fix(gateway): prevent duplicate streamed tool calls

.changeset/restore-failed-question-replies.md

@@ -1,5 +1,7 @@
 ---
 "kilo-code": patch
+"@kilocode/cli": patch
+"@kilocode/kilo-gateway": patch
 ---
 
-Dismiss answered question prompts immediately and restore them when submission fails.
+Prevent streamed tool calls from executing twice and leaving answered questions disabled in VS Code.

bun.lock

@@ -171,7 +171,7 @@
         "@clack/prompts": "1.0.0-alpha.1",
         "@kilocode/plugin": "workspace:*",
         "@kilocode/sdk": "workspace:*",
-        "@openrouter/ai-sdk-provider": "2.8.1",
+        "@openrouter/ai-sdk-provider": "2.9.0",
         "ai": "catalog:",
         "open": "10.1.2",
         "zod": "catalog:",
@@ -4649,8 +4649,6 @@
 
     "@kilocode/kilo-docs/typescript": ["typescript@5.9.3", "", { "bin": { "tsc": "bin/tsc", "tsserver": "bin/tsserver" } }, "sha512-jl1vZzPDinLr9eUt3J/t7V6FgNEw9QjvBPdysz9KfQDD41fQrC2Y4vKQdiaUpFT4bXlb1RHhLpp8wtm6M5TgSw=="],
 
-    "@kilocode/kilo-gateway/@openrouter/ai-sdk-provider": ["@openrouter/ai-sdk-provider@2.8.1", "", { "peerDependencies": { "ai": "^6.0.0", "zod": "^3.25.0 || ^4.0.0" } }, "sha512-Y6j3yivgoEUf/kutD/k5GX/mzZfioRFoSx0gbQ+mIOzMaH/vJv1rCkztiuvlLw5xRYQil7oxHUZvmSfXqOx1NQ=="],
-
     "@kilocode/kilo-indexing/glob": ["glob@13.0.6", "", { "dependencies": { "minimatch": "^10.2.2", "minipass": "^7.1.3", "path-scurry": "^2.0.2" } }, "sha512-Wjlyrolmm8uDpm/ogGyXZXb1Z+Ca2B8NbJwqBVg0axK9GbBeoS7yGV6vjXnYdGm6X53iehEuxxbyiKp8QmN4Vw=="],
 
     "@manypkg/find-root/find-up": ["find-up@4.1.0", "", { "dependencies": { "locate-path": "^5.0.0", "path-exists": "^4.0.0" } }, "sha512-PpOwAdQ/YlXQ2vj8a3h8IipDuYRi3wceVQQGYWxNINccq40Anw7BlsEXCMbt1Zt+OLA6Fq9suIpIWD0OsnISlw=="],

packages/kilo-gateway/package.json

@@ -39,7 +39,7 @@
     "@ai-sdk/openai": "3.0.53",
     "@ai-sdk/openai-compatible": "2.0.41",
     "@ai-sdk/mistral": "3.0.27",
-    "@openrouter/ai-sdk-provider": "2.8.1",
+    "@openrouter/ai-sdk-provider": "2.9.0",
     "@clack/prompts": "1.0.0-alpha.1",
     "ai": "catalog:",
     "open": "10.1.2",

packages/kilo-gateway/test/provider.test.ts

@@ -1,5 +1,7 @@
 import { describe, expect, test } from "bun:test"
-import { buildRequestHeaders } from "../src/provider"
+import { streamText, tool } from "ai"
+import { z } from "zod"
+import { buildRequestHeaders, createKilo } from "../src/provider"
 
 describe("Kilo provider request headers", () => {
   test("request headers override provider defaults", () => {
@@ -21,3 +23,89 @@ describe("Kilo provider request headers", () => {
     expect(headers.get("x-request-only")).toBe("kept-too")
   })
 })
+
+describe("Kilo provider tool streaming", () => {
+  test("executes a tool once when complete arguments are followed by whitespace", async () => {
+    const chunks = [
+      {
+        choices: [
+          {
+            index: 0,
+            delta: {
+              role: "assistant",
+              content: null,
+              tool_calls: [
+                {
+                  index: 0,
+                  id: "call_question",
+                  type: "function",
+                  function: { name: "question", arguments: '{"answer":"lasagna"}' },
+                },
+              ],
+            },
+            logprobs: null,
+            finish_reason: null,
+          },
+        ],
+      },
+      {
+        choices: [
+          {
+            index: 0,
+            delta: {
+              tool_calls: [{ index: 0, function: { arguments: " " } }],
+            },
+            logprobs: null,
+            finish_reason: null,
+          },
+        ],
+      },
+      {
+        choices: [{ index: 0, delta: {}, logprobs: null, finish_reason: "tool_calls" }],
+      },
+      {
+        choices: [],
+        usage: { prompt_tokens: 10, completion_tokens: 20, total_tokens: 30 },
+      },
+    ].map((chunk) =>
+      JSON.stringify({
+        id: "chatcmpl-question",
+        object: "chat.completion.chunk",
+        created: 1711357598,
+        model: "qwen/qwen3.6-plus",
+        ...chunk,
+      }),
+    )
+    const body = [...chunks.map((chunk) => `data: ${chunk}\n\n`), "data: [DONE]\n\n"].join("")
+    const fetcher = async () =>
+      new Response(body, {
+        status: 200,
+        headers: { "content-type": "text/event-stream" },
+      })
+    const provider = createKilo({
+      apiKey: "test",
+      baseURL: "https://gateway.test/api/openrouter/",
+      fetch: fetcher as typeof fetch,
+    })
+    const calls: Array<{ answer: string }> = []
+    const result = streamText({
+      model: provider.languageModel("qwen/qwen3.6-plus"),
+      prompt: "Ask a question",
+      tools: {
+        question: tool({
+          description: "Ask the user a question",
+          inputSchema: z.object({ answer: z.string() }),
+          execute: async (input) => {
+            calls.push(input)
+            return input.answer
+          },
+        }),
+      },
+    })
+    const events = []
+    for await (const event of result.fullStream) events.push(event)
+
+    expect(events.filter((event) => event.type === "tool-call")).toHaveLength(1)
+    expect(calls).toEqual([{ answer: "lasagna" }])
+  })
+})

packages/kilo-vscode/src/kilo-provider/handlers/question.ts

@@ -71,7 +71,6 @@ export async function handleQuestionReply(
       { requestID, answers, directory: ctx.getWorkspaceDirectory(sid) },
       { throwOnError: true },
     )
-    ctx.postMessage({ type: "questionResolved", requestID })
     return true
   } catch (error) {
     console.error("[Kilo New] KiloProvider: Failed to reply to question:", error)
@@ -95,7 +94,6 @@ export async function handleQuestionReject(
 
   try {
     await ctx.client.question.reject({ requestID, directory: ctx.getWorkspaceDirectory(sid) }, { throwOnError: true })
-    ctx.postMessage({ type: "questionResolved", requestID })
     return true
   } catch (error) {
     console.error("[Kilo New] KiloProvider: Failed to reject question:", error)

packages/kilo-vscode/tests/unit/question-dock-contract.test.ts

@@ -11,7 +11,6 @@ import path from "node:path"
 
 const ROOT = path.resolve(import.meta.dir, "../..")
 const FILE = path.join(ROOT, "webview-ui/src/components/chat/QuestionDock.tsx")
-const ASSISTANT = path.join(ROOT, "webview-ui/src/components/chat/AssistantMessage.tsx")
 
 function readFile(filePath: string): string {
   return fs.readFileSync(filePath, "utf-8")
@@ -59,9 +58,4 @@ describe("QuestionDock explicit submit contract", () => {
   it("keeps the footer Submit button wired to submit()", () => {
     expect(source).toContain('<Button variant="primary" size="small" onClick={submit} disabled={store.sending}>')
   })
-
-  it("remounts the dock when an inline tool receives a new request", () => {
-    const assistant = readFile(ASSISTANT)
-    expect(assistant).toMatch(/<Show\s+when=\{activeQuestion\(\)\}\s+keyed/)
-  })
 })

packages/kilo-vscode/tests/unit/question-handler.test.ts

@@ -5,7 +5,6 @@ import { handleQuestionReject, handleQuestionReply } from "../../src/kilo-provid
 describe("question handlers", () => {
   it("routes replies using the question session when provided", async () => {
     const calls: Array<Record<string, unknown>> = []
-    const messages: unknown[] = []
     const client = {
       question: {
         reply: async (input: Record<string, unknown>) => {
@@ -20,9 +19,7 @@ describe("question handlers", () => {
       {
         client,
         currentSessionId: "ses-root",
-        postMessage(message) {
-          messages.push(message)
-        },
+        postMessage() {},
         getWorkspaceDirectory(sessionId) {
           return sessionId ? `/repo/${sessionId}` : "/repo"
         },
@@ -40,7 +37,6 @@ describe("question handlers", () => {
         directory: "/repo/ses-worktree",
       },
     ])
-    expect(messages).toEqual([{ type: "questionResolved", requestID: "req-1" }])
   })
 
   it("falls back to the current session when no question session is provided", async () => {
@@ -74,7 +70,6 @@ describe("question handlers", () => {
 
   it("routes rejects using the question session when provided", async () => {
     const calls: Array<Record<string, unknown>> = []
-    const messages: unknown[] = []
     const client = {
       question: {
         reply: async () => true,
@@ -89,9 +84,7 @@ describe("question handlers", () => {
       {
         client,
         currentSessionId: "ses-root",
-        postMessage(message) {
-          messages.push(message)
-        },
+        postMessage() {},
         getWorkspaceDirectory(sessionId) {
           return sessionId ? `/repo/${sessionId}` : "/repo"
         },
@@ -107,6 +100,5 @@ describe("question handlers", () => {
         directory: "/repo/ses-worktree",
       },
     ])
-    expect(messages).toEqual([{ type: "questionResolved", requestID: "req-3" }])
   })
 })

packages/kilo-vscode/tests/unit/question-queue.test.ts

@@ -1,29 +0,0 @@
-import { describe, expect, it } from "bun:test"
-import type { QuestionRequest } from "../../webview-ui/src/types/messages"
-import { removeQuestion, restoreQuestion } from "../../webview-ui/src/context/question-queue"
-
-const first: QuestionRequest = {
-  id: "req-1",
-  sessionID: "ses-1",
-  questions: [],
-}
-
-const second: QuestionRequest = {
-  id: "req-2",
-  sessionID: "ses-1",
-  questions: [],
-}
-
-describe("question queue", () => {
-  it("removes only the question being answered", () => {
-    const result = removeQuestion([first, second], first.id)
-
-    expect(result.question).toBe(first)
-    expect(result.questions).toEqual([second])
-  })
-
-  it("restores a failed question without duplicating it", () => {
-    expect(restoreQuestion([second], first)).toEqual([second, first])
-    expect(restoreQuestion([first, second], first)).toEqual([first, second])
-  })
-})

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

@@ -225,7 +225,6 @@ export const AssistantMessage: Component<AssistantMessageProps> = (props) => {
               <div data-component="tool-part-wrapper" data-part-type={part.type}>
                 <Show
                   when={activeQuestion()}
-                  keyed
                   fallback={
                     <Show
                       when={activeSuggestion()}
@@ -269,7 +268,7 @@ export const AssistantMessage: Component<AssistantMessageProps> = (props) => {
                     </Show>
                   }
                 >
-                  {(req) => <QuestionDock request={req} />}
+                  {(req) => <QuestionDock request={req()} />}
                 </Show>
               </div>
             </Show>

packages/kilo-vscode/webview-ui/src/context/question-queue.ts

@@ -1,15 +0,0 @@
-import type { QuestionRequest } from "../types/messages"
-
-export function removeQuestion(questions: QuestionRequest[], id: string) {
-  const question = questions.find((item) => item.id === id)
-  if (!question) return { question, questions }
-  return {
-    question,
-    questions: questions.filter((item) => item.id !== id),
-  }
-}
-
-export function restoreQuestion(questions: QuestionRequest[], question: QuestionRequest | undefined) {
-  if (!question || questions.some((item) => item.id === question.id)) return questions
-  return [...questions, question]
-}

packages/kilo-vscode/webview-ui/src/context/session.tsx

@@ -71,7 +71,6 @@ import { state as todoState } from "./todo-revert"
 import { getVariant, sessionVariantKeys, transferVariants, variantKey } from "./session-variant-store"
 import { KILO_AUTO, parseModelString } from "../../../src/shared/provider-model"
 import { visibleMessages as filterVisibleMessages } from "./session-queue"
-import { removeQuestion, restoreQuestion } from "./question-queue"
 
 const RECENT_LIMIT = 5
 const MESSAGE_PAGE_LIMIT = 80
@@ -343,7 +342,6 @@ export const SessionProvider: ParentComponent = (props) => {
 
   // Pending questions
   const [questions, setQuestions] = createSignal<QuestionRequest[]>([])
-  const respondingQuestions = new Map<string, QuestionRequest>()
 
   // Tracks question IDs that failed so the UI can reset sending state
   const [questionErrors, setQuestionErrors] = createSignal<Set<string>>(new Set())
@@ -945,7 +943,6 @@ export const SessionProvider: ParentComponent = (props) => {
       case "clearPendingPrompts":
         setPermissions([])
         setQuestions([])
-        respondingQuestions.clear()
         setSuggestions([])
         setRespondingPermissions(new Set<string>())
         setSuggestionErrors(new Set<string>())
@@ -1457,7 +1454,6 @@ export const SessionProvider: ParentComponent = (props) => {
   }
 
   function handleQuestionRequest(question: QuestionRequest) {
-    if (respondingQuestions.has(question.id)) return
     setQuestions((prev) => {
       const idx = prev.findIndex((q) => q.id === question.id)
       if (idx === -1) return [...prev, question]
@@ -1468,7 +1464,6 @@ export const SessionProvider: ParentComponent = (props) => {
   }
 
   function handleQuestionResolved(requestID: string) {
-    respondingQuestions.delete(requestID)
     setQuestions((prev) => prev.filter((q) => q.id !== requestID))
     setQuestionErrors((prev) => {
       const next = new Set(prev)
@@ -1478,9 +1473,6 @@ export const SessionProvider: ParentComponent = (props) => {
   }
 
   function handleQuestionError(requestID: string) {
-    const question = respondingQuestions.get(requestID)
-    respondingQuestions.delete(requestID)
-    setQuestions((prev) => restoreQuestion(prev, question))
     setQuestionErrors((prev) => new Set(prev).add(requestID))
   }
 
@@ -2173,10 +2165,8 @@ export const SessionProvider: ParentComponent = (props) => {
 
   function replyToQuestion(requestID: string, answers: string[][]) {
     clearQuestionError(requestID)
-    const next = removeQuestion(questions(), requestID)
-    const sessionID = next.question?.sessionID ?? currentSessionID() ?? ""
-    if (next.question) respondingQuestions.set(requestID, next.question)
-    setQuestions(next.questions)
+    const question = questions().find((item) => item.id === requestID)
+    const sessionID = question?.sessionID ?? currentSessionID() ?? ""
     vscode.postMessage({
       type: "questionReply",
       requestID,
@@ -2187,10 +2177,8 @@ export const SessionProvider: ParentComponent = (props) => {
 
   function rejectQuestion(requestID: string) {
     clearQuestionError(requestID)
-    const next = removeQuestion(questions(), requestID)
-    const sessionID = next.question?.sessionID ?? currentSessionID() ?? ""
-    if (next.question) respondingQuestions.set(requestID, next.question)
-    setQuestions(next.questions)
+    const question = questions().find((item) => item.id === requestID)
+    const sessionID = question?.sessionID ?? currentSessionID() ?? ""
     vscode.postMessage({
       type: "questionReject",
       requestID,