api-fetch-wrapper

api-fetch-wrapper

The case that breaks first in production — a skill that hits an external service. This example covers the four things every such skill needs: secret handling, retry policy, error normalisation, and no-leak-on-failure logging.

Contract

Input (stdin, JSON):

{ "latitude": 52.52, "longitude": 13.41 }

Output (stdout, JSON):

{
  "temperatureC": 18.4,
  "windSpeedKmh": 12.6,
  "fetchedAt": "2026-05-19T08:00:00.000Z"
}

Errors — written to stderr as {"error": "...", "cause": "..."} and exit code 1. Error messages NEVER include the raw upstream response body (it might echo a secret); they include the upstream status code and a short canonical reason.

Required environment

| Var | Purpose | |---|---| | OPEN_METEO_URL | Optional override of the upstream host. Defaults to https://api.open-meteo.com. | | (none for auth) | Open-Meteo is keyless. For an auth'd API, the same skeleton reads process.env.YOUR_API_KEY and passes it via Authorization header. |

Run locally

cd examples/api-fetch-wrapper
bun install
echo '{"latitude":52.52,"longitude":13.41}' | bun run src/index.ts

Adapt this

• Different API — replace the URL, query params, and response shape. The retry / error-mapping skeleton stays.
• API-key auth — read process.env.YOUR_API_KEY once at the top, pass it via Authorization: Bearer ${apiKey}, and ensure the key never appears in error messages.
• Stronger retry — bump MAX_ATTEMPTS, add exponential backoff with jitter, or distinguish 5xx (retry) vs 4xx (do not retry) explicitly.
• Cache — wrap the fetch in a TTL cache when the upstream rate-limits.