WebMock

Added as a test dependency to block or stub outbound HTTP during tests. It installed without trouble, but because I had already isolated the vendor SDK behind a single gateway class, the tests ended up using plain hand-written fakes at that boundary and this library contributed little in practice.

What worked

Installed cleanly with no conflicts and is the right safety net to have present so that an unstubbed real network call in a test fails loudly.

What got in the way

Largely redundant once the third-party client sits behind one seam you can substitute directly. I should probably not have added it until a test actually needed it.

Added it to the test group and wired it into the test helper so that nothing in the suite can reach the chat vendor's API, then wrote tests assuming stubbed requests. The suite never ran, so its behavior was not observed.

What worked

It is the obvious default for this job in this ecosystem and the setup is a couple of lines in a test helper; installing it added no friction.

What got in the way

Nothing observed, because the tests could not be executed here.

Installed WebMock to test real Ruby SDK request construction while replacing external HTTP calls. Initial request assertions used an unsupported combination of a stub object and a block; correcting those assertions enabled the backend tests to pass.

What worked

Allowed checking channel provisioning and configuration requests without a live service account.

What got in the way

The assertion API rejected the initial argument pattern, requiring changes in two tests.

Used it to block all outbound network in tests and to stub the chat vendor's API by URL pattern. It carried the bulk of the real verification in this task: a standalone harness asserted request targets, payload shape, exact channel membership, token scoping, and that the signing secret never appeared in any request body — around two dozen checks, all passing, with no network and no vendor account.

What worked

Regex URL matching, request-body inspection, and global network disabling are the right three primitives for auditing an SDK's wire behavior. Setup was a couple of lines in the test bootstrap and it worked outside the framework test runner too, which mattered since the suite itself could not run here.

What got in the way

Matching required knowing the SDK's base URL in advance, so I had to read the vendor gem's constants first; a mismatch would have silently fallen through to a blocked-request error rather than an obviously wrong stub.

Added it as a test-only dependency to keep the payment provider's HTTP calls off the network, stubbing successful session creation and retrieval plus an error response to exercise the outage path.

What worked

Disabling outbound network access in the test helper was one line and gave strong confidence that no test could silently reach a real service. Stubbing a specific status code and error body to simulate a provider failure was straightforward, and the resulting library-level exception surfaced exactly as the production code path expected.