fix(server): repair remaining 16 server-checks CI failures

Group F (PR regression): drop the dead sessionGenerationBusy gate in
  scheduleRestartSessionWithRuntimeConfig. It only cleared on a status
  'idle' event that a normal turn never emits, so any runtime restart
  after the first turn was parked forever and the 2nd startSession never
  fired. The new active-turn deferral (deferredRuntimeRestarts, drained by
  the turn's 'result' callback) now correctly guards mid-turn restarts.
  Removes now-dead pendingDeferredRuntimeRestarts / drainPendingRuntimeRestart.
Group A: install ripgrep in the server-checks job so project-scoped
  markdown config discovery (output-styles, slash-commands, agents) works.
Group B: clear inherited NVM_DIR in shell-env tests so the CI runner's real
  /home/runner/.nvm can't shadow the fake .zshrc value under test.
Group C: pin/restore ProviderService.serverPort (process-wide static) in
  conversation-service and cron-scheduler tests so leaked random ports from
  other suites don't break the hardcoded 3456 ANTHROPIC_BASE_URL assertions.
Group D: assert process.platform instead of hardcoded 'darwin' in full-flow.
Group E: assert response shape (key present) not an exact empty servers
  array for /api/mcp + /api/plugins H5 auth checks; they run at repo root
  which has a tracked .mcp.json.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

.github/workflows/pr-quality.yml

@@ -96,6 +96,9 @@ jobs:
         with:
           bun-version: latest
 
+      - name: Install ripgrep
+        run: sudo apt-get update && sudo apt-get install -y ripgrep
+
       - name: Install root dependencies
         run: bun install
 

src/server/__tests__/conversation-service.test.ts

@@ -31,10 +31,19 @@ describe('ConversationService', () => {
   let originalPath: string | undefined
   let originalShell: string | undefined
   let originalZdotdir: string | undefined
+  let originalNvmDir: string | undefined
   let originalDisableTerminalShellEnv: string | undefined
+  let originalServerPort: number | undefined
 
   beforeEach(async () => {
     tmpDir = await fs.mkdtemp(path.join(os.tmpdir(), 'cc-haha-conversation-service-'))
+    // ProviderService.serverPort is a process-wide static; other server tests
+    // (e.g. conversations.test.ts starting a server on an OS-assigned port) can
+    // leave it set to a random value. Pin it to the default so the managed
+    // ANTHROPIC_BASE_URL assertions below are deterministic, and restore it
+    // afterwards.
+    originalServerPort = ProviderService.getServerPort()
+    ProviderService.setServerPort(3456)
     originalConfigDir = process.env.CLAUDE_CONFIG_DIR
     originalApiKey = process.env.ANTHROPIC_API_KEY
     originalAuthToken = process.env.ANTHROPIC_AUTH_TOKEN
@@ -54,6 +63,7 @@ describe('ConversationService', () => {
     originalPath = process.env.PATH
     originalShell = process.env.SHELL
     originalZdotdir = process.env.ZDOTDIR
+    originalNvmDir = process.env.NVM_DIR
     originalDisableTerminalShellEnv = process.env.CC_HAHA_DISABLE_TERMINAL_SHELL_ENV
 
     process.env.CLAUDE_CONFIG_DIR = tmpDir
@@ -74,6 +84,11 @@ describe('ConversationService', () => {
     delete process.env.CC_HAHA_TRACE_PROVIDER_NAME
     delete process.env.CC_HAHA_TRACE_PROVIDER_FORMAT
     process.env.CC_HAHA_DISABLE_TERMINAL_SHELL_ENV = '1'
+    // The CI runner exports a real NVM_DIR (/home/runner/.nvm). buildChildEnv's
+    // shell-env merge lets inherited process env win over shell-captured values,
+    // so a leaked NVM_DIR would shadow the fake .zshrc value the shell-capture
+    // tests assert. Clear it; the per-test fake shell is the only intended source.
+    delete process.env.NVM_DIR
     resetTerminalShellEnvironmentCacheForTests()
   })
 
@@ -135,10 +150,14 @@ describe('ConversationService', () => {
     if (originalZdotdir === undefined) delete process.env.ZDOTDIR
     else process.env.ZDOTDIR = originalZdotdir
 
+    if (originalNvmDir === undefined) delete process.env.NVM_DIR
+    else process.env.NVM_DIR = originalNvmDir
+
     if (originalDisableTerminalShellEnv === undefined) delete process.env.CC_HAHA_DISABLE_TERMINAL_SHELL_ENV
     else process.env.CC_HAHA_DISABLE_TERMINAL_SHELL_ENV = originalDisableTerminalShellEnv
 
     resetTerminalShellEnvironmentCacheForTests()
+    if (originalServerPort !== undefined) ProviderService.setServerPort(originalServerPort)
     await fs.rm(tmpDir, { recursive: true, force: true })
   })
 

src/server/__tests__/cron-scheduler-launcher.test.ts

@@ -22,6 +22,7 @@ const originalClaudeCodeEntrypoint = process.env.CLAUDE_CODE_ENTRYPOINT
 const originalHome = process.env.HOME
 const originalShell = process.env.SHELL
 const originalZdotdir = process.env.ZDOTDIR
+const originalNvmDir = process.env.NVM_DIR
 const originalDisableTerminalShellEnv = process.env.CC_HAHA_DISABLE_TERMINAL_SHELL_ENV
 
 const isWindows = process.platform === 'win32'
@@ -101,6 +102,11 @@ function restoreEnv(): void {
   } else {
     delete process.env.ZDOTDIR
   }
+  if (originalNvmDir) {
+    process.env.NVM_DIR = originalNvmDir
+  } else {
+    delete process.env.NVM_DIR
+  }
   if (originalDisableTerminalShellEnv) {
     process.env.CC_HAHA_DISABLE_TERMINAL_SHELL_ENV = originalDisableTerminalShellEnv
   } else {
@@ -111,16 +117,28 @@ function restoreEnv(): void {
 
 describe('cron scheduler launcher resolution', () => {
   let tmpDir: string
+  let originalServerPort: number | undefined
 
   beforeEach(async () => {
     tmpDir = await createTmpDir()
     process.env.CLAUDE_CONFIG_DIR = path.join(tmpDir, 'config')
     process.env.CC_HAHA_DISABLE_TERMINAL_SHELL_ENV = '1'
+    // The CI runner exports a real NVM_DIR (/home/runner/.nvm); the shell-env
+    // merge lets inherited process env shadow shell-captured values, which would
+    // break the shell-capture assertions. Clear it so the fake shell is the only
+    // source.
+    delete process.env.NVM_DIR
+    // ProviderService.serverPort is a process-wide static other tests may leave
+    // at a random value; pin it so the managed ANTHROPIC_BASE_URL assertions are
+    // deterministic, restore afterwards.
+    originalServerPort = ProviderService.getServerPort()
+    ProviderService.setServerPort(3456)
     resetTerminalShellEnvironmentCacheForTests()
   })
 
   afterEach(async () => {
     restoreEnv()
+    if (originalServerPort !== undefined) ProviderService.setServerPort(originalServerPort)
     await cleanupTmpDir(tmpDir)
   })
 

src/server/__tests__/e2e/full-flow.test.ts

@@ -112,7 +112,7 @@ describe('E2E: Full Flow', () => {
 
   it('should return diagnostics', async () => {
     const { data } = await api('GET', '/api/status/diagnostics')
-    expect(data.platform).toBe('darwin')
+    expect(data.platform).toBe(process.platform)
     expect(data.configDir).toBe(tmpDir)
   })
 

src/server/__tests__/h5-access-auth.test.ts

@@ -191,8 +191,13 @@ function expectWebSocketUpgradeThenClose(url: string): Promise<void> {
 }
 
 const settingsSurfaceEndpoints = [
-  { path: '/api/mcp', expected: { servers: [] } },
-  { path: '/api/plugins', expected: { plugins: [] } },
+  // Assert the response *shape* (key present), not an exact empty array: the
+  // server reads MCP/plugin config from the current working directory, so when
+  // the suite runs at the repo root (which has a tracked .mcp.json) `servers`
+  // is non-empty. These cases exist to verify H5-token auth (401/401/200), not
+  // the discovered config contents.
+  { path: '/api/mcp', expectedKey: 'servers' },
+  { path: '/api/plugins', expectedKey: 'plugins' },
   { path: '/api/agents', expectedKey: 'activeAgents' },
 ] as const
 
@@ -713,11 +718,7 @@ describe('remote H5 auth and CORS integration', () => {
       expect(validTokenResponse.status).toBe(200)
       expect(validTokenResponse.headers.get('Access-Control-Allow-Origin')).toBe(PHONE_ORIGIN)
       const body = await validTokenResponse.json()
-      if ('expected' in endpoint) {
-        expect(body).toMatchObject(endpoint.expected)
-      } else {
-        expect(body).toHaveProperty(endpoint.expectedKey)
-      }
+      expect(body).toHaveProperty(endpoint.expectedKey)
     }
   })
 

src/server/ws/handler.ts

@@ -119,19 +119,6 @@ const prewarmIdleTimers = new Map<string, ReturnType<typeof setTimeout>>()
 const DEFAULT_PREWARM_IDLE_TIMEOUT_MS = 5 * 60_000
 const VALID_EFFORT_LEVELS = new Set(['low', 'medium', 'high', 'max'])
 
-// Track whether a session is currently mid-turn (anything other than 'idle').
-// Updated inside sendMessage by observing outbound `status` events. Used to
-// avoid yanking the CLI subprocess out from under an in-progress generation
-// when the user changes runtime config (model / effort / thinking) mid-stream.
-const sessionGenerationBusy = new Set<string>()
-
-// When a runtime config change arrives while the session is busy, we record
-// the WS that should drive the eventual restart and apply it on the next
-// transition to 'idle'. The actual override values are read from
-// `runtimeOverrides` at restart time, so a later set_runtime_config naturally
-// supersedes an earlier deferred one.
-const pendingDeferredRuntimeRestarts = new Map<string, ServerWebSocket<WebSocketData>>()
-
 async function sendRepositoryStartupStatus(
   ws: ServerWebSocket<WebSocketData>,
   sessionId: string,
@@ -1331,8 +1318,6 @@ function cleanupSessionRuntimeState(sessionId: string) {
   runtimeTransitionPromises.delete(sessionId)
   sessionStartupPromises.delete(sessionId)
   lastResolvedStartupWorkDirs.delete(sessionId)
-  sessionGenerationBusy.delete(sessionId)
-  pendingDeferredRuntimeRestarts.delete(sessionId)
   clearPrewarmState(sessionId)
 }
 
@@ -2133,49 +2118,21 @@ function toApiRetryServerMessage(cliMsg: any): ServerMessage | null {
 }
 
 function sendMessage(ws: ServerWebSocket<WebSocketData>, message: ServerMessage) {
-  // Track per-session generation busy state by observing outbound `status`
-  // events. This lets handleSetRuntimeConfig defer a restart that would
-  // otherwise kill the CLI subprocess in the middle of a streaming response.
-  const sessionId = ws.data.sessionId
-  if (sessionId && message.type === 'status') {
-    if (message.state === 'idle') {
-      const wasBusy = sessionGenerationBusy.delete(sessionId)
-      if (wasBusy) drainPendingRuntimeRestart(sessionId)
-    } else {
-      sessionGenerationBusy.add(sessionId)
-    }
-  }
   ws.send(JSON.stringify(message))
 }
 
-// Apply a pending deferred runtime restart that was queued while the session
-// was busy. Reads the latest override from `runtimeOverrides` at restart time,
-// so multiple toggles during streaming naturally collapse into a single
-// restart with the most recent config.
-function drainPendingRuntimeRestart(sessionId: string) {
-  const ws = pendingDeferredRuntimeRestarts.get(sessionId)
-  if (!ws) return
-  pendingDeferredRuntimeRestarts.delete(sessionId)
-  if (!conversationService.hasSession(sessionId)) return
-  void enqueueRuntimeTransition(sessionId, async () => {
-    if (!conversationService.hasSession(sessionId)) return
-    await restartSessionWithRuntimeConfig(ws, sessionId)
-  })
-}
-
-// Schedule a runtime-config restart, but defer until the session is idle to
-// avoid interrupting an in-progress generation. The override values are
-// already in `runtimeOverrides[sessionId]` (and persisted) before this is
-// called, so getRuntimeSettings will read them when the deferred restart
-// finally runs.
+// Restart the CLI subprocess to apply a runtime-config change. The override
+// values are already in `runtimeOverrides[sessionId]` (and persisted) before
+// this is called, so getRuntimeSettings will read them at restart time.
+//
+// Mid-turn protection is handled upstream by the active-turn deferral
+// (`shouldDeferRuntimeRestartForActiveTurn` + `deferredRuntimeRestarts`, drained
+// by the turn's `result` callback in `bindActiveUserTurnCompletion`), which
+// gates on the real turn lifecycle rather than on outbound status events.
 async function scheduleRestartSessionWithRuntimeConfig(
   ws: ServerWebSocket<WebSocketData>,
   sessionId: string,
 ): Promise<void> {
-  if (sessionGenerationBusy.has(sessionId)) {
-    pendingDeferredRuntimeRestarts.set(sessionId, ws)
-    return
-  }
   await restartSessionWithRuntimeConfig(ws, sessionId)
 }
 

src/utils/mcpStdioEnvironment.test.ts

@@ -51,10 +51,16 @@ describe('MCP stdio environment', () => {
       PATH: process.env.PATH,
       SHELL: process.env.SHELL,
       ZDOTDIR: process.env.ZDOTDIR,
+      NVM_DIR: process.env.NVM_DIR,
       CC_HAHA_DISABLE_TERMINAL_SHELL_ENV:
         process.env.CC_HAHA_DISABLE_TERMINAL_SHELL_ENV,
     }
     delete process.env.CC_HAHA_DISABLE_TERMINAL_SHELL_ENV
+    // The CI runner exports a real NVM_DIR (/home/runner/.nvm). The env merge
+    // lets inherited process env win over shell-captured values, so a leaked
+    // NVM_DIR would shadow the fake .zshrc value asserted below. Clear it so the
+    // only source of NVM_DIR is the fake shell config.
+    delete process.env.NVM_DIR
     resetMcpStdioEnvironmentCacheForTests()
   })