Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
test(e2e): read the rollout baseline before the save that rolls the agent
Alfredo Lopez44e0cc0dee43
diff --git a/tests/e2e/playwright/desktop/codex-subscription-approved-tools.spec.ts b/tests/e2e/playwright/desktop/codex-subscription-approved-tools.spec.tsindex fd8eafd20..0474c26d1 100644--- a/tests/e2e/playwright/desktop/codex-subscription-approved-tools.spec.ts+++ b/tests/e2e/playwright/desktop/codex-subscription-approved-tools.spec.ts@@ -193,23 +193,30 @@ test('authenticated user without agent access cannot select the protected agentsawait expect(page.getByTestId('chat-input')).toHaveCount(0)await expect(page.getByTestId('send-button')).toHaveCount(0)expect(await Promise.all(cases.map(readEvidence))).toEqual(before)+ journeyPassed = true+ } finally {// Sign out before the window closes. The session token lives in the macOS// Keychain, keyed by the REST+RPC origin, which a fresh --user-data-dir does// not isolate: leaving it behind signs the next launch in as this- // unauthorized user, and the next test's login form never renders. Signing- // out is part of this journey, so a failure here fails this test rather- // than the one that inherits the session.- await page.getByTestId('nav-settings-menu').click()- await page.getByTestId('logout-btn').click()- await expect(page.getByLabel('Email', { exact: true })).toBeVisible()- journeyPassed = true- } finally {+ // unauthorized user, and the next test's login form never renders. This has+ // to run on the failure path too, because a journey that fails before+ // signing out leaks the session exactly as one that never tried.+ let cleanupError: string | undefined+ try {+ await signOutDesktop(await app.firstWindow())+ } catch {+ cleanupError = 'Desktop sign-out failed; the session stays in the Keychain'+ }try {await app.close()} catch {- if (journeyPassed) throw new Error('Desktop cleanup failed')- // Preserve the original assertion failure when cleanup also fails.+ cleanupError = cleanupError+ ? `${cleanupError}; Desktop cleanup failed`+ : 'Desktop cleanup failed'}+ // Raise cleanup problems only when the journey passed; otherwise the+ // original assertion failure is the one worth reporting.+ if (cleanupError && journeyPassed) throw new Error(cleanupError)}})diff --git a/tests/e2e/playwright/helpers/approved-tools-workflow.ts b/tests/e2e/playwright/helpers/approved-tools-workflow.tsindex c41a3ab4e..feae78fd9 100644--- a/tests/e2e/playwright/helpers/approved-tools-workflow.ts+++ b/tests/e2e/playwright/helpers/approved-tools-workflow.ts@@ -73,6 +73,12 @@ export async function workflowJourney(page: Page, testInfo: TestInfo) {const scenario = workflowScenario()const mode = required('APPROVED_TOOLS_UPSTREAM_MODE')expect((await readEvidence(scenario)).calls).toEqual([])+ // Saving the model binding changes the runtime-token contract, so HCC rewrites+ // the pod-template annotation `clerum.io/runtime-token-revision`+ // (`host-context-controller/src/hostReconciler.ts:3013`) and the Deployment+ // rolls. The baseline belongs before that save: read afterwards it already+ // holds the post-rollout generation, which nothing can exceed.+ let rolloutBaseline = 0await test.step('Select the isolated workflow agent and bind its subscription visibly', async () => {await page.goto('/')await loginControlUiVisible(page)@@ -84,6 +90,7 @@ export async function workflowJourney(page: Page, testInfo: TestInfo) {await model.chooseSubscription(scenario.subscriptionName)await page.getByLabel('Current model', { exact: true }).click()await page.getByRole('option', { name: scenario.modelName, exact: true }).click()+ rolloutBaseline = readAgentDeploymentGeneration(scenario.agentName)const saved = await model.saveHost(scenario.agentName)expect(saved.spec?.model).toMatchObject({provider: 'codex-subscription',@@ -96,9 +103,6 @@ export async function workflowJourney(page: Page, testInfo: TestInfo) {await expect(page.getByRole('button', { name: 'Edit', exact: true })).toBeVisible()})- // The approval map is part of the agent contract, so saving it rolls the- // Deployment exactly as the model binding does. Recorded before the save.- const rolloutBaseline = readAgentDeploymentGeneration(scenario.agentName)await test.step('Require approval for the real native workflow trigger through Advanced settings', async () => {await page.getByRole('tab', { name: 'Advanced', exact: true }).click()await expect(page).toHaveURL(new RegExp(`/agents/${scenario.agentName}/advanced$`))@@ -137,7 +141,11 @@ export async function workflowJourney(page: Page, testInfo: TestInfo) {await expect(advanced.getByText('workflow_trigger', { exact: true })).toBeVisible()})- await test.step('Wait for the agent rollout the approval change triggered', async () => {+ // The approval map is not part of the runtime-token contract: HCC never reads+ // `spec.approval` (no match in `host-context-controller/src`), and the only+ // consumer is `mcp-host/src/core/extensions/mcpApprovalGateController.ts`.+ // What has to converge before Desktop launches is the model binding's rollout.+ await test.step('Wait for the agent rollout the model binding triggered', async () => {await waitForAgentRollout(scenario.agentName, rolloutBaseline)})diff --git a/tests/e2e/playwright/package-lock.json b/tests/e2e/playwright/package-lock.jsonindex 5ea3f49e9..5869bd991 100644--- a/tests/e2e/playwright/package-lock.json+++ b/tests/e2e/playwright/package-lock.json@@ -1,12 +1,12 @@{"name": "clerum-playwright-tests",- "version": "0.1.59",+ "version": "0.1.60","lockfileVersion": 3,"requires": true,"packages": {"": {"name": "clerum-playwright-tests",- "version": "0.1.59",+ "version": "0.1.60","devDependencies": {"@playwright/test": "^1.50.0","@types/node": "^22.0.0",diff --git a/tests/e2e/playwright/package.json b/tests/e2e/playwright/package.jsonindex 0ce24c66d..277ba51f8 100644--- a/tests/e2e/playwright/package.json+++ b/tests/e2e/playwright/package.json@@ -1,6 +1,6 @@{"name": "clerum-playwright-tests",- "version": "0.1.59",+ "version": "0.1.60","private": true,"description": "Playwright E2E tests for Control UI and Desktop App","scripts": {