fix(x): surface coding tool errors

apps/x/apps/renderer/src/components/ai-elements/tool.tsx

@@ -22,7 +22,7 @@ import {
 import { type ComponentProps, type ReactNode, isValidElement, useState } from "react";
 import { AnimatePresence, motion } from "motion/react";
 import type { ToolCall, ToolGroup as ToolGroupType } from "@/lib/chat-conversation";
-import { getToolActionsSummary, getToolDisplayName, getToolGroupSummary, toToolState } from "@/lib/chat-conversation";
+import { getToolActionsSummary, getToolDisplayName, getToolErrorText, getToolGroupSummary, toToolState } from "@/lib/chat-conversation";
 
 const formatToolValue = (value: unknown) => {
   if (typeof value === "string") return value;
@@ -329,7 +329,7 @@ export const ToolGroupComponent = ({ group, isToolOpen, onToolOpenChange }: Tool
                   <ToolTabbedContent
                     input={tool.input as ToolUIPart["input"]}
                     output={tool.result as ToolUIPart["output"]}
-                    errorText={tool.status === 'error' ? 'Tool error' : undefined}
+                    errorText={getToolErrorText(tool)}
                   />
                 </ToolContent>
               </Tool>

apps/x/apps/renderer/src/components/chat-sidebar.tsx

@@ -51,6 +51,7 @@ import {
   getWebSearchCardData,
   getComposioConnectCardData,
   getToolDisplayName,
+  getToolErrorText,
   groupConversationItems,
   isChatMessage,
   isErrorMessage,
@@ -484,7 +485,7 @@ export function ChatSidebar({
         )
       }
       const toolTitle = getToolDisplayName(item)
-      const errorText = item.status === 'error' ? 'Tool error' : ''
+      const errorText = getToolErrorText(item)
       const output = normalizeToolOutput(item.result, item.status)
       const input = normalizeToolInput(item.input)
       return (

apps/x/apps/renderer/src/components/code/code-chat.tsx

@@ -10,7 +10,7 @@ import { Conversation, ConversationContent, ConversationScrollButton } from '@/c
 import { MessageResponse } from '@/components/ai-elements/message'
 import { Shimmer } from '@/components/ai-elements/shimmer'
 import { Tool, ToolContent, ToolHeader } from '@/components/ai-elements/tool'
-import { toToolState, getToolDisplayName, getWebSearchCardData, type ToolCall } from '@/lib/chat-conversation'
+import { toToolState, getToolDisplayName, getToolErrorText, getWebSearchCardData, type ToolCall } from '@/lib/chat-conversation'
 import { CodeRunPermissionRequest, CodingRunTimeline } from '@/components/coding-run'
 import { PermissionRequest } from '@/components/ai-elements/permission-request'
 import { AskHumanRequest } from '@/components/ai-elements/ask-human-request'
@@ -84,7 +84,7 @@ function RowboatToolCall({ item, onOpenDiff }: { item: ToolCall; onOpenDiff: (pa
       <Tool open={open || item.status === 'running'} onOpenChange={setOpen}>
         <ToolHeader title={AGENT_LABEL[agent ?? ''] ?? 'Coding agent'} type="tool-code_agent_run" state={toToolState(item.status)} />
         <ToolContent>
-          <CodingRunTimeline events={item.codeRunEvents ?? []} onOpenDiff={onOpenDiff} />
+          <CodingRunTimeline events={item.codeRunEvents ?? []} error={getToolErrorText(item)} onOpenDiff={onOpenDiff} />
         </ToolContent>
       </Tool>
     )

apps/x/apps/renderer/src/components/coding-run.tsx

@@ -1,5 +1,6 @@
 import { useMemo, useState } from 'react'
 import {
+  AlertCircle,
   CheckCircle2,
   Circle,
   CircleDot,
@@ -17,7 +18,7 @@ import {
 import type { CodeRunEvent, PermissionAsk, PermissionDecision } from '@x/shared/src/code-mode.js'
 import { cn } from '@/lib/utils'
 import { Tool, ToolContent, ToolHeader } from '@/components/ai-elements/tool'
-import { toToolState, type ToolCall } from '@/lib/chat-conversation'
+import { getToolErrorText, toToolState, type ToolCall } from '@/lib/chat-conversation'
 
 // ── Timeline reduction ──────────────────────────────────────────────
 // The raw ACP stream is a flat list of events; collapse it into ordered rows,
@@ -111,14 +112,26 @@ const basename = (p: string) => p.split(/[\\/]/).pop() || p
 
 export function CodingRunTimeline({
   events,
+  error,
   onOpenDiff,
 }: {
   events: CodeRunEvent[]
+  error?: string
   // When set, changed-file names become clickable (the Code section opens the diff).
   onOpenDiff?: (path: string) => void
 }) {
   const rows = useMemo(() => reduceEvents(events), [events])
   if (rows.length === 0) {
+    if (error) {
+      return (
+        <div className="px-4 py-3">
+          <div className="flex items-start gap-2 rounded-md border border-destructive/30 bg-destructive/5 px-3 py-2 text-sm text-destructive">
+            <AlertCircle className="mt-0.5 size-3.5 shrink-0" />
+            <span className="min-w-0 whitespace-pre-wrap break-words">{error}</span>
+          </div>
+        </div>
+      )
+    }
     return <div className="px-4 py-3 text-xs text-muted-foreground">Starting the agent…</div>
   }
   return (
@@ -133,12 +146,17 @@ export function CodingRunTimeline({
         }
         if (row.kind === 'tool') {
           const running = row.status !== 'completed' && row.status !== 'failed'
+          const failed = row.status === 'failed'
           return (
             <div key={row.id} className="flex flex-col gap-1">
               <div className="flex items-center gap-2 text-sm">
-                {running
+                {running ? (
-                  ? <Loader className="size-3.5 shrink-0 animate-spin text-muted-foreground" />
+                  <Loader className="size-3.5 shrink-0 animate-spin text-muted-foreground" />
-                  : <CheckCircle2 className="size-3.5 shrink-0 text-green-600" />}
+                ) : failed ? (
+                  <AlertCircle className="size-3.5 shrink-0 text-destructive" />
+                ) : (
+                  <CheckCircle2 className="size-3.5 shrink-0 text-green-600" />
+                )}
                 {toolKindIcon(row.toolKind, row.title)}
                 <span className="truncate text-foreground/90">{row.title ?? row.toolKind ?? 'Tool call'}</span>
               </div>
@@ -189,6 +207,12 @@ export function CodingRunTimeline({
           </div>
         )
       })}
+      {error && (
+        <div className="flex items-start gap-2 rounded-md border border-destructive/30 bg-destructive/5 px-3 py-2 text-sm text-destructive">
+          <AlertCircle className="mt-0.5 size-3.5 shrink-0" />
+          <span className="min-w-0 whitespace-pre-wrap break-words">{error}</span>
+        </div>
+      )}
     </div>
   )
 }
@@ -258,12 +282,13 @@ export function CodingRunBlock({
     (item.result as { agent?: string } | undefined)?.agent ??
     (item.input as { agent?: string } | undefined)?.agent
   const title = AGENT_LABEL[agent ?? ''] ?? 'Coding agent'
+  const error = getToolErrorText(item)
   return (
     <>
       <Tool open={open} onOpenChange={onOpenChange}>
         <ToolHeader title={title} type="tool-code_agent_run" state={toToolState(item.status)} />
         <ToolContent>
-          <CodingRunTimeline events={item.codeRunEvents ?? []} />
+          <CodingRunTimeline events={item.codeRunEvents ?? []} error={error} />
         </ToolContent>
       </Tool>
       {item.pendingCodePermission && (

apps/x/apps/renderer/src/components/compact-conversation.tsx

@@ -7,6 +7,7 @@ import {
     isErrorMessage,
     isToolCall,
     getToolDisplayName,
+    getToolErrorText,
     toToolState,
     normalizeToolOutput,
 } from '@/lib/chat-conversation'
@@ -62,7 +63,7 @@ function CompactToolRow({ tool }: { tool: ToolCall }) {
     const [open, setOpen] = useState(false)
     const title = getToolDisplayName(tool)
     const state = toToolState(tool.status)
-    const errorText = tool.status === 'error' && typeof tool.result === 'string' ? tool.result : undefined
+    const errorText = getToolErrorText(tool)
     return (
         <Tool open={open} onOpenChange={setOpen} className="mb-0 text-xs">
             <ToolHeader title={title} type={`tool-${tool.name}` as `tool-${string}`} state={state} />

apps/x/apps/renderer/src/lib/chat-conversation.ts

@@ -121,6 +121,19 @@ export const normalizeToolOutput = (
   return output
 }
 
+export const getToolErrorText = (tool: ToolCall): string | undefined => {
+  if (tool.status !== 'error') return undefined
+  if (typeof tool.result === 'string' && tool.result.trim()) return tool.result
+  if (tool.result !== undefined) {
+    try {
+      return JSON.stringify(tool.result, null, 2)
+    } catch {
+      // Fall through to the generic label for non-serializable legacy values.
+    }
+  }
+  return 'Tool error'
+}
+
 export type WebSearchCardResult = { title: string; url: string; description: string }
 
 export type WebSearchCardData = {

apps/x/apps/renderer/src/lib/session-chat/turn-view.test.ts

@@ -190,6 +190,35 @@ describe('buildTurnConversation', () => {
     expect(items.map((i) => i.status)).toEqual(['pending', 'running'])
   })
 
+  it('uses the tool result envelope to determine error status', () => {
+    const state = reduceTurn([
+      created(T1, S1),
+      requested(T1, 0),
+      completed(
+        T1,
+        0,
+        assistantCalls(
+          toolCallPart('flagged', 'echo'),
+          toolCallPart('normal', 'echo'),
+        ),
+      ),
+      invocation(T1, 'flagged', 'echo'),
+      {
+        type: 'tool_result',
+        turnId: T1,
+        ts: TS,
+        toolCallId: 'flagged',
+        toolName: 'echo',
+        source: 'sync',
+        result: { output: 'boom', isError: true },
+      },
+      invocation(T1, 'normal', 'echo'),
+      toolResult(T1, 'normal', 'echo', { success: false, message: 'payload data' }),
+    ])
+    const items = buildTurnConversation(state).filter(isToolCall)
+    expect(items.map((i) => i.status)).toEqual(['error', 'completed'])
+  })
+
   it('renders user attachments and a failed turn as an error item', () => {
     const input = {
       role: 'user' as const,

apps/x/apps/renderer/src/lib/session-chat/turn-view.ts

@@ -200,7 +200,7 @@ function extractAttachments(content: UserContent): MessageAttachment[] | undefin
 }
 
 function toolStatus(tc: ToolCallState): ToolCall['status'] {
-  if (tc.result) return 'completed'
+  if (tc.result) return tc.result.result.isError ? 'error' : 'completed'
   if (tc.permission && !tc.permission.resolved) return 'pending'
   return 'running'
 }

apps/x/packages/core/src/application/assistant/skills/code-with-agents/skill.ts

@@ -79,7 +79,7 @@ After \`code_agent_run\` returns:
 - Pass through the agent's \`summary\` as-is. Do not rewrite it.
 - Refer to file paths as plain text. Do NOT use \`\`\`file:path\`\`\` reference blocks. (This overrides the global "always wrap paths in filepath blocks" rule — for code-mode output, plain text.)
 - Only add your own explanation if it failed:
-  - \`success: false\` with a message — surface the message. If it mentions the agent isn't installed or signed in, tell the user to install or sign in via **Settings → Code Mode**.
+  - A tool error with a message — surface the message. If it mentions the agent isn't installed or signed in, tell the user to install or sign in via **Settings → Code Mode**.
   - \`stopReason: "cancelled"\` — the run was stopped; acknowledge briefly and ask if they want to continue.
 
 ---

apps/x/packages/core/src/application/lib/builtin-tools.test.ts

@@ -0,0 +1,54 @@
+import { afterEach, describe, expect, it, vi } from "vitest";
+import container from "../../di/container.js";
+import { InMemoryAbortRegistry } from "../../runs/abort-registry.js";
+import { BuiltinTools } from "./builtin-tools.js";
+import type { ToolContext } from "./exec-tool.js";
+
+function context(signal: AbortSignal): ToolContext {
+    return {
+        runId: "turn-1",
+        toolCallId: "tool-1",
+        signal,
+        abortRegistry: new InMemoryAbortRegistry(),
+        publish: async () => {},
+        codePolicy: "ask",
+    };
+}
+
+function mockCodeServices(runPrompt: () => Promise<never>): void {
+    vi.spyOn(container, "resolve").mockImplementation(((name: string) => {
+        if (name === "codeModeManager") return { runPrompt };
+        if (name === "codePermissionRegistry") {
+            return { cancelRun: vi.fn(), request: vi.fn() };
+        }
+        throw new Error(`Unexpected dependency: ${name}`);
+    }) as typeof container.resolve);
+}
+
+describe("code_agent_run", () => {
+    afterEach(() => vi.restoreAllMocks());
+
+    it("throws genuine coding-agent failures for the runtime to mark as errors", async () => {
+        mockCodeServices(async () => {
+            throw new Error("spawn Electron ENOENT");
+        });
+
+        await expect(BuiltinTools.code_agent_run.execute(
+            { agent: "codex", cwd: "/repo", prompt: "Fix it" },
+            context(new AbortController().signal),
+        )).rejects.toThrow("Coding agent failed: spawn Electron ENOENT");
+    });
+
+    it("returns an ordinary cancellation result when the turn was aborted", async () => {
+        const controller = new AbortController();
+        controller.abort();
+        mockCodeServices(async () => {
+            throw new Error("cancelled");
+        });
+
+        await expect(BuiltinTools.code_agent_run.execute(
+            { agent: "codex", cwd: "/repo", prompt: "Fix it" },
+            context(controller.signal),
+        )).resolves.toMatchObject({ success: false, stopReason: "cancelled" });
+    });
+});

apps/x/packages/core/src/application/lib/builtin-tools.ts

@@ -868,7 +868,7 @@ export const BuiltinTools: z.infer<typeof BuiltinToolsSchema> = {
         }),
         execute: async ({ agent, cwd, prompt }: { agent: 'claude' | 'codex', cwd: string, prompt: string }, ctx?: ToolContext) => {
             if (!ctx) {
-                return { success: false, message: 'code_agent_run requires run context (runId / streaming).' };
+                throw new Error('code_agent_run requires run context (runId / streaming).');
             }
             // The composer chip is the source of truth for the agent. The model's `agent`
             // argument is only a fallback for the ask-human flow (code mode not active, no
@@ -955,10 +955,7 @@ export const BuiltinTools: z.infer<typeof BuiltinToolsSchema> = {
                         changedFiles: [...changedFiles],
                     };
                 }
-                return {
+                throw new Error(`Coding agent failed: ${error instanceof Error ? error.message : String(error)}`);
-                    success: false,
-                    message: `Coding agent failed: ${error instanceof Error ? error.message : String(error)}`,
-                };
             } finally {
                 ctx.signal.removeEventListener('abort', onAbort);
             }