Improve CLI E2E test assertions and parallel isolation

- Fix non-falsifiable assertions in /exit, /logout, Ctrl+C tests
- Verify --agent flag shows agent name in UI
- Verify autocomplete shows actual command names
- Add globalCleanupRan flag to prevent parallel tests killing each other

Author: brandonkachen

cli/src/__tests__/e2e/cli-ui.test.ts

@@ -189,14 +189,17 @@ describe('CLI UI Tests', () => {
     )
 
     test(
-      'accepts --agent flag without crashing',
+      '--agent flag sets the specified agent',
       async () => {
         const session = await launchCLI({ args: ['--agent', 'ask'] })
 
         try {
-          await session.waitForText(/ask|codebuff|login/i, { timeout: 15000 })
+          // Wait for the CLI to load and show the agent indicator
+          await session.waitForText(/ask/i, { timeout: 15000 })
 
           const text = await session.text()
+          // Verify the agent name appears in the UI (mode indicator shows agent)
+          expect(text.toLowerCase()).toContain('ask')
           expect(text.toLowerCase()).not.toContain('unknown option')
         } finally {
           await session.press(['ctrl', 'c'])
@@ -253,10 +256,9 @@ describe('CLI UI Tests', () => {
             // Process may have exited before message was captured - that's OK
           }
 
-          // Verify CLI responded to Ctrl+C
-          // If we get here without error, the test passed - the process either:
-          // 1. Showed the goodbye message (caught above)
-          // 2. Exited cleanly before we could capture the message
+          // Verify CLI showed the goodbye message (graceful exit indicator)
+          const text = await session.text()
+          expect(text.toLowerCase()).toMatch(/goodbye|exiting|continue this session/)
         } finally {
           session.close()
         }
@@ -349,7 +351,7 @@ describe('CLI UI Tests', () => {
 
   describe('slash commands', () => {
     test(
-      'typing / triggers autocomplete menu',
+      'typing / triggers autocomplete menu with command suggestions',
       async () => {
         const session = await launchCLI({ args: [] })
 
@@ -360,14 +362,13 @@ describe('CLI UI Tests', () => {
           // Type a slash to trigger command suggestions
           await session.type('/')
 
-          // Wait for autocomplete to show - it should display a list with "/" prefix
-          // The autocomplete shows command names, so we look for the slash in input
-          // plus any command-like pattern in the suggestions
-          await session.waitForText('/', { timeout: 5000 })
+          // Wait for autocomplete to show command suggestions
+          // The autocomplete should display actual command names like new, exit, usage
+          await session.waitForText(/new|exit|usage|init|logout/i, { timeout: 5000 })
 
           const text = await session.text()
-          // Verify the slash was typed and CLI is responsive
-          expect(text).toContain('/')
+          // Verify autocomplete shows at least one command name
+          expect(text.toLowerCase()).toMatch(/new|exit|usage|init|logout/)
         } finally {
           await session.press(['ctrl', 'c'])
           session.close()

cli/src/__tests__/e2e/full-stack.test.ts

@@ -181,7 +181,7 @@ describe('E2E: Slash Commands', () => {
   )
 
   test(
-    'typing / displays autocomplete with slash in input',
+    'typing / displays autocomplete with command suggestions',
     async () => {
       const session = await ctx.createSession()
 
@@ -190,11 +190,13 @@ describe('E2E: Slash Commands', () => {
 
       // Type / to trigger suggestions
       await session.cli.type('/')
-      await session.cli.waitForText('/', { timeout: 5000 })
+
+      // Wait for autocomplete to show command names
+      await session.cli.waitForText(/new|exit|usage|init|logout/i, { timeout: 5000 })
 
       const text = await session.cli.text()
-      // Verify the slash appears in the input
-      expect(text).toContain('/')
+      // Verify autocomplete shows at least one command name
+      expect(text.toLowerCase()).toMatch(/new|exit|usage|init|logout/)
     },
     TIMEOUT_MS,
   )
@@ -230,7 +232,7 @@ describe('E2E: User Authentication', () => {
   )
 
   test(
-    '/logout command is processed by CLI',
+    '/logout command triggers logout flow',
     async () => {
       const session = await ctx.createSession(E2E_TEST_USERS.default)
 
@@ -242,15 +244,12 @@ describe('E2E: User Authentication', () => {
       await session.cli.waitForText('/logout', { timeout: 5000 })
       await session.cli.press('enter')
 
-      // Wait for the CLI to process the command - the UI should change
-      // Give the command time to execute
-      await sleep(2000)
+      // Wait for logout confirmation or login prompt to appear
+      await session.cli.waitForText(/logged out|log out|login|sign in|goodbye/i, { timeout: 10000 })
 
-      const textAfter = await session.cli.text()
-      // The command should have been processed (UI changed from before)
-      // We can't guarantee specific output text since /logout behavior may vary
-      // but we verify the command was accepted (didn't error or crash)
-      expect(textAfter.length).toBeGreaterThan(0)
+      const text = await session.cli.text()
+      // Verify logout-related UI appeared
+      expect(text.toLowerCase()).toMatch(/logged out|log out|login|sign in|goodbye/)
     },
     TIMEOUT_MS,
   )
@@ -444,7 +443,7 @@ describe('E2E: Additional Slash Commands', () => {
   )
 
   test(
-    '/exit command is processed by CLI',
+    '/exit command triggers graceful exit',
     async () => {
       const session = await ctx.createSession()
 
@@ -456,13 +455,12 @@ describe('E2E: Additional Slash Commands', () => {
       await session.cli.waitForText('/exit', { timeout: 5000 })
       await session.cli.press('enter')
 
-      // Wait for the CLI to process the command
-      await sleep(2000)
+      // Wait for exit/goodbye message to appear
+      await session.cli.waitForText(/goodbye|exiting|continue this session/i, { timeout: 10000 })
 
       const text = await session.cli.text()
-      // /exit should either show goodbye/exit message or the CLI should terminate
-      // Either outcome is valid - we verify the command was accepted
-      expect(text.length).toBeGreaterThan(0)
+      // Verify exit message appeared
+      expect(text.toLowerCase()).toMatch(/goodbye|exiting|continue this session/)
     },
     TIMEOUT_MS,
   )
@@ -516,18 +514,19 @@ describe('E2E: CLI Flags', () => {
   )
 
   test(
-    '--agent flag starts CLI with specified agent',
+    '--agent flag starts CLI with specified agent visible in UI',
     async () => {
       const session = await ctx.createSession(E2E_TEST_USERS.default, [
         '--agent',
         'ask',
       ])
 
-      // CLI should start successfully and show main interface
-      await session.cli.waitForText(/directory/i, { timeout: 15000 })
+      // CLI should show the agent name in the UI
+      await session.cli.waitForText(/ask/i, { timeout: 15000 })
 
       const text = await session.cli.text()
-      expect(text.toLowerCase()).toContain('directory')
+      // Verify the agent name appears in the UI (mode indicator shows agent)
+      expect(text.toLowerCase()).toContain('ask')
     },
     TIMEOUT_MS,
   )

cli/src/__tests__/e2e/test-cli-utils.ts

@@ -177,6 +177,9 @@ export interface E2ETestContext {
   cleanup: () => Promise<void>
 }
 
+// Track if global cleanup has already run in this process
+let globalCleanupRan = false
+
 /**
  * Create a full e2e test context with database, server, and CLI utilities
  */
@@ -189,9 +192,14 @@ export async function createE2ETestContext(describeId: string): Promise<E2ETestC
   } = await import('./test-db-utils')
   const { startE2EServer, stopE2EServer, cleanupOrphanedServers } = await import('./test-server-utils')
 
-  // Clean up any leftovers from previous runs (important on CI retries)
-  cleanupOrphanedContainers()
-  cleanupOrphanedServers()
+  // Only run global cleanup once per process to avoid killing sibling test contexts
+  // This cleanup is for leftover containers/servers from crashed previous runs,
+  // not for cleaning up between parallel describe blocks in the same run
+  if (!globalCleanupRan) {
+    globalCleanupRan = true
+    cleanupOrphanedContainers()
+    cleanupOrphanedServers()
+  }
 
   // Start database
   const db = await createE2EDatabase(describeId)