference pipeline. The flow is the same whether you are importing Hash, Range, Set, Pathname, Time, Date, Enumerable-aware projections, or a brand-new Refined-tier predicate.

Background

Rigor's constant-fold dispatcher consults two complementary surfaces:

• Hand-rolled allow lists in lib/rigor/inference/method_dispatcher/constant_folding.rb — INTEGER_UNARY / STRING_BINARY / NIL_UNARY etc. These are the trusted floor.
• Generated catalogues under data/builtins/ruby_core/<topic>.yml — produced offline from the CRuby reference checkout by tool/extract_builtin_catalog.rb and consumed by lib/rigor/inference/builtins/method_catalog.rb plus the per-topic singletons (STRING_CATALOG, ARRAY_CATALOG, …).

The catalog tier is the additive superset; the hand-rolled tier remains the safety net. Adding a new class means deciding for each method whether the catalog's static classification is correct and, when it is not, how to express the override (blocklist, hand-rolled add, RBS-tier rule).

The principled background lives in:

• docs/adr/3-type-representation.md — type-object layout and the OQ3 working decision (Difference + Refined).
• docs/adr/5-robustness-principle.md — strict-on-returns, lenient-on-parameters.
• docs/type-specification/imported-built-in-types.md — the canonical kebab-case refinement names.

Read those before extending the catalogue if you have not already; the decision points below assume the principle's framing.

Workflow

The flow has six stages. The first four are mechanical; the last two are decision-heavy.

Stage 0 — Run the scaffold script (recommended)

tool/scaffold_builtin_catalog.rb automates the mechanical 70 % of stages 1–4 and 7. Run it once and the manual work that remains is just the per-class judgement calls — blocklist curation, fixture body, and the changelog fragment.