[Bug] Test ordering coupling: shared workflowId, sequential state in runner.test.ts

[chronoai-shining]

Severity: Medium

Several test files have implicit ordering dependencies between it() blocks. Tests pass for the wrong reason; a reordering would surface flakes.

1. runner.test.ts — sequential state across describes

sisyphus-api/test/runner/runner.test.ts:62-148:

• :62 starts a research session.
• :75 "returns 409 when workflow already running" — only passes because :62 left a session active. (Also see the previous issue's note about whether 409 is even real.)
• :86 "stops a running workflow" — depends on the session from :62 still being live.
• :95 "returns 404 when no session running" — depends on :86 having just stopped it.
• :135 history-list test asserts total >= 1 "From start/stop tests" — explicit ordering coupling in the assertion.

If file order or it() order changes, half the suite fails or passes for the wrong reason.

2. workflow.test.ts + connector + compile tests share a top-level let workflowId: string

test/workflows/workflow.test.ts:49, 74, 88, 104 — workflowId is declared inside the describe, assigned only by the first it("POST /workflows ..."). Subsequent it() blocks read it. If the POST test fails, every subsequent test fails with cryptic undefined 404s. Same pattern at :133 (connectorId) and :191 (workflowId inside the compile describe).

3. schemas-crud.test.ts — same pattern

createdId shared across the suite (schemas-crud.test.ts:25, 47, 77, 92, 144); the very first POST controls the rest of the suite's outcome.

Remediation

Two patterns to apply:

A) beforeEach factory. Reset the relevant collection + factory a fresh resource at the start of every it:

let workflowId: string;
beforeEach(async () => {
  const res = await request(app).post('/workflows').send(validBody);
  workflowId = res.body.id;
});

B) Reset the in-memory activeSessions Map and the sessions collection in beforeEach of runner.test.ts. Each test should set up its own state.

Drop the "From start/stop tests" assertion (runner.test.ts:135) — pin the count via local setup, not implicit ordering.

Coupled with #8 (test setup refactor) — fixing both is the same surgery.