fix(chat): keep tool failures in expanded history

Author: Sg312

apps/sim/app/workspace/[workspaceId]/home/components/message-content/components/agent-group/activity-stream.test.tsx

@@ -121,26 +121,43 @@ describe.each(['mothership', 'workflow', 'browser', 'deploy'])('%s activity', (a
     expect(header()?.textContent).toBe('Reading third')
   })
 
-  it.each(['error', 'cancelled', 'interrupted', 'rejected', 'skipped'] as const)(
+  it.each(['cancelled', 'interrupted', 'skipped'] as const)(
     'shows %s immediately and cancels a pending cosmetic update',
     (status) => {
       render([tool('first')])
       advance(100)
       render([tool('first', 'success'), tool('second')])
       render([tool('first', 'success'), tool('second', status)])
-      const prefix =
-        status === 'error' || status === 'rejected'
-          ? 'Failed'
-          : status === 'skipped'
-            ? 'Skipped'
-            : 'Stopped'
+      const prefix = status === 'skipped' ? 'Skipped' : 'Stopped'
       expect(header()?.textContent).toBe(`${prefix} reading second`)
       expect(container.querySelector('[class*="shimmer"]')).toBeNull()
       advance(1500)
       expect(header()?.textContent).toBe(`${prefix} reading second`)
     }
   )
 
+  it.each(['error', 'rejected'] as const)(
+    'keeps a %s attempt in history without promoting it into the summary',
+    (status) => {
+      render([tool('first')])
+      advance(100)
+      render([tool('first', 'success'), tool('second')])
+      render([tool('first', 'success'), tool('second', status)])
+      const label = agentName === 'mothership' ? 'Read first' : 'Reading first'
+      expect(header()?.textContent).toBe(label)
+      advance(1500)
+      expect(header()?.textContent).toBe(label)
+      render([tool('first', 'success'), tool('second', status)], false)
+      expect(header()?.textContent).toBe(
+        agentName === 'mothership' ? 'Read first + 1' : 'Read files'
+      )
+      act(() => container.querySelector<HTMLElement>('[role="button"]')?.click())
+      expect(container.querySelector('[data-state="open"]')?.textContent).toContain(
+        'Failed reading second'
+      )
+    }
+  )
+
   it('shows final completion immediately and never replays the held action', () => {
     render([tool('first')])
     advance(100)

apps/sim/app/workspace/[workspaceId]/home/components/message-content/components/agent-group/agent-group-view.tsx

@@ -1,9 +1,9 @@
 'use client'
 
 import { type ComponentType, type ReactNode, useState } from 'react'
-import type { ToolActivity } from '@/lib/mothership/generated/protocol'
 import { ThinkingLoader } from '@/components/ui/thinking-loader'
 import { isBrowserAgentAvailable } from '@/lib/browser-agent/transport'
+import type { ToolActivity } from '@/lib/mothership/generated/protocol'
 import { RETIRED_BROWSER_REQUEST_TAKEOVER_ID } from '@/lib/mothership/tools/retired-tools'
 import { getToolDisplayTitle, getToolStatusDisplayTitle } from '@/lib/mothership/tools/tool-display'
 import { ActivityStream } from '@/app/workspace/[workspaceId]/home/components/message-content/components/agent-group/activity-stream'
@@ -275,6 +275,8 @@ export function AgentGroupView({
     meaningfulItems.some(
       (item) =>
         item.type !== 'tool' ||
+        item.data.status === ToolCallStatus.error ||
+        item.data.status === ToolCallStatus.rejected ||
         needsToolInput(item.data) ||
         item.data.toolName === RETIRED_BROWSER_REQUEST_TAKEOVER_ID
     )

apps/sim/app/workspace/[workspaceId]/home/components/message-content/components/agent-group/agent-group.test.ts

@@ -162,10 +162,8 @@ describe('AgentGroup inline main activity', () => {
   it.each([
     ['executing', 'Reading notes'],
     ['success', 'Read notes'],
-    ['error', 'Failed reading notes'],
     ['cancelled', 'Stopped reading notes'],
     ['skipped', 'Skipped reading notes'],
-    ['rejected', 'Failed reading notes'],
     ['interrupted', 'Stopped reading notes'],
   ] as const)('renders a single %s tool once without a disclosure', (status, expected) => {
     act(() =>
@@ -223,11 +221,11 @@ describe('AgentGroup inline main activity', () => {
 
   it.each([
     ['success', 'Checked search requirements'],
-    ['error', 'Failed searching'],
+    ['error', '1 tool call'],
     ['cancelled', 'Stopped searching'],
     ['skipped', 'Skipped searching'],
     ['interrupted', 'Stopped searching'],
-    ['rejected', 'Failed searching'],
+    ['rejected', '1 tool call'],
   ] as const)('uses an honest grouped activity label after %s', (status, expected) => {
     const item = tool(status)
     act(() =>
@@ -387,7 +385,7 @@ describe('AgentGroup inline main activity', () => {
     expect(container.textContent).not.toContain('Built API')
   })
 
-  it('retains earlier failed outcomes when completed activities collapse into one summary', () => {
+  it('keeps earlier failures in expanded history when completed activities collapse', () => {
     act(() =>
       root.render(
         createElement(AgentGroup, {
@@ -420,11 +418,90 @@ describe('AgentGroup inline main activity', () => {
         })
       )
     )
-    expect(container.querySelector('[role="status"]')?.textContent).toBe(
-      'Checked inputs + 1 · 1 failed'
-    )
+    expect(container.querySelector('[role="status"]')?.textContent).toBe('Checked inputs + 1')
+    act(() => container.querySelector<HTMLElement>('[role="button"]')?.click())
+    expect(container.textContent).toContain('Failed reading first document')
   })
 
+  it.each(['mothership', 'workflow'])(
+    'keeps a lone failed %s call accessible under a neutral disclosure',
+    (agentName) => {
+      act(() =>
+        root.render(
+          createElement(AgentGroup, {
+            agentName,
+            agentLabel: agentName,
+            items: [
+              {
+                type: 'tool',
+                data: {
+                  id: 'failed-read',
+                  toolName: 'read',
+                  displayTitle: 'Reading notes',
+                  status: 'error',
+                },
+              },
+            ],
+          })
+        )
+      )
+      expect(container.textContent).not.toContain('Failed')
+      expect(container.textContent).toContain('1 tool call')
+      const disclosure = container.querySelector<HTMLElement>('[role="button"]')
+      expect(disclosure?.getAttribute('aria-expanded')).toBe('false')
+      act(() => disclosure?.click())
+      expect(container.textContent).toContain('Failed reading notes')
+    }
+  )
+
+  it.each(['executing', 'success'] as const)(
+    'keeps failures out of a %s summary while retaining them in expanded history',
+    (status) => {
+      const items: AgentGroupItem[] = [
+        {
+          type: 'tool',
+          data: {
+            id: 'read-failed',
+            toolName: 'read',
+            displayTitle: 'Reading image.png',
+            status: 'error',
+          },
+        },
+        {
+          type: 'tool',
+          data: { id: 'view', toolName: 'read', displayTitle: 'Viewing image.png', status },
+        },
+        {
+          type: 'tool',
+          data: {
+            id: 'later-failed',
+            toolName: 'read',
+            displayTitle: 'Reading notes',
+            status: 'error',
+          },
+        },
+      ]
+      act(() =>
+        root.render(
+          createElement(AgentGroup, {
+            agentName: 'mothership',
+            agentLabel: 'Sim',
+            items,
+            activity: { id: 'inspect', completedTitle: 'Inspected images' },
+            isStreaming: status === 'executing',
+            isLaneOpen: status === 'executing',
+          })
+        )
+      )
+      expect(container.textContent).toBe(
+        status === 'executing' ? 'Viewing image.png' : 'Viewed image.png + 2'
+      )
+      act(() => container.querySelector<HTMLElement>('[role="button"]')?.click())
+      expect(container.textContent).toContain('Failed reading image.png')
+      expect(container.textContent).toContain('Failed reading notes')
+    }
+  )
+
   it('paces the active status in place and expands the full completed history', () => {
     vi.useFakeTimers()
     const first: AgentGroupItem = {

apps/sim/app/workspace/[workspaceId]/home/components/message-content/components/agent-group/tool-activity-group.test.ts

@@ -64,13 +64,13 @@ describe('getToolActivitySummary', () => {
         tool('apply_file_edit', 'error'),
         tool('terminal_run', 'interrupted'),
       ])
-    ).toBe('Tool activity · 1 stopped')
+    ).toBe('2 tool calls · 1 stopped')
   })
 
   it('uses a neutral summary when every call failed', () => {
     expect(
       getToolActivitySummary([tool('run_workflow', 'error'), tool('terminal', 'rejected')])
-    ).toBe('Tool activity')
+    ).toBe('2 tool calls')
   })
 
   it('does not infer tool failures from workflow results', () => {
@@ -89,7 +89,7 @@ describe('getToolActivitySummary', () => {
   })
 
   it.each([
-    ['rejected', 'Failed running checks'],
+    ['rejected', '1 tool call'],
     ['skipped', 'Skipped running checks'],
     ['interrupted', 'Stopped running checks'],
   ] as const)('labels a single %s tool as finished', (status, expected) => {
@@ -134,9 +134,9 @@ describe('getToolActivitySummary', () => {
 
   it('keeps individual failures explicit without adding aggregate failure badges', () => {
     const rejected = { ...tool('terminal', 'rejected'), displayTitle: 'Running checks' }
-    expect(getToolActivitySummary([rejected])).toBe('Failed running checks')
+    expect(getToolActivitySummary([rejected])).toBe('1 tool call')
     expect(getToolActivitySummary([rejected, tool('read', 'skipped')])).toBe(
-      'Tool activity · 1 skipped'
+      '2 tool calls · 1 skipped'
     )
   })
 

apps/sim/app/workspace/[workspaceId]/home/components/message-content/components/agent-group/tool-activity-group.tsx

@@ -16,10 +16,19 @@ import { type ToolCallData, ToolCallStatus } from '@/app/workspace/[workspaceId]
 
 const MAX_SUMMARY_ACTIONS = 3
 
+function isFailedTool(tool: ToolCallData): boolean {
+  return tool.status === ToolCallStatus.error || tool.status === ToolCallStatus.rejected
+}
+
+function toolCountLabel(tools: ToolCallData[]): string {
+  return `${tools.length} tool ${tools.length === 1 ? 'call' : 'calls'}`
+}
+
 /** Summarize completed actions without describing failed or skipped work as successful. */
 export function getToolActivitySummary(tools: ToolCallData[]): string {
   if (tools.length === 1) {
     const tool = tools[0]
+    if (isFailedTool(tool)) return toolCountLabel(tools)
     return getToolStatusDisplayTitle(
       tool.displayTitle,
       tool.status,
@@ -32,7 +41,7 @@ export function getToolActivitySummary(tools: ToolCallData[]): string {
     MAX_SUMMARY_ACTIONS
   )
   const summary = labels.join(', ')
-  const summaryLabel = summary ? summary[0].toUpperCase() + summary.slice(1) : 'Tool activity'
+  const summaryLabel = summary ? summary[0].toUpperCase() + summary.slice(1) : toolCountLabel(tools)
   return [
     additionalActions > 0 ? `${summaryLabel} +${additionalActions} more` : summaryLabel,
     ...getToolActivityInterruptions(tools),
@@ -56,6 +65,7 @@ export function getActiveToolActivityTitle(
   tool: ToolCallData,
   tools: ToolCallData[]
 ): string {
+  if (isFailedTool(tool)) return toolCountLabel(tools)
   return tool.status === ToolCallStatus.executing || tool.status === ToolCallStatus.success
     ? [label, ...getToolActivityInterruptions(tools)].join(' · ')
     : label
@@ -70,7 +80,9 @@ export function getActivityStatusTool(tools: ToolCallData[]): ToolCallData | und
           ? tool
           : newest,
       undefined
-    ) ?? tools.at(-1)
+    ) ??
+    tools.filter((tool) => !isFailedTool(tool)).at(-1) ??
+    tools.at(-1)
   )
 }
 
@@ -116,12 +128,14 @@ export function ToolActivityGroup({
   )
   const completedActivityLabel = groupedActivity?.completedTitle
     ? failedActivityTool
-      ? getToolStatusDisplayTitle(
-          failedActivityTool.displayTitle,
-          failedActivityTool.status,
-          failedActivityTool.toolName,
-          failedActivityTool.activityDescription
-        )
+      ? isFailedTool(failedActivityTool)
+        ? undefined
+        : getToolStatusDisplayTitle(
+            failedActivityTool.displayTitle,
+            failedActivityTool.status,
+            failedActivityTool.toolName,
+            failedActivityTool.activityDescription
+          )
       : groupedActivity.completedTitle
     : undefined
   const completedLabel = completedActivityLabel
@@ -160,7 +174,7 @@ export function ToolActivityGroup({
           }}
           activityKey={statusTool.id}
           attentionKey={attentionKey}
-          collapsible={tools.length > 1}
+          collapsible={tools.length > 1 || isFailedTool(statusTool)}
           expanded={expanded}
           onToggle={() => setExpanded(!expanded)}
           isStreaming={working && autoScrollActivity}