title: Testing Examples description: Runnable workflow-testing examples for Loopstack — runWorkflow() in-process workflow tests with scripted HITL answers, parkView() park assertions, coverage() gates, testTool() tool unit tests, fixture auto record/replay, and createContractFake() DI fakes. Companion package to the Testing guide.

@loopstack/testing-examples

Runnable testing examples for the Loopstack automation framework — the patterns from the Testing guide and Testing Methodology as working code.

The example under test is a small deterministic triage flow (src/triage/): a classify_ticket tool grades a support ticket, the workflow presents the severity as a yes/no approval prompt, then waits for the human decision. No LLM — the whole flow is deterministic, so every test is exact.

Scenarios are named for the acceptance criterion they check (it('C1: …')), so the test report reads as a criteria matrix — the discipline from the methodology guide.

What each spec demonstrates

| Spec | Pattern | | ------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | classify-ticket.tool.spec.ts | Tool unit test — testTool().forTool(...), assert on the envelope data (the resultSchema is validated by call()) | | triage-ticket.workflow.spec.ts | In-process workflow test — runWorkflow() with scripted answers; outcome assertions on status/path/result/document(); parkView() for the human touchpoint; a coverage() gate closing the suite | | triage-ticket-replay.spec.ts | Record/replay — the fixture: option (auto record/replay, CI-guarded) as the default, plus an inline replay() fixture for fully-scripted determinism | | triage-ticket-contract-fake.spec.ts | Contract fake — createContractFake() substitutes the tool with a resultSchema-validated response and asserts how the workflow called it |

Run them:

npm test

Recording fixtures from live runs

The fixture: option records a fresh fixture when the file is missing and replays it when present. Record one from a real run:

npm run start:dev                                    # start the backend
loopstack run triage_ticket --arg text="Production is down!" --trace
loopstack runs <run-id> --record src/triage/__tests__/__recordings__/triage.json

Replay is transition-scoped: recordings are matched by the transition they were captured in, then consumed in sequence. Each entry's metadata (tool, workflow, transition, args, and config when present) is asserted against the actual call — when the outgoing args or config drift from the recording the run fails with a drift error naming the tool and both values, the signal to re-validate live and re-record. In CI a missing fixture is an error (not a silent live run) unless LOOPSTACK_RECORD=1 is set.

About

Author: Jakob Klippel

License: MIT