neverthrow-return-types（从不抛出返回类型）

Neverthrow Return Types

Goal

Model every compatible function and method with neverthrow return types.

This skill governs return signatures only. If the task also changes how thrown exceptions or rejected promises are captured, use neverthrow-wrap-exceptions alongside this skill.

Apply this rule to:

• standalone functions
• exported module functions
• class methods
• object literal methods
• factory-returned methods
• service, repository, and domain APIs

Do not treat nesting as an exception. A method inside a class or object follows the same rule as a top-level function.

The only allowed exception is an incompatible boundary enforced by the surrounding technology, library, or framework.

Detect Compatibility First

1. Identify the real contract of the function or method before editing.

- Check implemented interfaces, overridden members, framework callback types, decorators, lifecycle hooks, route handler signatures, component signatures, and public SDK contracts. - If the contract already requires Result or ResultAsync, keep it. - If the contract requires another shape, the boundary is incompatible.

1. Treat these as common incompatible boundaries unless the local technology explicitly supports neverthrow returns:

- UI render functions that must return elements or nodes. - Framework route handlers, middleware, loaders, actions, controllers, or resolvers that must return Response, framework reply objects, void, or Promise of those values. - Event listeners, test callbacks, constructors, getters, setters, and other APIs with fixed runtime signatures. - Interface implementations or overridden methods whose declared return type cannot be widened to Result or ResultAsync.

1. When incompatibility exists, keep neverthrow inside the boundary.

- Move business logic into internal helper functions or methods that return Result or ResultAsync. - Convert the final Ok or Err into the framework-native return value only at the outermost boundary. - Do not fall back to ad hoc failure shapes such as null, sentinel values, or framework-specific shortcuts merely because the boundary itself cannot return neverthrow.

1. If the target project does not already depend on neverthrow, add the dependency only when the task allows dependency changes.

- If dependency changes are out of scope, state that the policy cannot be fully applied yet.

Choose the Return Type

1. Use Result<T, E> for synchronous work.
2. Use ResultAsync<T, E> for asynchronous work.
3. Prefer explicit domain error types for E.

- Use discriminated unions, tagged objects, or stable error classes that match the local style. - Avoid unknown, any, and vague string errors unless the codebase already standardizes on them.

1. Use ok(...), err(...), okAsync(...), and errAsync(...) to construct success and failure values explicitly.

Implementation Rules

1. Do not mark a function or method async if its public contract should be ResultAsync<T, E>.

- Return ResultAsync directly so callers can compose with .map, .mapErr, .andThen, and .orElse. - Avoid Promise<Result<T, E>> unless a framework boundary explicitly requires a native Promise.

1. When touching a module, update all edited compatible functions and methods in that scope.

- Do not convert only top-level functions while leaving neighboring compatible class or object methods on non-neverthrow signatures if they are part of the same requested change.

Boundary Pattern

When the outer API cannot return neverthrow, use this pattern:

function createUser(
  input: CreateUserInput,
): ResultAsync<User, CreateUserError> {
  return validateInput(input).asyncAndThen(insertUser)
}

export async function post(request: Request): Promise<Response> {
  const result = await createUser(await request.json())

  return result.match(
    (user) => Response.json(user, { status: 201 }),
    (error) => toErrorResponse(error),
  )
}

Validate Before Finishing

1. Verify every edited compatible function or method now returns Result or ResultAsync.
2. Verify incompatible boundaries adapt from internal neverthrow results instead of bypassing them.
3. Verify imports come from neverthrow and match actual usage.
4. Verify error types are explicit and stable enough for callers.
5. Run the normal local validation for the stack when it is safe and in scope, such as tests, linting, or type checks.

Report the Outcome

When finishing the task:

• State which functions or methods now return Result or ResultAsync.
• State which boundaries remained non-neverthrow and why they were incompatible.
• State how Err values are typed and mapped.