(MOT-4107) refactor(harness): author integration scenarios as Rust builder modules

.github/workflows/ci.yml

@@ -489,9 +489,6 @@ jobs:
       - name: Integration crate unit tests
         run: cargo test --manifest-path harness/evals/integration/Cargo.toml
 
-      - name: Validate integration scenarios
-        run: make -C harness integration-validate
-
       - name: Run integration scenarios
         run: make -C harness integration-e2e III_BIN="${{ steps.engine.outputs.bin }}" INTEGRATION_REPEAT=2
 

console/web/e2e/durable-hydration.spec.ts

@@ -0,0 +1,44 @@
+import { expect, expectPassingResult, openSession, test } from './harness-stack'
+
+test.use({ scenario: 'streamed-text' })
+
+test('hydrates a durable transcript again after a page reload', async ({
+  page,
+  stack,
+}) => {
+  const completed = stack.waitForTurnCompleted()
+  await stack.trigger('harness::send', stack.ready.send)
+  expect(await completed).toMatchObject({ status: 'completed' })
+
+  await openSession(page, stack.ready)
+  await expect(
+    page.locator('[data-message-role="user"]', {
+      hasText: stack.ready.message,
+    }),
+  ).toHaveCount(1)
+  await expect(
+    page.locator('[data-message-role="assistant"]', {
+      hasText: 'fixture complete',
+    }),
+  ).toHaveCount(1)
+
+  await page.reload()
+  await page
+    .getByRole('button', {
+      name: `open ${stack.ready.session.title}`,
+      exact: true,
+    })
+    .click()
+  await expect(
+    page.locator('[data-message-role="user"]', {
+      hasText: stack.ready.message,
+    }),
+  ).toHaveCount(1)
+  await expect(
+    page.locator('[data-message-role="assistant"]', {
+      hasText: 'fixture complete',
+    }),
+  ).toHaveCount(1)
+
+  expectPassingResult(await stack.finish())
+})

console/web/e2e/exactly-once-function.spec.ts

@@ -0,0 +1,34 @@
+import { expect, expectPassingResult, openSession, test } from './harness-stack'
+
+test.use({ scenario: 'exactly-once-function' })
+
+test('renders one completed function call and its durable result', async ({
+  page,
+  stack,
+}) => {
+  const completed = stack.waitForTurnCompleted()
+  await stack.trigger('harness::send', stack.ready.send)
+  expect(await completed).toMatchObject({ status: 'completed' })
+
+  await openSession(page, stack.ready)
+  const functionId = stack.ready.functions.record
+  expect(functionId).toBeTruthy()
+  const card = page.locator('[data-message-role="function-call"]', {
+    hasText: functionId,
+  })
+  await expect(card).toHaveCount(1)
+  await expect(card).toHaveAttribute('data-function-id', functionId)
+  await expect(card).toHaveAttribute('data-function-status', 'done')
+  await expect(
+    page.locator('[data-message-role="assistant"]', {
+      hasText: 'recorded once',
+    }),
+  ).toHaveCount(1)
+
+  const result = await stack.finish()
+  expectPassingResult(result)
+  const recordCalls = (result.evidence?.recorder_events ?? []).filter(
+    (event) => event.kind === 'target_call' && event.function_id === functionId,
+  )
+  expect(recordCalls).toHaveLength(1)
+})