fix false missing Hermes CLI detection after deployment

tests/server/health-controller.test.ts

@@ -16,6 +16,7 @@ type LoadHealthControllerOptions = {
   bridgeReadinessError?: Error
   managerError?: Error
   runtimeStateError?: Error
+  cliProbe?: ReturnType<typeof vi.fn>
 }
 
 const defaultBridgeReadiness = {
@@ -65,6 +66,12 @@ async function loadHealthController(options: LoadHealthControllerOptions = {}) {
     isDockerContainer: () => options.isDocker === true,
   }))
 
+  const cliProbe = options.cliProbe || vi.fn().mockResolvedValue({ available: true, reason: 'ready' })
+  vi.doMock('../../packages/server/src/services/hermes/hermes-process', () => ({
+    probeHermesCli: cliProbe,
+    resolveHermesBin: () => '/opt/hermes/bin/hermes',
+  }))
+
   const health = await import('../../packages/server/src/controllers/health')
 
   return {
@@ -303,3 +310,43 @@ describe('health controller version metadata', () => {
   })
 
 })
+
+describe('CLI status cache', () => {
+  afterEach(() => { vi.useRealTimers(); vi.resetModules() })
+
+  it('coalesces concurrent requests and allows an explicit recheck after failure', async () => {
+    let finish!: (value: any) => void
+    const probe = vi.fn().mockReturnValueOnce(new Promise(resolve => { finish = resolve }))
+      .mockResolvedValue({ available: true, reason: 'ready' })
+    const { hermesCliStatus } = await loadHealthController({ cliProbe: probe })
+    const first = createMockCtx()
+    const second = createMockCtx()
+    const pending = [hermesCliStatus(first), hermesCliStatus(second)]
+    expect(probe).toHaveBeenCalledOnce()
+    finish({ available: false, reason: 'timeout' })
+    await Promise.all(pending)
+    expect(first.body).toMatchObject({ available: false, reason: 'timeout' })
+    expect(second.body).toEqual(first.body)
+    await hermesCliStatus(createMockCtx())
+    expect(probe).toHaveBeenCalledOnce()
+    const retry = { ...createMockCtx(), query: { refresh: '1' } }
+    await hermesCliStatus(retry)
+    expect(retry.body).toMatchObject({ available: true, reason: 'ready' })
+    expect(probe).toHaveBeenCalledTimes(2)
+  })
+
+  it('expires failures quickly and starts successful cache lifetime after the probe completes', async () => {
+    vi.useFakeTimers()
+    const probe = vi.fn().mockResolvedValueOnce({ available: false, reason: 'permission_denied' })
+      .mockImplementationOnce(() => new Promise(resolve => setTimeout(() => resolve({ available: true, reason: 'ready' }), 8000)))
+    const { hermesCliStatus } = await loadHealthController({ cliProbe: probe })
+    await hermesCliStatus(createMockCtx())
+    await vi.advanceTimersByTimeAsync(5001)
+    const pending = hermesCliStatus(createMockCtx())
+    await vi.advanceTimersByTimeAsync(8000)
+    await pending
+    await vi.advanceTimersByTimeAsync(29000)
+    await hermesCliStatus(createMockCtx())
+    expect(probe).toHaveBeenCalledTimes(2)
+  })
+})