Add agent prompt sent telemetry

Co-authored-by: Orca <help@stably.ai>

docs/reference/agent-prompt-sent-telemetry.md

@@ -0,0 +1,74 @@
+# Agent Prompt Sent Telemetry
+
+This change adds `agent_prompt_sent` so onboarding and retention analysis can tell the difference between an agent being started and Orca initiating or completing a covered prompt submission to that agent.
+
+## Why this exists
+
+`agent_started` only proves Orca launched an agent-backed PTY with telemetry attached. It does not prove the user reached the product value moment of asking the agent to do work.
+
+For onboarding completion, that distinction matters. A user who creates a workspace and starts an agent may still drop before sending a prompt. `agent_prompt_sent` gives us a cleaner signal for "workspace plus real agent intent." It is not proof that the model processed the prompt; folded argv/flag prompts are recorded after local startup is queued, while post-ready terminal input is recorded only after the write path accepts it.
+
+## Event contract
+
+Event: `agent_prompt_sent`
+
+Allowed properties:
+
+| Property | Meaning |
+| --- | --- |
+| `agent_kind` | Bounded agent family, using the same mapping as `agent_started`. |
+| `launch_source` | Bounded source for where the agent prompt flow came from, such as onboarding or quick launch. |
+| `request_kind` | Whether this is a new request or a supported bounded request type. |
+| `nth_repo_added` | Optional repo-onboarding cohort marker when available. |
+
+The event does not include prompt text, file paths, repository names, branch names, raw command text, or raw error details.
+
+## When it fires
+
+The event records only non-empty, non-draft prompt/message paths that Orca submits or queues for local agent startup:
+
+- A prompt folded into an agent launch command after local startup is queued.
+- A prompt sent after the agent is ready through the workspace startup path and verified terminal input accepts it.
+- A post-ready pasted prompt that Orca submits on behalf of the user after verified terminal input accepts it.
+
+The implementation covers renderer paths that launch an agent and submit the user's first prompt through Orca's launch/startup helpers:
+
+- `src/renderer/src/hooks/useComposerState.ts`
+- `src/renderer/src/lib/launch-agent-in-new-tab.ts`
+- `src/renderer/src/lib/new-workspace.ts`
+
+That includes onboarding/new workspace launches, tab-bar quick launch, and other `launchAgentInNewTab` submit-after-ready surfaces such as notes send, conflict resolution, source-control recovery, and task pages.
+
+For local PTYs, "accepted" means main accepted the write through `pty:writeAccepted`. For remote web runtime PTYs, it means the runtime `terminal.send` RPC returned `accepted: true`. SSH relay PTYs cannot acknowledge writes yet, so these paths still attempt the prompt write but intentionally do not emit `agent_prompt_sent` when acceptance cannot be known.
+
+## When it does not fire
+
+The event intentionally does not fire for:
+
+- Bare agent starts with no user prompt.
+- Draft-prefill flows where Orca places text in the composer but does not submit it.
+- Empty or whitespace-only prompts.
+- Incidental terminal input outside the agent prompt submission paths.
+
+## How to use it
+
+For onboarding retention, prefer splitting users like this:
+
+| Segment | Interpretation |
+| --- | --- |
+| `workspace_created`, no `agent_started` | Workspace created, but no agent launched. |
+| `agent_started`, no `agent_prompt_sent` | Agent process launched, but no clear user message. |
+| `agent_prompt_sent` | User submitted a real agent request. Stronger activation signal. |
+
+This should make the "workspace/setup action only" and "agent started only" buckets less ambiguous. It also gives us a better way to identify users who technically completed onboarding but never reached a meaningful agent interaction.
+
+## Validation
+
+The worker validated the change with:
+
+- `pnpm test src/shared/telemetry-events.test.ts src/main/telemetry/validator.test.ts src/renderer/src/lib/agent-prompt-sent-telemetry.test.ts src/renderer/src/lib/agent-paste-draft.test.ts src/renderer/src/lib/launch-agent-in-new-tab.test.ts src/renderer/src/lib/new-workspace.test.ts src/renderer/src/runtime/runtime-terminal-inspection.test.ts`
+- `pnpm run typecheck:web`
+- `pnpm run typecheck:node`
+- `git diff --check`
+
+PostHog first-seen validation remains release-dependent. After this ships, update `docs/reference/telemetry-availability.md` with the first observed production timestamp for `agent_prompt_sent`.

docs/reference/telemetry-availability.md

@@ -39,13 +39,14 @@ D1+/D3+/D7+ retention means the user fired `app_opened` at least once after 24/7
 - `source = 'unknown'` is not a real product surface. It means the caller omitted a source or the value failed schema validation.
 - `workspace_created` means create-worktree IPC succeeded. It is not a general "usable workspace exists" or "workspace revealed" marker.
 - `agent_started` means PTY spawn succeeded with agent telemetry attached. It is not first-repo activation and does not prove the user sent a prompt.
+- `agent_prompt_sent` means Orca submitted a non-draft prompt/message through an agent launch or post-ready agent input path. It excludes agent auto-start, bare shells, and draft prefill.
 - Workspace-outcome joins are native Electron coverage unless the query explicitly proves remote/web instrumentation. Remote runtime and web paths can bypass native repo/worktree telemetry, so do not interpret missing workspace outcome rows as product drop-off for SSH, remote, or web users.
 
 ## Rollouts
 
 ### 2026-05-08 - Repo Cohort Property
 
-Scope: `nth_repo_added` on repo/activation/retention events. Current schemas declare it on `app_opened`, `repo_added`, `add_repo_setup_step_action`, `add_repo_existing_workspaces_detected`, `workspace_created`, `workspace_create_failed`, `setup_script_prompt_shown`, `setup_script_prompt_action`, `agent_started`, and `agent_error`. The original rollout covered `app_opened`, `repo_added`, `add_repo_setup_step_action`, `workspace_created`, `workspace_create_failed`, `agent_started`, and `agent_error`; later events have their own first-seen timestamps below.
+Scope: `nth_repo_added` on repo/activation/retention events. Current schemas declare it on `app_opened`, `repo_added`, `add_repo_setup_step_action`, `add_repo_existing_workspaces_detected`, `workspace_created`, `workspace_create_failed`, `setup_script_prompt_shown`, `setup_script_prompt_action`, `agent_started`, `agent_prompt_sent`, and `agent_error`. The original rollout covered `app_opened`, `repo_added`, `add_repo_setup_step_action`, `workspace_created`, `workspace_create_failed`, `agent_started`, and `agent_error`; later events have their own first-seen timestamps below.
 
 | Field | Value |
 | --- | --- |
@@ -65,6 +66,7 @@ PostHog evidence checked at `2026-05-23T23:34:32Z`:
 | `app_opened` | `2026-05-08T18:40:00.354Z` (`1.3.42-rc.1`) |
 | `workspace_created` | `2026-05-08T19:15:06.913Z` |
 | `agent_started` | `2026-05-08T19:15:07.130Z` |
+| `agent_prompt_sent` | `TBD` |
 | `repo_added` | `2026-05-08T20:01:06.364Z` (`1.3.42`) |
 | `add_repo_setup_step_action` | `2026-05-08T20:01:20.897Z` |
 | `workspace_create_failed` | `2026-05-08T23:04:32.315Z` |

src/main/telemetry/validator.test.ts

@@ -31,6 +31,16 @@ describe('validate', () => {
     expect(result.ok).toBe(true)
   })
 
+  it('accepts a well-formed agent_prompt_sent payload', () => {
+    const result = validate('agent_prompt_sent', {
+      agent_kind: 'claude-code',
+      launch_source: 'new_workspace_composer',
+      request_kind: 'new',
+      nth_repo_added: 1
+    })
+    expect(result.ok).toBe(true)
+  })
+
   it('drops unknown event names', () => {
     const result = validate('not_a_real_event' as never, {})
     expect(result.ok).toBe(false)

src/renderer/src/hooks/useComposerState.ts

@@ -19,6 +19,8 @@ import { filterEnabledTuiAgents, isTuiAgentEnabled } from '../../../shared/tui-a
 import { tuiAgentToAgentKind } from '@/lib/telemetry'
 import { isGitRepoKind } from '../../../shared/repo-kind'
 import { callRuntimeRpc, getActiveRuntimeTarget } from '@/runtime/runtime-rpc-client'
+import type { EventProps } from '../../../shared/telemetry-events'
+import { trackFoldedStartupPromptSent } from '@/lib/agent-prompt-sent-telemetry'
 import type {
   GitHubWorkItem,
   GitPushTarget,
@@ -1870,7 +1872,8 @@ export function useComposerState(options: UseComposerStateOptions): UseComposerS
         launch_source: telemetrySource === 'onboarding' ? 'onboarding' : 'new_workspace_composer',
         request_kind: 'new'
       }
-      activateAndRevealWorktree(worktree.id, {
+      const composerPromptTelemetry: EventProps<'agent_prompt_sent'> = composerTelemetry
+      const activation = activateAndRevealWorktree(worktree.id, {
         sidebarRevealBehavior: 'auto',
         setup: result.setup,
         issueCommand,
@@ -1893,9 +1896,16 @@ export function useComposerState(options: UseComposerStateOptions): UseComposerS
           : {})
       })
       if (startupPlan) {
+        trackFoldedStartupPromptSent({
+          prompt: submitStartupPrompt,
+          startupPlan,
+          promptTelemetry: composerPromptTelemetry,
+          startupQueued: activation !== false && activation.primaryTabId !== null
+        })
         void ensureAgentStartupInTerminal({
           worktreeId: worktree.id,
-          startup: startupPlan
+          startup: startupPlan,
+          promptTelemetry: composerPromptTelemetry
         })
       }
       setSidebarOpen(true)
@@ -2138,7 +2148,8 @@ export function useComposerState(options: UseComposerStateOptions): UseComposerS
                   telemetrySource === 'onboarding' ? 'onboarding' : 'new_workspace_composer',
                 request_kind: 'new'
               }
-        activateAndRevealWorktree(worktree.id, {
+        const quickPromptTelemetry: EventProps<'agent_prompt_sent'> | null = quickTelemetry
+        const activation = activateAndRevealWorktree(worktree.id, {
           sidebarRevealBehavior: 'auto',
           setup: result.setup,
           ...(startupPlan
@@ -2160,9 +2171,16 @@ export function useComposerState(options: UseComposerStateOptions): UseComposerS
             : {})
         })
         if (startupPlan) {
+          trackFoldedStartupPromptSent({
+            prompt: quickPrompt,
+            startupPlan,
+            promptTelemetry: quickPromptTelemetry,
+            startupQueued: activation !== false && activation.primaryTabId !== null
+          })
           void ensureAgentStartupInTerminal({
             worktreeId: worktree.id,
-            startup: startupPlan
+            startup: startupPlan,
+            ...(quickPromptTelemetry ? { promptTelemetry: quickPromptTelemetry } : {})
           })
         }
         setSidebarOpen(true)

src/renderer/src/lib/agent-paste-draft.test.ts

@@ -10,7 +10,7 @@ const testState = vi.hoisted(() => ({
   unsubscribe: vi.fn(),
   subscribeToPtyData: vi.fn(),
   isRemoteRuntimePtyId: vi.fn(),
-  sendRuntimePtyInput: vi.fn(),
+  sendRuntimePtyInputVerified: vi.fn(),
   subscribeToRuntimeTerminalData: vi.fn()
 }))
 
@@ -26,7 +26,7 @@ vi.mock('@/components/terminal-pane/pty-dispatcher', () => ({
 
 vi.mock('@/runtime/runtime-terminal-inspection', () => ({
   isRemoteRuntimePtyId: testState.isRemoteRuntimePtyId,
-  sendRuntimePtyInput: testState.sendRuntimePtyInput
+  sendRuntimePtyInputVerified: testState.sendRuntimePtyInputVerified
 }))
 
 vi.mock('@/runtime/runtime-terminal-stream', () => ({
@@ -58,7 +58,8 @@ describe('pasteDraftWhenAgentReady', () => {
     )
     testState.isRemoteRuntimePtyId.mockReset()
     testState.isRemoteRuntimePtyId.mockReturnValue(false)
-    testState.sendRuntimePtyInput.mockReset()
+    testState.sendRuntimePtyInputVerified.mockReset()
+    testState.sendRuntimePtyInputVerified.mockResolvedValue(true)
     testState.subscribeToRuntimeTerminalData.mockReset()
   })
 
@@ -77,16 +78,20 @@ describe('pasteDraftWhenAgentReady', () => {
 
     testState.ptyObserver?.(CODEX_COMPOSER_PROMPT_RENDER)
     await flushMicrotasks()
-    expect(testState.sendRuntimePtyInput).not.toHaveBeenCalled()
+    expect(testState.sendRuntimePtyInputVerified).not.toHaveBeenCalled()
 
     testState.ptyObserver?.(DECSET_BRACKETED_PASTE)
     await flushMicrotasks()
-    expect(testState.sendRuntimePtyInput).not.toHaveBeenCalled()
+    expect(testState.sendRuntimePtyInputVerified).not.toHaveBeenCalled()
 
     testState.ptyObserver?.(CODEX_COMPOSER_PROMPT_RENDER)
 
     await expect(promise).resolves.toBe(true)
-    expect(testState.sendRuntimePtyInput).toHaveBeenCalledWith({}, 'pty-1', PASTED_ISSUE_URL)
+    expect(testState.sendRuntimePtyInputVerified).toHaveBeenCalledWith(
+      {},
+      'pty-1',
+      PASTED_ISSUE_URL
+    )
     expect(vi.getTimerCount()).toBe(0)
   })
 
@@ -103,7 +108,11 @@ describe('pasteDraftWhenAgentReady', () => {
     )
 
     await expect(promise).resolves.toBe(true)
-    expect(testState.sendRuntimePtyInput).toHaveBeenCalledWith({}, 'pty-1', PASTED_ISSUE_URL)
+    expect(testState.sendRuntimePtyInputVerified).toHaveBeenCalledWith(
+      {},
+      'pty-1',
+      PASTED_ISSUE_URL
+    )
   })
 
   it('keeps the render-quiet wait for agents without the Codex ready signal', async () => {
@@ -116,15 +125,19 @@ describe('pasteDraftWhenAgentReady', () => {
 
     testState.ptyObserver?.(DECSET_BRACKETED_PASTE)
     await flushMicrotasks()
-    expect(testState.sendRuntimePtyInput).not.toHaveBeenCalled()
+    expect(testState.sendRuntimePtyInputVerified).not.toHaveBeenCalled()
 
     await vi.advanceTimersByTimeAsync(1499)
-    expect(testState.sendRuntimePtyInput).not.toHaveBeenCalled()
+    expect(testState.sendRuntimePtyInputVerified).not.toHaveBeenCalled()
 
     await vi.advanceTimersByTimeAsync(1)
 
     await expect(promise).resolves.toBe(true)
-    expect(testState.sendRuntimePtyInput).toHaveBeenCalledWith({}, 'pty-1', PASTED_ISSUE_URL)
+    expect(testState.sendRuntimePtyInputVerified).toHaveBeenCalledWith(
+      {},
+      'pty-1',
+      PASTED_ISSUE_URL
+    )
   })
 
   it('does not paste for agents that already use native draft prefill', async () => {
@@ -137,7 +150,7 @@ describe('pasteDraftWhenAgentReady', () => {
     ).resolves.toBe(false)
 
     expect(testState.subscribeToPtyData).not.toHaveBeenCalled()
-    expect(testState.sendRuntimePtyInput).not.toHaveBeenCalled()
+    expect(testState.sendRuntimePtyInputVerified).not.toHaveBeenCalled()
   })
 
   it('can force paste and submit for native-prefill agents', async () => {
@@ -154,7 +167,39 @@ describe('pasteDraftWhenAgentReady', () => {
     await vi.advanceTimersByTimeAsync(1500)
 
     await expect(promise).resolves.toBe(true)
-    expect(testState.sendRuntimePtyInput).toHaveBeenCalledWith({}, 'pty-1', `${PASTED_ISSUE_URL}\r`)
+    expect(testState.sendRuntimePtyInputVerified).toHaveBeenCalledWith(
+      {},
+      'pty-1',
+      `${PASTED_ISSUE_URL}\r`
+    )
+  })
+
+  it('reports false when verified input delivery fails', async () => {
+    testState.sendRuntimePtyInputVerified.mockResolvedValue(false)
+    const promise = pasteDraftWhenAgentReady({
+      tabId: 'tab-1',
+      content: ISSUE_URL,
+      agent: 'codex'
+    })
+    await flushMicrotasks()
+
+    testState.ptyObserver?.(`${DECSET_BRACKETED_PASTE}${CODEX_COMPOSER_PROMPT_RENDER}`)
+
+    await expect(promise).resolves.toBe(false)
+  })
+
+  it('reports false when verified input delivery rejects', async () => {
+    testState.sendRuntimePtyInputVerified.mockRejectedValue(new Error('runtime timeout'))
+    const promise = pasteDraftWhenAgentReady({
+      tabId: 'tab-1',
+      content: ISSUE_URL,
+      agent: 'codex'
+    })
+    await flushMicrotasks()
+
+    testState.ptyObserver?.(`${DECSET_BRACKETED_PASTE}${CODEX_COMPOSER_PROMPT_RENDER}`)
+
+    await expect(promise).resolves.toBe(false)
   })
 })
 

src/renderer/src/lib/agent-paste-draft.ts

@@ -4,7 +4,6 @@ import { useAppStore } from '@/store'
 import { subscribeToPtyData } from '@/components/terminal-pane/pty-dispatcher'
 import {
   isRemoteRuntimePtyId,
-  sendRuntimePtyInput,
   sendRuntimePtyInputVerified
 } from '@/runtime/runtime-terminal-inspection'
 import { subscribeToRuntimeTerminalData } from '@/runtime/runtime-terminal-stream'
@@ -95,12 +94,15 @@ export async function pasteDraftWhenAgentReady(args: {
     return false
   }
 
-  sendRuntimePtyInput(
-    useAppStore.getState().settings,
-    ptyId,
-    `${BRACKETED_PASTE_BEGIN}${content}${BRACKETED_PASTE_END}${submit ? '\r' : ''}`
-  )
-  return true
+  try {
+    return await sendRuntimePtyInputVerified(
+      useAppStore.getState().settings,
+      ptyId,
+      `${BRACKETED_PASTE_BEGIN}${content}${BRACKETED_PASTE_END}${submit ? '\r' : ''}`
+    )
+  } catch {
+    return false
+  }
 }
 
 export async function submitPromptToAgentTab(args: {

src/renderer/src/lib/agent-prompt-sent-telemetry.test.ts

@@ -0,0 +1,79 @@
+import { beforeEach, describe, expect, it, vi } from 'vitest'
+
+const mockTrack = vi.hoisted(() => vi.fn())
+
+vi.mock('@/lib/telemetry', () => ({
+  track: mockTrack
+}))
+
+import { trackFoldedStartupPromptSent } from './agent-prompt-sent-telemetry'
+
+describe('trackFoldedStartupPromptSent', () => {
+  beforeEach(() => {
+    vi.clearAllMocks()
+  })
+
+  it('tracks a folded startup prompt once with bounded telemetry', () => {
+    trackFoldedStartupPromptSent({
+      prompt: 'fix the spinner',
+      startupPlan: {
+        followupPrompt: null,
+        draftPrompt: null
+      },
+      startupQueued: true,
+      promptTelemetry: {
+        agent_kind: 'codex',
+        launch_source: 'new_workspace_composer',
+        request_kind: 'new'
+      }
+    })
+
+    expect(mockTrack).toHaveBeenCalledOnce()
+    expect(mockTrack).toHaveBeenCalledWith('agent_prompt_sent', {
+      agent_kind: 'codex',
+      launch_source: 'new_workspace_composer',
+      request_kind: 'new'
+    })
+  })
+
+  it('does not track empty prompts, follow-up prompts, drafts, or bare shells', () => {
+    const promptTelemetry = {
+      agent_kind: 'codex',
+      launch_source: 'new_workspace_composer',
+      request_kind: 'new'
+    } as const
+
+    trackFoldedStartupPromptSent({
+      prompt: '   ',
+      startupPlan: { followupPrompt: null, draftPrompt: null },
+      startupQueued: true,
+      promptTelemetry
+    })
+    trackFoldedStartupPromptSent({
+      prompt: 'fix the spinner',
+      startupPlan: { followupPrompt: 'fix the spinner', draftPrompt: null },
+      startupQueued: true,
+      promptTelemetry
+    })
+    trackFoldedStartupPromptSent({
+      prompt: 'https://example.test/issue/1',
+      startupPlan: { followupPrompt: null, draftPrompt: 'https://example.test/issue/1' },
+      startupQueued: true,
+      promptTelemetry
+    })
+    trackFoldedStartupPromptSent({
+      prompt: 'fix the spinner',
+      startupPlan: { followupPrompt: null, draftPrompt: null },
+      startupQueued: true,
+      promptTelemetry: null
+    })
+    trackFoldedStartupPromptSent({
+      prompt: 'fix the spinner',
+      startupPlan: { followupPrompt: null, draftPrompt: null },
+      startupQueued: false,
+      promptTelemetry
+    })
+
+    expect(mockTrack).not.toHaveBeenCalled()
+  })
+})

src/renderer/src/lib/agent-prompt-sent-telemetry.ts

@@ -0,0 +1,25 @@
+import { track } from '@/lib/telemetry'
+import type { AgentStartupPlan } from '@/lib/tui-agent-startup'
+import type { EventProps } from '../../../shared/telemetry-events'
+
+export function trackFoldedStartupPromptSent(args: {
+  prompt: string
+  startupPlan: Pick<AgentStartupPlan, 'followupPrompt' | 'draftPrompt'>
+  promptTelemetry: EventProps<'agent_prompt_sent'> | null
+  startupQueued: boolean
+}): void {
+  const { prompt, startupPlan, promptTelemetry, startupQueued } = args
+  if (!startupQueued) {
+    return
+  }
+  if (!promptTelemetry) {
+    return
+  }
+  if (prompt.trim().length === 0) {
+    return
+  }
+  if (startupPlan.followupPrompt !== null || startupPlan.draftPrompt) {
+    return
+  }
+  track('agent_prompt_sent', promptTelemetry)
+}

src/renderer/src/lib/launch-agent-in-new-tab.test.ts

@@ -6,6 +6,7 @@ const mockSetActiveTabType = vi.fn()
 const mockSetTabBarOrder = vi.fn()
 const mockSetAgentStatus = vi.fn()
 const mockPasteDraftWhenAgentReady = vi.fn()
+const mockTrack = vi.fn()
 
 const LEAF_ID = '11111111-1111-4111-8111-111111111111'
 
@@ -50,7 +51,7 @@ vi.mock('@/lib/agent-paste-draft', () => ({
 }))
 
 vi.mock('@/lib/telemetry', () => ({
-  track: vi.fn(),
+  track: mockTrack,
   tuiAgentToAgentKind: (agent: string) => agent
 }))
 
@@ -88,6 +89,37 @@ describe('launchAgentInNewTab', () => {
     )
   })
 
+  it('tracks one prompt-sent event for argv prompt launches', async () => {
+    const { launchAgentInNewTab } = await import('./launch-agent-in-new-tab')
+
+    launchAgentInNewTab({
+      agent: 'codex',
+      worktreeId: 'wt-1',
+      prompt: 'fix the spinner',
+      launchSource: 'onboarding'
+    })
+
+    expect(mockTrack).toHaveBeenCalledTimes(1)
+    expect(mockTrack).toHaveBeenCalledWith('agent_prompt_sent', {
+      agent_kind: 'codex',
+      launch_source: 'onboarding',
+      request_kind: 'new'
+    })
+  })
+
+  it('does not track prompt-sent for draft launches', async () => {
+    const { launchAgentInNewTab } = await import('./launch-agent-in-new-tab')
+
+    launchAgentInNewTab({
+      agent: 'claude',
+      worktreeId: 'wt-1',
+      prompt: 'review this before sending',
+      promptDelivery: 'draft'
+    })
+
+    expect(mockTrack).not.toHaveBeenCalledWith('agent_prompt_sent', expect.anything())
+  })
+
   it('seeds working after Command Code submit-after-ready prompt delivery', async () => {
     const { launchAgentInNewTab } = await import('./launch-agent-in-new-tab')
 
@@ -120,5 +152,26 @@ describe('launchAgentInNewTab', () => {
       prompt: 'large generated prompt',
       agentType: 'command-code'
     })
+    expect(mockTrack).toHaveBeenCalledTimes(1)
+    expect(mockTrack).toHaveBeenCalledWith('agent_prompt_sent', {
+      agent_kind: 'command-code',
+      launch_source: 'tab_bar_quick_launch',
+      request_kind: 'new'
+    })
+  })
+
+  it('does not track prompt-sent when submit-after-ready delivery fails', async () => {
+    mockPasteDraftWhenAgentReady.mockResolvedValue(false)
+    const { launchAgentInNewTab } = await import('./launch-agent-in-new-tab')
+
+    launchAgentInNewTab({
+      agent: 'command-code',
+      worktreeId: 'wt-1',
+      prompt: 'large generated prompt',
+      promptDelivery: 'submit-after-ready'
+    })
+    await Promise.resolve()
+
+    expect(mockTrack).not.toHaveBeenCalledWith('agent_prompt_sent', expect.anything())
   })
 })

src/renderer/src/lib/launch-agent-in-new-tab.ts

@@ -12,7 +12,7 @@ import { pasteDraftWhenAgentReady } from '@/lib/agent-paste-draft'
 import { TUI_AGENT_CONFIG } from '../../../shared/tui-agent-config'
 import { makePaneKey } from '../../../shared/stable-pane-id'
 import type { TuiAgent } from '../../../shared/types'
-import type { LaunchSource } from '../../../shared/telemetry-events'
+import type { EventProps, LaunchSource } from '../../../shared/telemetry-events'
 
 export type LaunchAgentInNewTabArgs = {
   agent: TuiAgent
@@ -93,6 +93,11 @@ export function launchAgentInNewTab(args: LaunchAgentInNewTabArgs): LaunchAgentI
   const trimmedPrompt = prompt?.trim() ?? ''
   const hasPrompt = trimmedPrompt.length > 0
   const isFollowupPath = TUI_AGENT_CONFIG[agent].promptInjectionMode === 'stdin-after-start'
+  const promptTelemetry: EventProps<'agent_prompt_sent'> = {
+    agent_kind: tuiAgentToAgentKind(agent),
+    launch_source: launchSource ?? 'tab_bar_quick_launch',
+    request_kind: 'new'
+  }
 
   // Why: argv/flag agents fold the prompt into the launch command and
   // auto-submit — keeping behavior consistent with the composer/tab-bar `+`
@@ -189,6 +194,14 @@ export function launchAgentInNewTab(args: LaunchAgentInNewTabArgs): LaunchAgentI
       request_kind: 'new'
     }
   })
+  if (
+    hasPrompt &&
+    promptDelivery === 'auto-submit' &&
+    startupPlan.followupPrompt === null &&
+    pasteDraftAfterLaunch === null
+  ) {
+    track('agent_prompt_sent', promptTelemetry)
+  }
 
   // Why: schedule the bracketed-paste-after-ready follow-up immediately after
   // the startup command is queued. Fire-and-forget so callers keep their
@@ -234,6 +247,9 @@ export function launchAgentInNewTab(args: LaunchAgentInNewTabArgs): LaunchAgentI
       }
     }).then((delivered) => {
       if (delivered) {
+        if (submitPastedPrompt) {
+          track('agent_prompt_sent', promptTelemetry)
+        }
         if (agent === 'command-code' && submitPastedPrompt) {
           // Why: Command Code has no prompt-submit hook; when Orca submits a
           // generated prompt after readiness, seed working at delivery time.

src/renderer/src/lib/new-workspace.test.ts

@@ -1,5 +1,48 @@
-import { describe, expect, it } from 'vitest'
-import { getWorkspaceSeedName, isGitLabIssueUrl } from './new-workspace'
+import { beforeEach, describe, expect, it, vi } from 'vitest'
+
+const {
+  mockInspectRuntimeTerminalProcess,
+  mockSendRuntimePtyInputVerified,
+  mockPasteDraftWhenAgentReady,
+  mockTrack,
+  store
+} = vi.hoisted(() => ({
+  mockInspectRuntimeTerminalProcess: vi.fn(),
+  mockSendRuntimePtyInputVerified: vi.fn(),
+  mockPasteDraftWhenAgentReady: vi.fn(),
+  mockTrack: vi.fn(),
+  store: {
+    settings: {},
+    activeTabIdByWorktree: { 'wt-1': 'tab-1' } as Record<string, string>,
+    tabsByWorktree: { 'wt-1': [{ id: 'tab-1' }] } as Record<string, { id: string }[]>,
+    ptyIdsByTabId: { 'tab-1': ['pty-1'] } as Record<string, string[]>
+  }
+}))
+
+vi.mock('@/store', () => ({
+  useAppStore: {
+    getState: () => store
+  }
+}))
+
+vi.mock('@/runtime/runtime-terminal-inspection', () => ({
+  inspectRuntimeTerminalProcess: mockInspectRuntimeTerminalProcess,
+  sendRuntimePtyInputVerified: mockSendRuntimePtyInputVerified
+}))
+
+vi.mock('@/lib/agent-paste-draft', () => ({
+  pasteDraftWhenAgentReady: mockPasteDraftWhenAgentReady
+}))
+
+vi.mock('@/lib/telemetry', () => ({
+  track: mockTrack
+}))
+
+import {
+  ensureAgentStartupInTerminal,
+  getWorkspaceSeedName,
+  isGitLabIssueUrl
+} from './new-workspace'
 
 describe('getWorkspaceSeedName', () => {
   it('prefers an explicit name', () => {
@@ -124,3 +167,112 @@ describe('isGitLabIssueUrl', () => {
     expect(isGitLabIssueUrl('https://github.com/group/project/issues/123')).toBe(false)
   })
 })
+
+describe('ensureAgentStartupInTerminal prompt telemetry', () => {
+  beforeEach(() => {
+    vi.clearAllMocks()
+    store.settings = {}
+    store.activeTabIdByWorktree = { 'wt-1': 'tab-1' }
+    store.tabsByWorktree = { 'wt-1': [{ id: 'tab-1' }] }
+    store.ptyIdsByTabId = { 'tab-1': ['pty-1'] }
+    mockInspectRuntimeTerminalProcess.mockResolvedValue({
+      foregroundProcess: 'aider',
+      hasChildProcesses: true
+    })
+    mockSendRuntimePtyInputVerified.mockResolvedValue(true)
+    mockPasteDraftWhenAgentReady.mockResolvedValue(true)
+  })
+
+  it('tracks after a follow-up prompt is accepted by the terminal runtime', async () => {
+    await ensureAgentStartupInTerminal({
+      worktreeId: 'wt-1',
+      startup: {
+        agent: 'aider',
+        launchCommand: 'aider',
+        expectedProcess: 'aider',
+        followupPrompt: 'fix the spinner'
+      },
+      promptTelemetry: {
+        agent_kind: 'aider',
+        launch_source: 'new_workspace_composer',
+        request_kind: 'new'
+      }
+    })
+
+    expect(mockSendRuntimePtyInputVerified).toHaveBeenCalledWith({}, 'pty-1', 'fix the spinner\r')
+    expect(mockTrack).toHaveBeenCalledWith('agent_prompt_sent', {
+      agent_kind: 'aider',
+      launch_source: 'new_workspace_composer',
+      request_kind: 'new'
+    })
+  })
+
+  it('does not track when follow-up prompt delivery is rejected by the terminal runtime', async () => {
+    mockSendRuntimePtyInputVerified.mockResolvedValue(false)
+
+    await ensureAgentStartupInTerminal({
+      worktreeId: 'wt-1',
+      startup: {
+        agent: 'aider',
+        launchCommand: 'aider',
+        expectedProcess: 'aider',
+        followupPrompt: 'fix the spinner'
+      },
+      promptTelemetry: {
+        agent_kind: 'aider',
+        launch_source: 'new_workspace_composer',
+        request_kind: 'new'
+      }
+    })
+
+    expect(mockTrack).not.toHaveBeenCalledWith('agent_prompt_sent', expect.anything())
+  })
+
+  it('does not track when follow-up prompt delivery rejects', async () => {
+    mockSendRuntimePtyInputVerified.mockRejectedValue(new Error('runtime timeout'))
+
+    await expect(
+      ensureAgentStartupInTerminal({
+        worktreeId: 'wt-1',
+        startup: {
+          agent: 'aider',
+          launchCommand: 'aider',
+          expectedProcess: 'aider',
+          followupPrompt: 'fix the spinner'
+        },
+        promptTelemetry: {
+          agent_kind: 'aider',
+          launch_source: 'new_workspace_composer',
+          request_kind: 'new'
+        }
+      })
+    ).resolves.toBeUndefined()
+
+    expect(mockTrack).not.toHaveBeenCalledWith('agent_prompt_sent', expect.anything())
+  })
+
+  it('does not track draft prompt delivery as a sent prompt', async () => {
+    await ensureAgentStartupInTerminal({
+      worktreeId: 'wt-1',
+      startup: {
+        agent: 'claude',
+        launchCommand: 'claude',
+        expectedProcess: 'claude',
+        followupPrompt: null,
+        draftPrompt: 'review this before sending'
+      },
+      promptTelemetry: {
+        agent_kind: 'claude-code',
+        launch_source: 'new_workspace_composer',
+        request_kind: 'new'
+      }
+    })
+
+    expect(mockPasteDraftWhenAgentReady).toHaveBeenCalledWith({
+      tabId: 'tab-1',
+      content: 'review this before sending',
+      agent: 'claude'
+    })
+    expect(mockTrack).not.toHaveBeenCalledWith('agent_prompt_sent', expect.anything())
+  })
+})

src/renderer/src/lib/new-workspace.ts

@@ -2,10 +2,12 @@ import { useAppStore } from '@/store'
 import { pasteDraftWhenAgentReady } from '@/lib/agent-paste-draft'
 import {
   inspectRuntimeTerminalProcess,
-  sendRuntimePtyInput
+  sendRuntimePtyInputVerified
 } from '@/runtime/runtime-terminal-inspection'
+import { track } from '@/lib/telemetry'
 import type { AgentStartupPlan } from '@/lib/tui-agent-startup'
 import { isShellProcess } from '@/lib/tui-agent-startup'
+import type { EventProps } from '../../../shared/telemetry-events'
 import type { OrcaHooks, TaskViewPresetId } from '../../../shared/types'
 import { resolveHookCommandSourcePolicy } from '../../../shared/hook-command-source-policy'
 import { isExpectedAgentProcess } from '../../../shared/agent-process-recognition'
@@ -208,8 +210,9 @@ export function getWorkspaceSeedName(args: {
 export async function ensureAgentStartupInTerminal(args: {
   worktreeId: string
   startup: AgentStartupPlan
+  promptTelemetry?: EventProps<'agent_prompt_sent'>
 }): Promise<void> {
-  const { worktreeId, startup } = args
+  const { worktreeId, startup, promptTelemetry } = args
   const draftPrompt = startup.draftPrompt ?? null
   if (startup.followupPrompt === null && draftPrompt === null) {
     return
@@ -244,7 +247,12 @@ export async function ensureAgentStartupInTerminal(args: {
   // session and submitted. Wait until the agent owns the PTY before writing.
   if (startup.followupPrompt) {
     await waitForAgentForeground(ptyId, startup.expectedProcess)
-    sendRuntimePtyInput(useAppStore.getState().settings, ptyId, `${startup.followupPrompt}\r`)
+    const sent = await sendFollowupPrompt(ptyId, startup.followupPrompt)
+    if (sent) {
+      if (promptTelemetry) {
+        track('agent_prompt_sent', promptTelemetry)
+      }
+    }
   }
 
   // Why: draftPrompt uses bracketed-paste so the URL lands atomically in the
@@ -259,6 +267,14 @@ export async function ensureAgentStartupInTerminal(args: {
   }
 }
 
+async function sendFollowupPrompt(ptyId: string, prompt: string): Promise<boolean> {
+  try {
+    return await sendRuntimePtyInputVerified(useAppStore.getState().settings, ptyId, `${prompt}\r`)
+  } catch {
+    return false
+  }
+}
+
 // Why: legacy followupPrompt path used `agentOwnsForeground` exclusively (with
 // a hasChildProcesses fallback after several polls). Preserve that behavior so
 // stdin-after-start agents still receive their prompt under the same

src/renderer/src/runtime/runtime-terminal-inspection.test.ts

@@ -1,5 +1,9 @@
 import { beforeEach, describe, expect, it, vi } from 'vitest'
-import { inspectRuntimeTerminalProcess, sendRuntimePtyInput } from './runtime-terminal-inspection'
+import {
+  inspectRuntimeTerminalProcess,
+  sendRuntimePtyInput,
+  sendRuntimePtyInputVerified
+} from './runtime-terminal-inspection'
 import {
   createCompatibleRuntimeStatusResponseIfNeeded,
   type RuntimeEnvironmentCallRequest
@@ -10,6 +14,7 @@ describe('runtime terminal owner routing', () => {
   const runtimeCall = vi.fn()
   const runtimeTransportCall = vi.fn()
   const localWrite = vi.fn()
+  const localWriteAccepted = vi.fn()
   const localForeground = vi.fn()
   const localHasChildren = vi.fn()
 
@@ -29,6 +34,7 @@ describe('runtime terminal owner routing', () => {
         runtimeEnvironments: { call: runtimeTransportCall },
         pty: {
           write: localWrite,
+          writeAccepted: localWriteAccepted,
           getForegroundProcess: localForeground,
           hasChildProcesses: localHasChildren
         }
@@ -83,4 +89,68 @@ describe('runtime terminal owner routing', () => {
       )
     ).resolves.toEqual({ foregroundProcess: null, hasChildProcesses: false })
   })
+
+  it('reports stale remote terminal handles as rejected during verified send', async () => {
+    runtimeCall.mockResolvedValue({
+      ok: false,
+      error: { code: 'terminal_handle_stale', message: 'terminal_handle_stale' }
+    })
+
+    await expect(
+      sendRuntimePtyInputVerified(
+        { activeRuntimeEnvironmentId: 'env-2' },
+        'remote:env-1@@terminal-stale',
+        'x'
+      )
+    ).resolves.toBe(false)
+  })
+
+  it('reports declined remote terminal sends as rejected during verified send', async () => {
+    runtimeCall.mockResolvedValue({
+      ok: true,
+      result: { send: { handle: 'terminal-1', accepted: false, bytesWritten: 0 } },
+      _meta: { runtimeId: 'runtime-1' }
+    })
+
+    await expect(
+      sendRuntimePtyInputVerified(
+        { activeRuntimeEnvironmentId: 'env-2' },
+        'remote:env-1@@terminal-1',
+        'x'
+      )
+    ).resolves.toBe(false)
+
+    expect(runtimeCall).toHaveBeenCalledWith({
+      selector: 'env-1',
+      method: 'terminal.send',
+      params: {
+        terminal: 'terminal-1',
+        text: 'x',
+        client: { id: 'orca-desktop', type: 'desktop' }
+      },
+      timeoutMs: 15_000
+    })
+  })
+
+  it('uses accepted local writes for verified input', async () => {
+    localWriteAccepted.mockResolvedValue(true)
+
+    await expect(
+      sendRuntimePtyInputVerified({ activeRuntimeEnvironmentId: null }, 'local-pty', 'x')
+    ).resolves.toBe(true)
+
+    expect(localWriteAccepted).toHaveBeenCalledWith('local-pty', 'x')
+    expect(localWrite).not.toHaveBeenCalled()
+  })
+
+  it('falls back to fire-and-forget writes when local acceptance cannot be verified', async () => {
+    localWriteAccepted.mockResolvedValue(false)
+
+    await expect(
+      sendRuntimePtyInputVerified({ activeRuntimeEnvironmentId: null }, 'local-pty', 'x')
+    ).resolves.toBe(false)
+
+    expect(localWriteAccepted).toHaveBeenCalledWith('local-pty', 'x')
+    expect(localWrite).toHaveBeenCalledWith('local-pty', 'x')
+  })
 })

src/renderer/src/runtime/runtime-terminal-inspection.ts

@@ -1,4 +1,5 @@
 import type { GlobalSettings } from '../../../shared/types'
+import type { RuntimeTerminalSend } from '../../../shared/runtime-types'
 import { RuntimeRpcCallError, callRuntimeRpc, getActiveRuntimeTarget } from './runtime-rpc-client'
 import {
   getRemoteRuntimePtyEnvironmentId,
@@ -106,13 +107,21 @@ export async function sendRuntimePtyInputVerified(
     : getActiveRuntimeTarget(settings)
   const terminal = getRemoteRuntimeTerminalHandle(ptyId)
   if (target.kind !== 'environment' || !terminal) {
-    window.api.pty.write(ptyId, data)
-    return true
+    const accepted = await window.api.pty.writeAccepted(ptyId, data)
+    if (!accepted) {
+      window.api.pty.write(ptyId, data)
+    }
+    return accepted
   }
 
   try {
-    await callRuntimeRpc(target, 'terminal.send', { terminal, text: data }, { timeoutMs: 15_000 })
-    return true
+    const result = await callRuntimeRpc<{ send: RuntimeTerminalSend }>(
+      target,
+      'terminal.send',
+      { terminal, text: data, client: { id: 'orca-desktop', type: 'desktop' } },
+      { timeoutMs: 15_000 }
+    )
+    return result.send.accepted === true
   } catch (error) {
     if (isTerminalGoneError(error)) {
       return false

src/shared/telemetry-events.test.ts

@@ -111,6 +111,28 @@ describe('agent_started schema', () => {
   })
 })
 
+describe('agent_prompt_sent schema', () => {
+  it('accepts a bounded prompt-send payload with cohort context', () => {
+    const parsed = eventSchemas.agent_prompt_sent.safeParse({
+      agent_kind: 'codex',
+      launch_source: 'onboarding',
+      request_kind: 'new',
+      nth_repo_added: 1
+    })
+    expect(parsed.success).toBe(true)
+  })
+
+  it('rejects prompt text via .strict()', () => {
+    const parsed = eventSchemas.agent_prompt_sent.safeParse({
+      agent_kind: 'claude-code',
+      launch_source: 'new_workspace_composer',
+      request_kind: 'new',
+      prompt: 'please inspect /Users/alice/private-repo'
+    })
+    expect(parsed.success).toBe(false)
+  })
+})
+
 describe('agent_hook_unattributed schema', () => {
   it('accepts the two bounded attribution failure reasons', () => {
     for (const reason of ['empty_pane_key', 'unknown_tab_id'] as const) {

src/shared/telemetry-events.ts

@@ -281,6 +281,14 @@ const agentStartedSchema = z
     nth_repo_added: nthRepoAddedSchema
   })
   .strict()
+const agentPromptSentSchema = z
+  .object({
+    agent_kind: agentKindSchema,
+    launch_source: launchSourceSchema,
+    request_kind: requestKindSchema,
+    nth_repo_added: nthRepoAddedSchema
+  })
+  .strict()
 
 // Enum-only by design for both fields. `error_message` and `error_stack` are
 // deliberately absent — `.strict()` rejects either key if a call site ever
@@ -1017,6 +1025,7 @@ export const eventSchemas = {
   setup_script_prompt_action: setupScriptPromptActionSchema,
 
   agent_started: agentStartedSchema,
+  agent_prompt_sent: agentPromptSentSchema,
   agent_error: agentErrorSchema,
   agent_hook_install_failed: agentHookInstallFailedSchema,
   agent_hook_unattributed: agentHookUnattributedSchema,
@@ -1107,6 +1116,7 @@ type _CohortExtendedRoster =
   | 'setup_script_prompt_shown'
   | 'setup_script_prompt_action'
   | 'agent_started'
+  | 'agent_prompt_sent'
   | 'agent_error'
 // Why: `z.object({}).strict()` infers a string index signature, which would
 // make every key appear present. Ignore index-signature-only keys here so