# When an outside service goes down

Prompt · Agent harness and modern stack · G04

Asks what happens to the app when Google, the AI Gateway or a service that sends it updates goes down.

## Prompt

```text
Assume each external API, LLM provider, or webhook source this application depends on either hangs for 30 seconds, returns 429 or 503 for an hour, or delivers a burst of out-of-order webhooks on recovery. Trace which transactions or connection pools are held open during external calls, and whether core journeys degrade gracefully or lock up.
```

**What it returns:** Each dependency with its timeout, retry policy, what is held open while waiting, and the user-visible result.

**Skip it when:** Features with no outside calls.

**Tools:** Repository; a local run with outside services made to fail on purpose.

**Changes:** Changes nothing

**Sessions:** One session

## Skills

- [diagnosing-bugs](https://www.atlasscope.dev/skills/diagnosing-bugs): Diagnosis loop for hard bugs and performance regressions.

## Related

- [Parallel work with no limit](https://www.atlasscope.dev/prompts/parallel-work-with-no-limit): Finds work run in parallel with no limit, which can swamp the database connections or an outside service.
- [Code that relies on lucky timing](https://www.atlasscope.dev/prompts/code-that-relies-on-lucky-timing): Finds code that only works because things usually happen in the right order.
- [Fallbacks that look like real data](https://www.atlasscope.dev/prompts/fallbacks-that-look-like-real-data): Finds screens that show a confident zero or an empty list when the real answer is 'we could not load this'.

Source: https://www.atlasscope.dev/prompts/when-an-outside-service-goes-down