l-blondy/up-fetch/upfetch

upfetch

Use up-fetch when you need a reusable fetch client with request-scoped defaults, automatic request/response shaping, runtime validation, retries, and lifecycle hooks.

Mental model

• up(fetchFn, getDefaultOptions?) creates the reusable client.
• getDefaultOptions(input, options, ctx) runs on every request.
• upfetch(input, options?, ctx?) performs one request.
• Keep SKILL.md high-level; load the relevant file under references/ for details.

Minimum pattern

import { up } from 'up-fetch'
import { z } from 'zod'
export const upfetch = up(fetch, () => ({
   baseUrl: 'https://api.example.com',
   headers: {
      Authorization: readToken() ? `Bearer ${readToken()}` : undefined,
   },
   timeout: 5000,
}))
const user = await upfetch('/users/1', {
   schema: z.object({
      id: z.number(),
      name: z.string(),
   }),
})

Workflow

1. Start with client setup and dynamic defaults.
2. If the request needs auth, params, body shaping, or merge semantics, read auth and request shaping.
3. If the response contract matters, read validation, parsing, and errors.
4. If retries, timeouts, or hook timing matter, read retries, timeouts, and lifecycle.
5. If streaming or runtime quirks matter, read streaming and runtime caveats.

High-value rules

• Pass a function as the second argument to up(), not a plain object.
• Read auth and other mutable defaults inside that function so values stay fresh.
• Use params and body instead of hand-serializing query strings or JSON.
• Use schema when you need runtime trust; TypeScript generics alone do not validate payloads.
• If you want error-as-value behavior, set reject: () => false before relying on parseResponse.
• Prefer globalThis.fetch over imported undici.fetch.

References

• Client setup and dynamic defaults
• Auth and request shaping
• Validation, parsing, and errors
• Retries, timeouts, and lifecycle
• Streaming and runtime caveats