fix(vscode): surface backend crash recovery

.changeset/calm-kilos-reconnect.md

@@ -0,0 +1,5 @@
+---
+"kilo-code": patch
+---
+
+Show a retryable connection error and preserve unsent prompts when the VS Code background CLI process exits.

packages/kilo-vscode/src/KiloProvider.ts

@@ -347,6 +347,16 @@ export class KiloProvider implements vscode.WebviewViewProvider, TelemetryProper
     }
   }
 
+  private postConnectionState(error = this.connectionService.getConnectionError()): void {
+    this.postMessage({
+      type: "connectionState",
+      state: this.connectionState,
+      ...(this.connectionState === "error" && {
+        error: getErrorMessage(error) || "Connection to CLI backend lost. Retry to reconnect.",
+      }),
+    })
+  }
+
   // Strip edit-tool metadata.filediff.before/after (multi-MB for edit-heavy
   // sessions) to keep session switches fast. Logic in kilo-provider/slim-metadata.ts.
   private slimPart<T>(part: T): T {
@@ -386,7 +396,7 @@ export class KiloProvider implements vscode.WebviewViewProvider, TelemetryProper
     }
 
     // Always push connection state first so the UI can render appropriately.
-    this.postMessage({ type: "connectionState", state: this.connectionState })
+    this.postConnectionState()
     pushTelemetryState((m) => this.postMessage(m))
 
     // Re-send ready so the webview can recover after refresh.
@@ -1228,9 +1238,9 @@ export class KiloProvider implements vscode.WebviewViewProvider, TelemetryProper
       )
 
       // Subscribe to connection state changes
-      this.unsubscribeState = this.connectionService.onStateChange(async (state) => {
+      this.unsubscribeState = this.connectionService.onStateChange(async (state, error) => {
         this.connectionState = state
-        this.postMessage({ type: "connectionState", state })
+        this.postConnectionState(error)
 
         if (state === "connected") {
           // Fire config warnings independently so a failure in the
@@ -1309,7 +1319,7 @@ export class KiloProvider implements vscode.WebviewViewProvider, TelemetryProper
           workspaceDirectory: this.getProjectDirectory(this.currentSession?.id),
         })
       }
-      this.postMessage({ type: "connectionState", state: this.connectionState })
+      this.postConnectionState()
 
       // connect() can resolve after SSE reaches "connected" but before this
       // provider subscribes to onStateChange(). In that case the initial

packages/kilo-vscode/src/services/cli-backend/connection-service.ts

@@ -7,7 +7,7 @@ import { resolveEventSessionId as resolveEventSessionIdPure } from "./connection
 
 export type ConnectionState = "connecting" | "connected" | "disconnected" | "error"
 type SSEEventListener = (event: Event, directory?: string) => void
-type StateListener = (state: ConnectionState) => void
+type StateListener = (state: ConnectionState, error?: Error) => void
 type SSEEventFilter = (event: Event, directory?: string) => boolean
 type NotificationDismissListener = (notificationId: string) => void
 type LanguageChangeListener = (locale: string) => void
@@ -65,6 +65,7 @@ export class KiloConnectionService {
   private info: { port: number } | null = null
   private config: ServerConfig | null = null
   private state: ConnectionState = "disconnected"
+  private error: Error | null = null
   private connectPromise: Promise<void> | null = null
   private healthPollTimer: ReturnType<typeof setInterval> | null = null
   private remoteService: import("../RemoteStatusService").RemoteStatusService | null = null
@@ -93,7 +94,7 @@ export class KiloConnectionService {
   private unsubRemote: (() => void) | null = null
 
   constructor(context: vscode.ExtensionContext) {
-    this.serverManager = new ServerManager(context)
+    this.serverManager = new ServerManager(context, (code) => this.handleServerExit(code))
   }
 
   /**
@@ -115,7 +116,7 @@ export class KiloConnectionService {
       await this.connectPromise
     } catch (error) {
       // If doConnect() fails before SSE can emit a state transition, avoid leaving consumers stuck in "connecting".
-      this.setState("error")
+      this.setState("error", this.error ?? (error instanceof Error ? error : new Error(String(error))))
       throw error
     } finally {
       this.connectPromise = null
@@ -126,7 +127,7 @@ export class KiloConnectionService {
    * Get the shared SDK client. Throws if not connected.
    */
   getClient(): KiloClient {
-    if (!this.client) {
+    if (!this.client || this.state !== "connected") {
       throw new Error("Not connected — call connect() first")
     }
     return this.client
@@ -139,11 +140,11 @@ export class KiloConnectionService {
    * or if the connection fails.
    */
   async getClientAsync(dir?: string): Promise<KiloClient> {
-    if (this.client) return this.client
+    if (this.client && this.state === "connected") return this.client
     const root = dir ?? vscode.workspace.workspaceFolders?.[0]?.uri.fsPath
     if (!root) throw new Error("No workspace folder open")
     await this.connect(root)
-    return this.client!
+    return this.getClient()
   }
 
   /**
@@ -189,6 +190,13 @@ export class KiloConnectionService {
     return this.state
   }
 
+  /**
+   * Last connection error. Cleared when a new connection attempt begins.
+   */
+  getConnectionError(): Error | null {
+    return this.error
+  }
+
   /**
    * Subscribe to SSE events. Returns unsubscribe function.
    */
@@ -503,12 +511,14 @@ export class KiloConnectionService {
     this.config = null
     this.info = null
     this.state = "disconnected"
+    this.error = null
   }
 
-  private setState(state: ConnectionState): void {
+  private setState(state: ConnectionState, error?: Error): void {
     this.state = state
+    this.error = state === "error" ? (error ?? this.error) : null
     for (const listener of this.stateListeners) {
-      listener(state)
+      listener(state, this.error ?? undefined)
     }
   }
 
@@ -558,10 +568,28 @@ export class KiloConnectionService {
     }
   }
 
-  private async doConnect(workspaceDir: string): Promise<void> {
-    // If we reconnect, ensure the previous SSE connection is cleaned up first.
+  private resetConnection(): void {
     this.stopHealthPoll()
-    this.sseClient?.dispose()
+    const sse = this.sseClient
+    this.sseClient = null
+    sse?.disconnect()
+    this.client = null
+    this.config = null
+    this.info = null
+  }
+
+  private handleServerExit(code: number | null): void {
+    console.warn("[Kilo New] ConnectionService: CLI background process exited:", code)
+    this.resetConnection()
+    this.setState(
+      "error",
+      new Error(`CLI background process exited with code ${code ?? "unknown"}. Retry to reconnect.`),
+    )
+  }
+
+  private async doConnect(workspaceDir: string): Promise<void> {
+    // Never expose a stale SDK client while its replacement server is starting.
+    this.resetConnection()
 
     const server = await this.serverManager.getServer()
     this.info = { port: server.port }
@@ -575,14 +603,15 @@ export class KiloConnectionService {
 
     // Create SDK client with Basic Auth header
     const authHeader = `Basic ${Buffer.from(`kilo:${server.password}`).toString("base64")}`
-    this.client = createKiloClient({
+    const client = createKiloClient({
       baseUrl: config.baseUrl,
       headers: {
         Authorization: authHeader,
       },
     })
-
-    this.sseClient = new SdkSSEAdapter(this.client)
+    const sse = new SdkSSEAdapter(client)
+    this.client = client
+    this.sseClient = sse
 
     // Wait until SSE yields its first server event before resolving connect().
     // Initial stream failures are handled by the adapter reconnect loop.
@@ -596,18 +625,29 @@ export class KiloConnectionService {
     let didConnect = false
 
     // Wire SSE events → broadcast to all registered listeners
-    this.sseClient.onEvent((event, directory) => {
+    sse.onEvent((event, directory) => {
+      if (this.sseClient !== sse) return
       for (const listener of this.eventListeners) {
         listener(event, directory)
       }
     })
 
-    this.sseClient.onError(() => {
-      this.setState("error")
+    sse.onError((error) => {
+      if (this.sseClient !== sse) return
+      this.setState("error", error)
     })
 
     // Wire SSE state → broadcast to all registered state listeners
-    this.sseClient.onStateChange((sseState) => {
+    sse.onStateChange((sseState) => {
+      if (this.sseClient !== sse) {
+        if (!didConnect && sseState === "disconnected") {
+          rejectConnected?.(new Error(`SSE connection ended in state: ${sseState}`))
+          resolveConnected = null
+          rejectConnected = null
+        }
+        return
+      }
+
       this.setState(sseState)
 
       if (sseState === "connected") {
@@ -625,7 +665,7 @@ export class KiloConnectionService {
       }
     })
 
-    this.sseClient.connect()
+    sse.connect()
 
     await connectedPromise
 

packages/kilo-vscode/src/services/cli-backend/server-manager.ts

@@ -17,6 +17,7 @@ export interface ServerInstance {
 const STARTUP_TIMEOUT_SECONDS = 30
 
 type WorkspaceFolderLike = { uri: { fsPath: string } }
+type ServerExitListener = (code: number | null) => void
 
 export function resolveServerCwd(folders: readonly WorkspaceFolderLike[] | undefined, storage: string): string {
   return folders?.[0]?.uri.fsPath ?? storage
@@ -31,7 +32,10 @@ export class ServerManager {
   private instance: ServerInstance | null = null
   private startupPromise: Promise<ServerInstance> | null = null
 
-  constructor(private readonly context: vscode.ExtensionContext) {}
+  constructor(
+    private readonly context: vscode.ExtensionContext,
+    private readonly onExit?: ServerExitListener,
+  ) {}
 
   /**
    * Get or start the server instance
@@ -171,6 +175,7 @@ export class ServerManager {
         console.log("[Kilo New] ServerManager: 🛑 Process exited with code:", code)
         if (this.instance?.process === serverProcess) {
           this.instance = null
+          this.onExit?.(code)
         }
         if (!resolved) {
           const { userMessage, userDetails } = toErrorMessage(

packages/kilo-vscode/tests/unit/kilo-provider-followup.test.ts

@@ -69,6 +69,7 @@ function connection() {
     getServerInfo: () => ({ port: 12345 }),
     getServerConfig: () => ({ baseUrl: "http://127.0.0.1:12345", password: "test" }),
     getConnectionState: () => "connected" as const,
+    getConnectionError: () => null,
     resolveEventSessionId: (event: Event) => (event.type === "session.created" ? event.properties.info.id : undefined),
     recordMessageSessionId: () => undefined,
     notifyNotificationDismissed: () => undefined,

packages/kilo-vscode/tests/unit/kilo-provider-load-messages.test.ts

@@ -98,6 +98,7 @@ function createConnection(client: ReturnType<typeof createClient>) {
     registerDirectoryProvider: () => () => undefined,
     getServerInfo: () => ({ port: 12345 }),
     getConnectionState: () => "connected" as const,
+    getConnectionError: () => null,
     resolveEventSessionId: () => undefined,
     recordMessageSessionId: () => undefined,
     notifyNotificationDismissed: () => undefined,

packages/kilo-vscode/tests/unit/kilo-provider-session-refresh.test.ts

@@ -91,6 +91,7 @@ function createConnection(client: ReturnType<typeof createClient>) {
     getServerInfo: () => ({ port: 12345 }),
     getServerConfig: () => ({ baseUrl: "http://127.0.0.1:12345", password: "test" }),
     getConnectionState: () => "connected" as const,
+    getConnectionError: () => null,
     resolveEventSessionId: () => undefined,
     recordMessageSessionId: () => undefined,
     notifyNotificationDismissed: () => undefined,

packages/kilo-vscode/tests/unit/prompt-input-connection-guard.test.ts

@@ -0,0 +1,20 @@
+import { describe, expect, it } from "bun:test"
+import { readFileSync } from "node:fs"
+import { join } from "node:path"
+
+describe("PromptInput connection guard", () => {
+  const path = join(__dirname, "..", "..", "webview-ui", "src", "components", "chat", "PromptInput.tsx")
+  const src = readFileSync(path, "utf8")
+
+  it("rechecks the connection after resolving async attachments and before clearing the draft", () => {
+    const attachments = src.indexOf("const gitFile = await git.resolveAttachment")
+    const guard = src.indexOf("if (isDisabled()) return", attachments)
+    const send = src.indexOf("session.sendMessage(message", guard)
+    const clear = src.indexOf("drafts.delete(key)", send)
+
+    expect(attachments).toBeGreaterThan(-1)
+    expect(guard).toBeGreaterThan(attachments)
+    expect(send).toBeGreaterThan(guard)
+    expect(clear).toBeGreaterThan(send)
+  })
+})

packages/kilo-vscode/tests/unit/sdk-sse-adapter.test.ts

@@ -110,6 +110,38 @@ describe("SdkSSEAdapter", () => {
   })
 })
 
+describe("KiloConnectionService backend crash", () => {
+  it("invalidates the stale SDK client and reports a retryable error", () => {
+    const service = new KiloConnectionService({} as any)
+    const states: Array<{ state: string; error?: string }> = []
+    ;(service as any).client = {}
+    ;(service as any).config = { baseUrl: "http://127.0.0.1:52512", password: "secret" }
+    ;(service as any).info = { port: 52512 }
+    ;(service as any).state = "connected"
+    service.onStateChange((state, error) => states.push({ state, error: error?.message }))
+    ;(service as any).handleServerExit(9)
+
+    expect(service.getConnectionState()).toBe("error")
+    expect(service.getConnectionError()?.message).toContain("CLI background process exited with code 9")
+    expect(service.getServerConfig()).toBeNull()
+    expect(service.getServerInfo()).toBeNull()
+    expect(() => service.getClient()).toThrow("Not connected")
+    expect(states).toEqual([
+      { state: "error", error: "CLI background process exited with code 9. Retry to reconnect." },
+    ])
+    service.dispose()
+  })
+
+  it("does not expose an SDK client while a replacement server is connecting", () => {
+    const service = new KiloConnectionService({} as any)
+    ;(service as any).client = {}
+    ;(service as any).state = "connecting"
+
+    expect(() => service.getClient()).toThrow("Not connected")
+    service.dispose()
+  })
+})
+
 describe("KiloConnectionService SSE startup", () => {
   it("waits through an initial SSE fetch failure until the stream opens", async () => {
     const original = globalThis.fetch

packages/kilo-vscode/webview-ui/src/components/chat/PromptInput.tsx

@@ -801,6 +801,7 @@ export const PromptInput: Component<PromptInputProps> = (props) => {
       return undefined
     })
     if (hasGit() && hasGitChangesMention(message) && !gitFile) return
+    if (isDisabled()) return
 
     const allFiles = [
       ...mentionFiles,