Typescript Engineer

TypeScript Engineer

Type-level design, compiler-error diagnosis, and strict-safety refactoring. This skill routes the user's intent to a set of focused rule files in references/; don't try to answer from SKILL.md alone on anything non-trivial.

Reference files live in ${CLAUDE_SKILL_DIR}/references/.

When NOT to use

• Runtime validation — use Zod / io-ts / Valibot (separate concern; types won't validate unknown input at the boundary).
• Refactors that change runtime behavior — this skill preserves behavior. If the change is behavioral, use a refactoring / testing skill.
• Build tooling issues (tsc not found, wrong tsconfig paths, module resolution not finding files) — that's config, not type-level design.
• JavaScript-only questions where types aren't involved.

Decision tree

Identify the user's goal first, then load the matching rule file on demand.

1. "Something doesn't compile / tsc is red"
   → start at references/error-diagnosis.md
   → then the rule file that matches the error category

2. "Design a type / API for X"
   → references/generics-basics.md (always the foundation)
   → then conditional-types.md / mapped-types.md / template-literal-types.md
     depending on whether you need branching, per-key transforms, or string ops

3. "Remove any / tighten types in this code"
   → references/type-narrowing.md (for input validation)
   → references/utility-types.md (for structural transforms)
   → references/generics-basics.md (when a function/class needs to be generic)

4. "Explain / teach concept X"
   → match X in the routing table below

Routing table

Match keywords in the user's request to load the right rule file.

Working style

• Check the version. Read typescript in package.json (or run npx tsc --version) before recommending anything version-gated. TypeScript 5.5 infers type predicates and 5.8 can ban enum outright, so the same advice is right or wrong depending on the target. typescript-versions.md lists what changed.
• Reproduce first. Run tsc --noEmit on the user's code before proposing a fix so you're reasoning about the real error, not a guess. Since 7.0 the compiler is a native binary and whole-project checks are fast — prefer a real run over reasoning from a snippet.
• Simplest type that works. Don't reach for conditional/mapped/template-literal machinery when a plain generic or utility type would do. Complexity has a cost to everyone who reads the code later.
• Validate type-level code. Use the Expect<Equal<A, B>> pattern (or similar) to prove the types are what you claim — see type-testing.md. Types that compile but are wrong are worse than runtime bugs — they silently lie.
• Explain why the type works. Dense types are hard to read; a one-line comment naming the technique (// distributive conditional over UnionKey) pays for itself.

One snippet per category

These are smell-tests — read them, then jump to the reference file for the full pattern.

Eliminate any with a generic

function getProperty<T, K extends keyof T>(obj: T, key: K): T[K] {
  return obj[key];
}
// getProperty({ name: "Alice" }, "name") → inferred as string

See generics-basics.md.

Narrow an unknown response at the boundary

function isUser(value: unknown): value is { id: number; name: string } {
  return (
    typeof value === "object" &&
    value !== null &&
    "id" in value &&
    "name" in value
  );
}

Two things to flag when you write one of these. in proves the keys exist but says nothing about their types, so the annotation is a promise the body doesn't keep — if the shape arrives over the network, validate it with Zod/Valibot instead. And on TS 5.5+ a guard like this often doesn't need the annotation at all; see typescript-versions.md.

See type-narrowing.md and assertion-functions.md.

Preserve literals while enforcing shape

const palette = {
  red: [255, 0, 0],
  green: [0, 255, 0],
} as const satisfies Record<string, readonly [number, number, number]>;
// palette.red → readonly [255, 0, 0]

See as-const-typeof.md.

Reference map

Core patterns — as-const-typeof · array-index-access · utility-types

Generics — generics-basics · builder-pattern · deep-inference

Type-level programming — conditional-types · infer-keyword · template-literal-types · mapped-types

Safety — opaque-types · type-narrowing · assertion-functions · function-overloads

Debugging — error-diagnosis · type-testing · typescript-versions