> ## Documentation Index
> Fetch the complete documentation index at: https://docs.interaone.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Testing

> Run InteraOne's existing Vitest coverage, package checks, integration tests, and manual end-to-end scenarios.

The gateway currently has the formal Vitest suite, including auth, analytics, JWT, middleware, error handling, validation, and an integration test. Other packages primarily expose build/lint/type-check workflows and need targeted manual or new automated coverage.

## Commands

```bash theme={null}
# Gateway tests
pnpm --filter @InteraOne/gateway test

# Watch or coverage when supported by the package scripts
pnpm --filter @InteraOne/gateway test:watch
pnpm --filter @InteraOne/gateway test:coverage

# Repository checks
pnpm lint
pnpm check-types
pnpm build
```

Use the exact script names from the current package manifest if they differ.

## Test layers

<Tabs>
  <Tab title="Unit">
    Validation, pure policy, adapters with mocked providers, formatting, protocol guards, chunking, and error mapping.
  </Tab>

  <Tab title="Service/integration">
    Tenant filters, role hierarchy, auth flows, model methods, queue producers/consumers, database indexes, and webhook signature validation.
  </Tab>

  <Tab title="End to end">
    Signup, tenant switching, member invite, widget conversation, AI stream, human handoff, knowledge reindex, ticket flow, and every configured channel.
  </Tab>
</Tabs>

## High-risk regression cases

* A record ID from organization A cannot be read or changed from organization B.
* A delayed AI job does not reply after human assignment or closure.
* Duplicate webhook/job delivery does not create duplicate tickets/messages.
* Widget messages from the wrong origin/window/version are ignored.
* Changing embedding dimensions fails safely and does not corrupt existing vectors.
* Owner-only and enterprise-only operations cannot be reached through the console or API by lower roles.

<Tip>
  A passing build is not a behavioral test. Add a regression test that fails before the fix whenever the package has a suitable test harness.
</Tip>