Refactor exported TypeScript types without breaking inference, runtime behavior, declarations, compiler budgets, or downstream consumers.
A type can compile and still be wrong for the people who import it. Conditional types can distribute unexpectedly. Overloads can accept the wrong union. Optional properties can mean something different under another compiler flag. A clean source build can still publish a broken declaration.
This skill turns a difficult public-type change into a controlled refactor. It starts with accepted and rejected consumer calls, inferred results, runtime boundaries, supported TypeScript versions, and the compatibility budget. It then classifies the actual failure, compares sound designs, aligns external data with runtime validation, and builds positive, negative, declaration, clean-consumer, and compiler-cost checks.
Use it for generic APIs, conditional and mapped types, overloads, declaration emit, optionality, recursive utilities, type-level performance, or any refactor where “it compiles” is not enough. The result includes an honest release verdict and exact closure evidence. A bundled checker validates the optional canonical JSON record; it does not run TypeScript or claim the code is correct.
Input
Review a registerMany API where an Entry[] widens each key to the whole PluginMap union. Consumers augment PluginMap, runtime keys come from JSON, and get() currently promises a value even for unknown keys.
Output
A blocked release verdict; correlation root cause; mapped-union and tuple-generic design comparison; runtime membership boundary; augmentation, union-key, negative, declaration, and clean-consumer probes; SemVer and migration gates.
Input
Redesign start(options: DeepPartial<Options>) where environment fallback may be missing, JSON can contain null or wrong header values, and consumers compile with exactOptionalPropertyTypes both on and off.
Output
Provide the exported surface, representative accepted and rejected calls, current and intended runtime behavior, TypeScript and tsconfig range, declaration or package entry points, compatibility budget, available tests, and any compiler diagnostics. The skill can diagnose, design, implement when authorized, or review a release. It keeps proposed checks separate from observed results and stops when missing evidence prevents a safe decision.
The exported symbols or declarations, package entry point, and representative accepted, rejected, and inference-sensitive consumer calls.
Supported TypeScript versions, relevant tsconfig inheritance and strict flags, module resolution, declaration settings, and the exact commands or fixtures available.
Actual parsing, validation, serialization, failure, and trust-boundary behavior for values represented by the public types.
The intended SemVer budget, downstream consumers or fixtures, migration constraints, and authority to edit or release the package.
SKILL.md; type-design and compatibility rubric; report and canonical JSON contract; read-only JSON checker; Agent Skills interface metadata
No reviews yet.
Separate raw input, environment, validated domain, and override contracts; parser pseudocode without invented dependencies; absence-versus-undefined behavior; runtime rejection cases; compiler/version matrix; compatibility verdict.
Input
Review recursive Paths<T> and ValueAt<T, P> types used across 180 packages. Timings are not comparable, recursive schemas fail with excessive depth, array and index-signature behavior is undecided, and runtime get() may return undefined.
Output
An insufficient-evidence verdict; distribution, recursion, runtime, and declaration findings; bounded-depth versus generated-path alternatives; comparable diagnostics protocol with a decision-required budget; release stop conditions and closure evidence.
Creator
Ssalven1
