ワンクリックで
typescript
Apply when writing TypeScript code. Strict types, discriminated unions, async patterns, and runtime safety.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
メニュー
Apply when writing TypeScript code. Strict types, discriminated unions, async patterns, and runtime safety.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Git-versioned agent memory — agents that never make the same mistake twice.
Apply when generating ideas, exploring solution space, or facilitating divergent thinking before committing to an approach.
Apply when closing out a feature branch — pre-merge checklist, rebase, CI verification, cleanup, and post-merge steps.
Apply when writing or refactoring code. Generic rules to prevent the most common review comments — function length, naming, error handling, security, and tooling.
Apply when designing database schemas, writing migrations, or reviewing table structure. Covers naming, keys, indexes, constraints, nullability, and migration safety.
Apply when diagnosing a bug, reproducing a failure, or performing root cause analysis. Covers systematic isolation, binary search, logging strategy, and hypothesis-driven investigation.
| name | typescript |
| description | Apply when writing TypeScript code. Strict types, discriminated unions, async patterns, and runtime safety. |
| license | MIT |
| version | 1.0.0 |
| tokens_target | 2500 |
| triggers | ["typescript code","strict types","async typescript"] |
| loads_after | ["code-quality"] |
| supersedes | [] |
Purpose: Prevents the TypeScript-specific mistakes LLMs make repeatedly — weak types, unsafe assertions, and patterns that compile but fail at runtime.
Where these rules don't strictly apply: test fixtures, generated types (e.g. from GraphQL codegen, OpenAPI generators, Prisma), declaration files (*.d.ts) for untyped third-party libraries, and migration scripts may legitimately differ. The rules below apply to production application code.
MUST: Never use any. Use unknown and narrow it. any disables the type checker entirely. unknown forces you to prove the type before use. Exception: third-party libraries without types and explicit dynamic-data boundaries (e.g. JSON parse at the API edge), with a comment explaining why.
// Wrong
function parse(data: any) { return data.name; }
// Correct
function parse(data: unknown): string {
if (typeof data === 'object' && data !== null && 'name' in data) {
return String((data as { name: unknown }).name);
}
throw new Error('Invalid data shape');
}
AVOID: Object or {} as a type. Both accept nearly everything. Use Record<string, unknown> for arbitrary objects or define an explicit interface.
// Wrong
function merge(a: {}, b: Object): {} { ... }
// Correct
function merge<T extends Record<string, unknown>>(a: T, b: Partial<T>): T { ... }
SHOULD: Use as only when you know more than the compiler — and document why. Prefer type guards or satisfies instead.
// Wrong — silences the error, hides the bug
const user = response.data as User;
// Correct — validate first
function isUser(v: unknown): v is User {
return typeof v === 'object' && v !== null && 'id' in v && 'email' in v;
}
const user = isUser(response.data) ? response.data : null;
SHOULD: Mark immutable data readonly. Prevents accidental mutation and communicates intent.
// Avoid
function process(ids: string[]) { ids.push('extra'); }
// Prefer
function process(ids: readonly string[]) { /* ids.push() is a compile error */ }
MUST: Enable strictNullChecks and handle every T | undefined. Optional chaining ?. returns undefined — always handle that branch.
// Wrong
const name = user?.profile.name.toUpperCase(); // crashes if name is undefined
// Correct
const name = user?.profile.name?.toUpperCase() ?? 'Anonymous';
SHOULD: Use branded types for IDs. Prevents passing a UserId where an OrderId is expected — both are string at runtime.
type UserId = string & { readonly _brand: 'UserId' };
type OrderId = string & { readonly _brand: 'OrderId' };
function createUserId(raw: string): UserId { return raw as UserId; }
function getUser(id: UserId): User { ... }
// getUser(orderId) → compile error
SHOULD: Use discriminated unions for state, not optional fields. Optional fields force you to reason about all combinations. A discriminated union makes illegal states unrepresentable.
// Avoid — 8 possible combinations, most invalid
type Request = { loading?: boolean; data?: User; error?: Error };
// Prefer — exactly 3 valid states
type Request =
| { status: 'idle' }
| { status: 'loading' }
| { status: 'success'; data: User }
| { status: 'error'; error: Error };
SHOULD: Use satisfies to validate shape without widening the type. as const preserves literals; satisfies validates against an interface without losing them.
const config = {
host: 'localhost',
port: 5432,
} satisfies DatabaseConfig;
// config.port is still typed as 5432, not number
SHOULD: Use const objects instead of enum. Enums emit runtime code, have surprising reverse-mapping behavior, and are not idiomatic TypeScript.
// Avoid
enum Direction { Up, Down, Left, Right }
// Prefer
const Direction = { Up: 'Up', Down: 'Down', Left: 'Left', Right: 'Right' } as const;
type Direction = typeof Direction[keyof typeof Direction];
AVOID: Barrel index.ts re-exports in large modules. They cause circular dependency chains that are hard to debug. Export directly from source files or use explicit named re-exports only.
// Wrong — index.ts re-exports everything, A imports B through index, B imports A through index
export * from './userService';
export * from './orderService';
// Correct — import directly
import { getUser } from './services/userService';
MUST: Type your thrown errors explicitly. catch (e) gives you unknown in strict mode. Narrow before accessing properties.
try {
await fetchUser(id);
} catch (e) {
// Wrong: e.message — e is unknown
// Correct:
const message = e instanceof Error ? e.message : String(e);
logger.error('fetchUser failed', { message, id });
}
SHOULD: Use Result types for expected failures instead of throwing. Throwing for control flow forces callers to know which functions throw and what.
type Result<T, E = Error> = { ok: true; value: T } | { ok: false; error: E };
function parseConfig(raw: string): Result<Config> {
try {
return { ok: true, value: JSON.parse(raw) };
} catch (e) {
return { ok: false, error: e instanceof Error ? e : new Error(String(e)) };
}
}
MUST: Use Promise.all for independent async operations, not sequential await. Sequential awaits multiply latency.
// Wrong — 300ms total if each takes 100ms
const user = await getUser(id);
const orders = await getOrders(id);
const prefs = await getPrefs(id);
// Correct — 100ms total
const [user, orders, prefs] = await Promise.all([
getUser(id),
getOrders(id),
getPrefs(id),
]);
SHOULD: Use Promise.allSettled when partial failure is acceptable. Promise.all rejects on the first failure. allSettled collects all results.
const results = await Promise.allSettled(ids.map(fetchItem));
const succeeded = results
.filter((r): r is PromiseFulfilledResult<Item> => r.status === 'fulfilled')
.map(r => r.value);
MUST: Type test helpers and mocks — never use as any to silence mock errors. Untyped mocks let type errors hide until runtime. Exception: prototyping spikes that are explicitly thrown away before merge.
// Wrong
const mockUser = { id: '1' } as any;
// Correct
const mockUser: User = { id: createUserId('1'), email: 'a@b.com', name: 'Alice' };
SHOULD: Test the discriminated union branches explicitly. Each status variant is a separate code path. One test per branch minimum.
it('renders error state', () => {
const state: Request = { status: 'error', error: new Error('timeout') };
render(<RequestView state={state} />);
expect(screen.getByRole('alert')).toHaveTextContent('timeout');
});
SHOULD: Use expectTypeOf or assertType for type-level tests. Runtime tests cannot catch type regressions.
import { expectTypeOf } from 'vitest';
expectTypeOf(createUserId('x')).toEqualTypeOf<UserId>();
These rules target the gap between "TypeScript that compiles" and "TypeScript that is safe". LLMs default to any, skip discriminated unions, and reach for as assertions because they are the path of least resistance. Each rule here closes a specific escape hatch that lets type errors reach production. The MUST/SHOULD/AVOID classification means safety-critical rules are strict and stylistic rules respect context.