diff --git a/src/main/agent-hooks/first-work-branch-rename.test.ts b/src/main/agent-hooks/first-work-branch-rename.test.ts
index d1fad66280f..b7aef25e138 100644
--- a/src/main/agent-hooks/first-work-branch-rename.test.ts
+++ b/src/main/agent-hooks/first-work-branch-rename.test.ts
@@ -272,7 +272,7 @@ describe('maybeAutoRenameBranchOnFirstWork', () => {
   })
 
   it('names the branch from the prompt a launch file carried, when main kept it', async () => {
-    const { deps, onRenamed } = makeDeps({ getLaunchFilePrompt: () => 'Fix auth from note #1' })
+    const { deps, onRenamed } = makeDeps({ showLaunchPrompt: () => 'Fix auth from note #1' })
     await maybeAutoRenameBranchOnFirstWork(
       workingEvent({ prompt: buildLaunchFilePointer('/tmp/orca-launch-file-a1/task-context.md') }),
       deps
diff --git a/src/main/agent-hooks/first-work-branch-rename.ts b/src/main/agent-hooks/first-work-branch-rename.ts
index e5abfe9cc66..3576334b368 100644
--- a/src/main/agent-hooks/first-work-branch-rename.ts
+++ b/src/main/agent-hooks/first-work-branch-rename.ts
@@ -68,8 +68,8 @@ export type FirstWorkBranchRenameDeps = {
   resolveWorktreeIdForTab: (tabId: string) => string | undefined
   /** Invalidate caches + notify the renderer so the new branch name surfaces. */
   onRenamed: (repoId: string) => void
-  /** The prompt a launch file carried for this pane, when its hook reports only the pointer. */
-  getLaunchFilePrompt?: (paneKey: string) => string | undefined
+  /** The prompt Orca shows for a pane's reported one (`launchPromptShownForPane`). */
+  showLaunchPrompt?: (paneKey: string, prompt: string) => string
 }
 
 // inFlight blocks concurrent generation; settled caches definitive verdicts (transient bails stay unsettled to retry later).
@@ -121,10 +121,10 @@ export async function maybeAutoRenameBranchOnFirstWork(
   // Why the pointer: a launch that carried its prompt in a file names only that file here, which
   // would name the branch after "read the task file". Its real prompt, when main kept it, names it
   // instead; otherwise (a worker brief) a later prompt of the user's own can.
-  const prompt =
-    hookPrompt && isLaunchFilePointer(hookPrompt)
-      ? deps.getLaunchFilePrompt?.(event.paneKey)?.trim()
-      : hookPrompt
+  const shown = hookPrompt
+    ? (deps.showLaunchPrompt?.(event.paneKey, hookPrompt) ?? hookPrompt).trim()
+    : undefined
+  const prompt = shown && !isLaunchFilePointer(shown) ? shown : undefined
   if (!prompt) {
     return
   }
diff --git a/src/main/agent-hooks/first-work-rename-runtime.ts b/src/main/agent-hooks/first-work-rename-runtime.ts
index 8c35b802bf6..82229db59a2 100644
--- a/src/main/agent-hooks/first-work-rename-runtime.ts
+++ b/src/main/agent-hooks/first-work-rename-runtime.ts
@@ -1,4 +1,4 @@
-import { launchFilePromptForPane } from './launch-file-prompt-by-pane'
+import { launchPromptShownForPane } from './launch-file-prompt-by-pane'
 import { existsSync } from 'node:fs'
 import { parseWorkspaceKey } from '../../shared/workspace-scope'
 import { getRepoIdFromWorktreeId } from '../../shared/worktree/id'
@@ -22,7 +22,7 @@ export function firstWorkRenameDeps(
   >
 ): FirstWorkBranchRenameDeps {
   return {
-    getLaunchFilePrompt: launchFilePromptForPane,
+    showLaunchPrompt: launchPromptShownForPane,
     getSettings: () => store.getSettings(),
     getRepo: (repoId) => store.getRepo(repoId),
     getAgentEnvResolvers: () => runtime.getCommitMessageAgentEnvironmentResolvers(),
diff --git a/src/main/agent-hooks/launch-file-prompt-by-pane.ts b/src/main/agent-hooks/launch-file-prompt-by-pane.ts
index 5229bc2d29c..8f881525580 100644
--- a/src/main/agent-hooks/launch-file-prompt-by-pane.ts
+++ b/src/main/agent-hooks/launch-file-prompt-by-pane.ts
@@ -1,4 +1,4 @@
-import type { LaunchFile } from '../../shared/launch-prompt-file'
+import { isLaunchFilePointer, type LaunchFile } from '../../shared/launch-prompt-file'
 
 // Why bounded: an entry outlives its launch until newer ones push it out; the rename reads only a
 // prompt's opening, and a pane's rename runs within minutes of its launch.
@@ -7,9 +7,9 @@ const KEPT_CHARS = 4_000
 const promptByPane = new Map<string, string>()
 
 /**
- * Keeps a launch file's prompt for the pane it launched, for the first-work rename: the agent's hook
- * reports only the pointer to the file. A sensitive file (a worker brief with its capability) is
- * never kept.
+ * Keeps a launch file's prompt for the pane it launched: the agent's hook reports only the pointer
+ * to the file, and Orca shows the prompt instead (`launchPromptShownForPane`). A sensitive file (a
+ * worker brief with its capability) is never kept.
  */
 export function rememberLaunchFilePrompt(
   paneKey: string,
@@ -28,6 +28,8 @@ export function rememberLaunchFilePrompt(
   }
 }
 
-export function launchFilePromptForPane(paneKey: string): string | undefined {
-  return promptByPane.get(paneKey)
+/** The one place Orca reads a pane's reported prompt: a launch file's pointer reads as the prompt
+ *  it points at, when this host kept it. */
+export function launchPromptShownForPane(paneKey: string, prompt: string): string {
+  return isLaunchFilePointer(prompt) ? (promptByPane.get(paneKey) ?? prompt) : prompt
 }
diff --git a/src/main/agent-hooks/server-launch-file-prompt.test.ts b/src/main/agent-hooks/server-launch-file-prompt.test.ts
new file mode 100644
index 00000000000..31fcfea9f1a
--- /dev/null
+++ b/src/main/agent-hooks/server-launch-file-prompt.test.ts
@@ -0,0 +1,39 @@
+import { describe, expect, it } from 'vitest'
+import { makePaneKey } from '../../shared/stable-pane-id'
+import { buildLaunchFilePointer, carryInLaunchFile } from '../../shared/launch-prompt-file'
+import { AgentHookServer } from './server'
+import { rememberLaunchFilePrompt } from './launch-file-prompt-by-pane'
+
+const LAUNCHED = makePaneKey('tab-launched', '55555555-5555-4555-8555-555555555555')
+const POINTER = buildLaunchFilePointer('/tmp/orca-launch-file-a1/task-context.md')
+
+describe('a status row whose agent reports its launch file’s pointer', () => {
+  // Why: the agent's own hook sees only the pointer sentence; every Orca reader of the row (status
+  // row, dashboard, phone) must see the prompt the user wrote.
+  it('shows the prompt the file carries', () => {
+    rememberLaunchFilePrompt(LAUNCHED, carryInLaunchFile('fix the flaky test', false).launchFile)
+    const server = new AgentHookServer()
+    server.ingestTerminalStatus({
+      paneKey: LAUNCHED,
+      tabId: 'tab-launched',
+      worktreeId: 'wt-1',
+      payload: { state: 'working', prompt: POINTER }
+    })
+    expect(server.getStatusSnapshot()).toEqual([
+      expect.objectContaining({ paneKey: LAUNCHED, prompt: 'fix the flaky test' })
+    ])
+  })
+
+  it('leaves a sensitive file’s pointer, whose content is never kept', () => {
+    const pane = makePaneKey('tab-sensitive', '66666666-6666-4666-8666-666666666666')
+    rememberLaunchFilePrompt(pane, carryInLaunchFile('token dcap_secret', true).launchFile)
+    const server = new AgentHookServer()
+    server.ingestTerminalStatus({
+      paneKey: pane,
+      tabId: 'tab-sensitive',
+      worktreeId: 'wt-1',
+      payload: { state: 'working', prompt: POINTER }
+    })
+    expect(server.getStatusSnapshot()).toEqual([expect.objectContaining({ prompt: POINTER })])
+  })
+})
diff --git a/src/main/agent-hooks/server/server-status-update.ts b/src/main/agent-hooks/server/server-status-update.ts
index ebb87c3c9c2..f9e2e6c7557 100644
--- a/src/main/agent-hooks/server/server-status-update.ts
+++ b/src/main/agent-hooks/server/server-status-update.ts
@@ -19,6 +19,7 @@ import {
 import { isStaleGrokTurnEnd } from './server-grok-status-rules'
 import { resolveCancelVerdictLatch } from './server-cancel-verdict-latch'
 import { AgentHookServerStatusApplication } from './server-status-application'
+import { launchPromptShownForPane } from '../launch-file-prompt-by-pane'
 
 export abstract class AgentHookServerStatusUpdate extends AgentHookServerStatusApplication {
   protected applyNormalizedStatus(
@@ -28,6 +29,12 @@ export abstract class AgentHookServerStatusUpdate extends AgentHookServerStatusA
     observedAt?: number,
     mutationBefore?: EnrichedAgentHookEventPayload
   ): EnrichedAgentHookEventPayload | undefined {
+    // Why here: every status row (local, relayed, replayed) passes through, so no reader of the
+    // prompt (status row, dashboard, mobile) shows a launch file's pointer instead of the prompt.
+    const prompt = launchPromptShownForPane(incoming.paneKey, incoming.payload.prompt)
+    if (prompt !== incoming.payload.prompt) {
+      incoming = { ...incoming, payload: { ...incoming.payload, prompt } }
+    }
     const transitioned = transitionHookPresence(
       incoming,
       this.state.lastStatusByPaneKey.get(incoming.paneKey)
diff --git a/src/main/agent-launch/__fixtures__/host-agent-startup-call-sites.txt b/src/main/agent-launch/__fixtures__/host-agent-startup-call-sites.txt
index c05556a2205..de078fd731e 100644
--- a/src/main/agent-launch/__fixtures__/host-agent-startup-call-sites.txt
+++ b/src/main/agent-launch/__fixtures__/host-agent-startup-call-sites.txt
@@ -4,7 +4,7 @@
 #   mobile-followup the phone's session-tab launch; attribution is a separate follow-up
 # The resolve-worktree-removal-target call also serves a bare typed command, which stays unattributed.
 src/main/runtime/orca-runtime-create-agent-session.ts 2 attributes
-src/main/runtime/orca-runtime-resolve-worktree-removal-target.ts 2 attributes
+src/main/runtime/orca-runtime-resolve-worktree-removal-target.ts 1 attributes
 src/main/runtime/runtime-worktree-agent-startup.ts 3 attributes
 src/main/runtime/orca-runtime-get-agent-session-execution-namespace.ts 1 resume
 src/main/runtime/orca-runtime-resolve-mobile-session-terminal-command.ts 1 mobile-followup
diff --git a/src/main/agent-launch/host-agent-startup-attribution.test.ts b/src/main/agent-launch/host-agent-startup-attribution.test.ts
index d9d15c5e61d..60851c0dd1e 100644
--- a/src/main/agent-launch/host-agent-startup-attribution.test.ts
+++ b/src/main/agent-launch/host-agent-startup-attribution.test.ts
@@ -22,9 +22,9 @@ const LISTED: ReadonlyMap<string, { calls: number; decision: string }> = new Map
     })
 )
 
-// planStartupWithLaunchPrompt wraps buildAgentStartupPlan in shared/, which this scan does not reach.
+// planLaunchPrompt plans every prompted launch; buildAgentStartupPlan only a prompt-less one.
 const BUILDER_CALL =
-  /\b(?:buildAgentStartupPlan|buildAgentDraftLaunchPlan|buildAgentResumeStartupPlan|planStartupWithLaunchPrompt)\s*\(/g
+  /\b(?:buildAgentStartupPlan|buildAgentDraftLaunchPlan|buildAgentResumeStartupPlan|planLaunchPrompt)\s*\(/g
 
 const DECIDE =
   'Decide attribution: a fresh agent the host builds must carry ' +
diff --git a/src/main/daemon/daemon-create-or-attach-request.ts b/src/main/daemon/daemon-create-or-attach-request.ts
index 653281f9c76..c74f2099152 100644
--- a/src/main/daemon/daemon-create-or-attach-request.ts
+++ b/src/main/daemon/daemon-create-or-attach-request.ts
@@ -21,7 +21,7 @@ export type CreateOrAttachRequest = {
     envToDelete?: string[]
     command?: string
     startupCommandDelivery?: StartupCommandDelivery
-    /** v39+: written by the daemon before it types `command`. */
+    /** v41+ (`LAUNCH_FILE_DAEMON_PROTOCOL_VERSION`): written by the daemon before it types `command`. */
     launchFile?: LaunchFile
     /** Where the daemon writes a WSL session's staged line and launch file; main resolves it. */
     wslLaunchDirectory?: WslLaunchDirectory
diff --git a/src/main/daemon/daemon-foreground-confirmation-protocol.test.ts b/src/main/daemon/daemon-foreground-confirmation-protocol.test.ts
index 3d87fa463af..b1e76a88b32 100644
--- a/src/main/daemon/daemon-foreground-confirmation-protocol.test.ts
+++ b/src/main/daemon/daemon-foreground-confirmation-protocol.test.ts
@@ -3,7 +3,7 @@ import { PREVIOUS_DAEMON_PROTOCOL_VERSIONS, PROTOCOL_VERSION } from './types'
 
 describe('foreground-confirmation daemon protocol', () => {
   it('rejects daemons from before the fresh-confirmation RPC', () => {
-    expect(PROTOCOL_VERSION).toBe(40)
+    expect(PROTOCOL_VERSION).toBe(41)
     expect(PREVIOUS_DAEMON_PROTOCOL_VERSIONS).toContain(19)
     expect(PREVIOUS_DAEMON_PROTOCOL_VERSIONS).toContain(22)
     expect(PREVIOUS_DAEMON_PROTOCOL_VERSIONS).toContain(23)
@@ -23,5 +23,6 @@ describe('foreground-confirmation daemon protocol', () => {
     expect(PREVIOUS_DAEMON_PROTOCOL_VERSIONS).toContain(37)
     expect(PREVIOUS_DAEMON_PROTOCOL_VERSIONS).toContain(38)
     expect(PREVIOUS_DAEMON_PROTOCOL_VERSIONS).toContain(39)
+    expect(PREVIOUS_DAEMON_PROTOCOL_VERSIONS).toContain(40)
   })
 })
diff --git a/src/main/daemon/daemon-launch-file-protocol.test.ts b/src/main/daemon/daemon-launch-file-protocol.test.ts
new file mode 100644
index 00000000000..a15373cbe9e
--- /dev/null
+++ b/src/main/daemon/daemon-launch-file-protocol.test.ts
@@ -0,0 +1,78 @@
+import './mock-descendant-sweep'
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
+import { rmSync } from 'node:fs'
+import { DaemonClient } from './client'
+import { DaemonPtyAdapter } from './daemon-pty-adapter'
+import { LAUNCH_FILE_DAEMON_PROTOCOL_VERSION, PROTOCOL_VERSION } from './daemon-protocol-version'
+import type { DaemonServer } from './daemon-server'
+import { createMockSubprocess, startDaemonAdapterHarness } from './daemon-pty-adapter-test-harness'
+import { carryInLaunchFile } from '../../shared/launch-prompt-file'
+
+describe('a launch file sent to a terminal daemon', () => {
+  let dir: string
+  let socketPath: string
+  let tokenPath: string
+  let server: DaemonServer
+  let adapter: DaemonPtyAdapter
+
+  beforeEach(async () => {
+    const harness = await startDaemonAdapterHarness(() => createMockSubprocess())
+    ;({ dir, socketPath, tokenPath, server, adapter } = harness)
+  })
+
+  afterEach(async () => {
+    vi.restoreAllMocks()
+    adapter?.dispose()
+    await server?.shutdown()
+    rmSync(dir, { recursive: true, force: true })
+  })
+
+  it('is a protocol the current daemon speaks', () => {
+    expect(PROTOCOL_VERSION).toBeGreaterThanOrEqual(LAUNCH_FILE_DAEMON_PROTOCOL_VERSION)
+  })
+
+  // Why: an older daemon drops `launchFile` and would type a line naming a file nobody wrote.
+  it('is refused, with the reason, by a daemon older than launch files', async () => {
+    vi.spyOn(DaemonClient.prototype, 'ensureConnected').mockResolvedValue()
+    const request = vi.spyOn(DaemonClient.prototype, 'request')
+    const legacy = new DaemonPtyAdapter({
+      socketPath,
+      tokenPath,
+      protocolVersion: LAUNCH_FILE_DAEMON_PROTOCOL_VERSION - 1
+    })
+    const { prompt, launchFile } = carryInLaunchFile('the whole task', false)
+    try {
+      await expect(
+        legacy.spawn({
+          sessionId: 'legacy-launch',
+          cols: 80,
+          rows: 24,
+          command: `claude '${prompt}'`,
+          launchAgent: 'claude',
+          launchFile
+        })
+      ).rejects.toThrow(/launch_file_unavailable/)
+      expect(request).not.toHaveBeenCalledWith(
+        'createOrAttach',
+        expect.anything(),
+        expect.anything()
+      )
+    } finally {
+      legacy.dispose()
+    }
+  })
+
+  it('is sent to the current daemon, which writes it', async () => {
+    const request = vi.spyOn(DaemonClient.prototype, 'request')
+    const { prompt, launchFile } = carryInLaunchFile('the whole task', false)
+    await adapter.spawn({
+      cols: 80,
+      rows: 24,
+      command: `claude '${prompt}'`,
+      launchAgent: 'claude',
+      launchFile
+    })
+    const createCall = request.mock.calls.find(([method]) => method === 'createOrAttach')
+    expect(createCall?.[1]).toMatchObject({ launchFile: { placeholder: launchFile.placeholder } })
+  })
+})
diff --git a/src/main/daemon/daemon-protocol-version.test.ts b/src/main/daemon/daemon-protocol-version.test.ts
index 1b77b8c20b5..aba47aeeea0 100644
--- a/src/main/daemon/daemon-protocol-version.test.ts
+++ b/src/main/daemon/daemon-protocol-version.test.ts
@@ -22,7 +22,7 @@ import {
 
 describe('daemon protocol version', () => {
   it('ships bounded history transfer after the 2031-unsubscribe fact', () => {
-    expect(PROTOCOL_VERSION).toBe(40)
+    expect(PROTOCOL_VERSION).toBe(41)
     expect(COLOR_QUERY_REPLY_COLORS_DAEMON_PROTOCOL_VERSION).toBe(38)
     expect(CODEX_NO_DAEMON_SHELL_LAUNCH_DAEMON_PROTOCOL_VERSION).toBe(37)
     expect(CONTENT_ADDRESSED_SHELL_WRAPPER_DAEMON_PROTOCOL_VERSION).toBe(36)
@@ -38,7 +38,7 @@ describe('daemon protocol version', () => {
     expect(AGENT_SESSION_CLAIM_DAEMON_PROTOCOL_VERSION).toBe(26)
     expect(AGENT_SESSION_CREATE_OPERATION_DAEMON_PROTOCOL_VERSION).toBe(26)
     expect(PREVIOUS_DAEMON_PROTOCOL_VERSIONS).toEqual(
-      Array.from({ length: 39 }, (_, index) => index + 1)
+      Array.from({ length: 40 }, (_, index) => index + 1)
     )
   })
 
diff --git a/src/main/daemon/daemon-protocol-version.ts b/src/main/daemon/daemon-protocol-version.ts
index 6dfec710e8b..c5875e22821 100644
--- a/src/main/daemon/daemon-protocol-version.ts
+++ b/src/main/daemon/daemon-protocol-version.ts
@@ -1,6 +1,8 @@
 // Why: daemons survive app updates, so wire behavior must be version-gated.
-// v40 stages long startup commands as sourced scripts; older owners stay attachable.
-export const PROTOCOL_VERSION = 40
+// v41 writes an agent's launch file before it types the line naming it; older owners stay attachable.
+export const PROTOCOL_VERSION = 41
+// Why: an older daemon drops `launchFile` and would type a line naming a file nobody wrote.
+export const LAUNCH_FILE_DAEMON_PROTOCOL_VERSION = 41
 // Why: older daemons reject `setColorQueryReplyColors` as an unknown request type.
 export const COLOR_QUERY_REPLY_COLORS_DAEMON_PROTOCOL_VERSION = 38
 export const CODEX_NO_DAEMON_SHELL_LAUNCH_DAEMON_PROTOCOL_VERSION = 37
@@ -33,7 +35,7 @@ export const CLEAN_DISCONNECT_PROTOCOL_VERSION = 24
 export const MODE_2031_UNSUBSCRIBE_FACT_PROTOCOL_VERSION = 29
 export const PREVIOUS_DAEMON_PROTOCOL_VERSIONS = [
   1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27,
-  28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39
+  28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40
 ] as const
 
 export function supportsColorQueryReplyColors(protocolVersion: number): boolean {
diff --git a/src/main/daemon/daemon-pty-session-spawn.ts b/src/main/daemon/daemon-pty-session-spawn.ts
index bade2bcc3e2..b40907a9b72 100644
--- a/src/main/daemon/daemon-pty-session-spawn.ts
+++ b/src/main/daemon/daemon-pty-session-spawn.ts
@@ -18,7 +18,7 @@ import { getRecoveredHistorySeedSegments } from './terminal-history-seed-segment
 import { AGENT_SESSION_CLAIM_DAEMON_PROTOCOL_VERSION, type CreateOrAttachResult } from './types'
 import { normalizeWslColdRestoreCwd } from './wsl-cold-restore-cwd'
 import { resolveWslSessionContext } from './wsl-session-context'
-import { resolveWslLaunchDirectory } from '../providers/wsl-launch-directory-resolution'
+import { resolveSpawnWslLaunchDirectory } from '../providers/wsl-launch-directory-resolution'
 import { resolveSafePtyDefaultCwd } from '../providers/pty-default-cwd'
 import { resolveUnixShellPath } from '../providers/local-pty-utils'
 import type { PtySpawnOptions, PtySpawnResult } from '../providers/types'
@@ -194,8 +194,8 @@ export abstract class DaemonPtySessionSpawn extends DaemonPtySpawnResult {
     }
 
     await this.ensureConnected()
-    const wslLaunchDirectory =
-      !attachOnly && opts.command ? await resolveWslLaunchDirectory(wslDistro) : undefined
+    const wslProbe = resolveSpawnWslLaunchDirectory(wslDistro, opts)
+    const wslLaunchDirectory = wslProbe ? await wslProbe : undefined
     // Pins the daemon's spawn to the distro the daemon writes the line and file into.
     if (wslLaunchDirectory) {
       opts = { ...opts, wslLaunchDirectory, terminalWindowsWslDistro: wslLaunchDirectory.distro }
diff --git a/src/main/daemon/daemon-pty-spawn-request.ts b/src/main/daemon/daemon-pty-spawn-request.ts
index 22222c8b206..f453741535f 100644
--- a/src/main/daemon/daemon-pty-spawn-request.ts
+++ b/src/main/daemon/daemon-pty-spawn-request.ts
@@ -7,7 +7,11 @@ import {
   type SnapshotCheckpointResult
 } from './daemon-pty-runtime-state'
 import { isDaemonGoneError } from './daemon-endpoint-errors'
-import { HISTORY_SEED_TRANSFER_PROTOCOL_VERSION } from './daemon-protocol-version'
+import {
+  HISTORY_SEED_TRANSFER_PROTOCOL_VERSION,
+  LAUNCH_FILE_DAEMON_PROTOCOL_VERSION
+} from './daemon-protocol-version'
+import { LaunchFileUnavailableError } from '../../shared/launch-file-writing'
 import type { ColdRestoreInfo } from './history-reader'
 import { NdjsonLineTooLongError } from './ndjson'
 import {
@@ -106,6 +110,15 @@ export abstract class DaemonPtySpawnRequest extends DaemonPtyRuntimeState {
       if (opts.signal?.aborted) {
         throw new Error('client_disconnected')
       }
+      // Why refuse: an older daemon (a same-id respawn routed to it) would drop the file and type
+      // a line naming it. The refusal reaches the user with the prompt to copy.
+      if (
+        !context.attachOnly &&
+        opts.launchFile &&
+        this.protocolVersion < LAUNCH_FILE_DAEMON_PROTOCOL_VERSION
+      ) {
+        throw new LaunchFileUnavailableError("this terminal's daemon is from an older Orca")
+      }
       const payload = {
         sessionId: context.sessionId,
         cols: context.effectiveCols,
diff --git a/src/main/daemon/daemon-pty-startup-delivery.test.ts b/src/main/daemon/daemon-pty-startup-delivery.test.ts
index b5ae04c5d1e..eb0b8e36654 100644
--- a/src/main/daemon/daemon-pty-startup-delivery.test.ts
+++ b/src/main/daemon/daemon-pty-startup-delivery.test.ts
@@ -9,7 +9,7 @@ import {
   type DaemonAdapterHarness,
   type SpawnSubprocess
 } from './daemon-pty-adapter-test-harness'
-import { buildLaunchFilePointer, planLaunchPrompt } from '../../shared/launch-prompt-file'
+import { buildLaunchFilePointer, carryInLaunchFile } from '../../shared/launch-prompt-file'
 
 const itOnPosix = process.platform === 'win32' ? it.skip : it
 
@@ -141,7 +141,7 @@ describe('DaemonPtyAdapter startup delivery', () => {
     vi.stubEnv('TMPDIR', stagingDir)
     nextShellPath = '/bin/zsh'
     try {
-      const { prompt, launchFile } = planLaunchPrompt('secret brief', { sensitive: true })
+      const { prompt, launchFile } = carryInLaunchFile('secret brief', true)
       await adapter.spawn({
         cols: 80,
         rows: 24,
diff --git a/src/main/daemon/terminal-host-wsl-launch-directory.test.ts b/src/main/daemon/terminal-host-wsl-launch-directory.test.ts
index 2b395205216..6e6354f385b 100644
--- a/src/main/daemon/terminal-host-wsl-launch-directory.test.ts
+++ b/src/main/daemon/terminal-host-wsl-launch-directory.test.ts
@@ -5,7 +5,7 @@ import { afterEach, beforeEach, describe, expect, it, vi, type Mock } from 'vite
 import { createMockSubprocess } from './daemon-pty-adapter-test-harness'
 import { TerminalHost } from './terminal-host'
 import type { TerminalHostOptions } from './terminal-host-options'
-import { buildLaunchFilePointer, planLaunchPrompt } from '../../shared/launch-prompt-file'
+import { buildLaunchFilePointer, carryInLaunchFile } from '../../shared/launch-prompt-file'
 import type { WslLaunchDirectory } from '../../shared/wsl-launch-directory'
 
 vi.mock('../pty-descendant-termination', () => ({ killWithDescendantSweep: vi.fn() }))
@@ -33,7 +33,7 @@ describe('a daemon WSL session with a launch file', () => {
   })
 
   function create(wslLaunchDirectory: WslLaunchDirectory | undefined) {
-    const { prompt, launchFile } = planLaunchPrompt('secret brief', { sensitive: true })
+    const { prompt, launchFile } = carryInLaunchFile('secret brief', true)
     return host.createOrAttach({
       sessionId: 'wsl-session',
       cols: 80,
diff --git a/src/main/ipc/pty/ipc/spawn-launch-refusal.test.ts b/src/main/ipc/pty/ipc/spawn-launch-refusal.test.ts
new file mode 100644
index 00000000000..659986dd26e
--- /dev/null
+++ b/src/main/ipc/pty/ipc/spawn-launch-refusal.test.ts
@@ -0,0 +1,39 @@
+import { describe, expect, it, vi } from 'vitest'
+import { describeLaunchFileUnavailable } from '../../../../shared/launch-prompt-file'
+import { takeTerminalLaunchRefusal } from '../../../runtime/terminal-launch-refusals'
+import type { PtySpawnIpcArgs, PtySpawnIpcDeps } from './spawn-types'
+
+const mocks = vi.hoisted(() => ({
+  handlers: new Map<string, (event: unknown, args: unknown) => Promise<unknown>>(),
+  runSpawn: vi.fn()
+}))
+
+vi.mock('../../pty-host-bindings', () => ({
+  getPtyIpc: () => ({
+    handle: (channel: string, handler: (event: unknown, args: unknown) => Promise<unknown>) =>
+      mocks.handlers.set(channel, handler)
+  })
+}))
+vi.mock('./spawn-run', () => ({ runPtyIpcSpawn: mocks.runSpawn }))
+
+import { installPtySpawnIpcHandler } from './spawn'
+
+describe('a pane spawn refused for what carries its prompt', () => {
+  // oxlint-disable-next-line typescript/consistent-type-assertions -- SAFETY: the handler reads only getLocalPtyStartupPromise before delegating to the mocked runner.
+  installPtySpawnIpcHandler({ getLocalPtyStartupPromise: () => null } as unknown as PtySpawnIpcDeps)
+  const spawn = (args: Partial<PtySpawnIpcArgs>): Promise<unknown> =>
+    mocks.handlers.get('pty:spawn')!({}, { cols: 80, rows: 24, ...args })
+
+  it('records the refusal for its tab, for a create waiting on that tab’s handle', async () => {
+    const refusal = describeLaunchFileUnavailable('disk full')
+    mocks.runSpawn.mockRejectedValueOnce(new Error(refusal))
+    await expect(spawn({ tabId: 'tab-refused' })).rejects.toThrow(refusal)
+    expect(takeTerminalLaunchRefusal('tab-refused')).toBe(refusal)
+  })
+
+  it('records nothing for any other spawn failure', async () => {
+    mocks.runSpawn.mockRejectedValueOnce(new Error('spawn ENOENT'))
+    await expect(spawn({ tabId: 'tab-other' })).rejects.toThrow('spawn ENOENT')
+    expect(takeTerminalLaunchRefusal('tab-other')).toBeUndefined()
+  })
+})
diff --git a/src/main/ipc/pty/ipc/spawn-options-launch-file.test.ts b/src/main/ipc/pty/ipc/spawn-options-launch-file.test.ts
index 77e41293405..4656ee4f765 100644
--- a/src/main/ipc/pty/ipc/spawn-options-launch-file.test.ts
+++ b/src/main/ipc/pty/ipc/spawn-options-launch-file.test.ts
@@ -1,7 +1,8 @@
 import { describe, expect, it, vi } from 'vitest'
 import { getDefaultSettings } from '../../../../shared/constants'
 import { buildPtyIpcSpawnOptions } from './spawn-options'
-import { launchFilePromptForPane } from '../../../agent-hooks/launch-file-prompt-by-pane'
+import { launchPromptShownForPane } from '../../../agent-hooks/launch-file-prompt-by-pane'
+import { buildLaunchFilePointer } from '../../../../shared/launch-prompt-file'
 import { createPtyIpcSpawnState } from './spawn-state'
 import type { PtySpawnIpcArgs, PtySpawnIpcDeps } from './spawn-types'
 
@@ -59,8 +60,8 @@ describe('renderer pty spawn: launch file', () => {
     expect(options.launchFile).toBeUndefined()
   })
 
-  // Why: the agent's hook reports only the pointer, so the first-work rename reads this instead.
-  it('keeps the prompt a launch file carries for its pane, never a sensitive one', async () => {
+  // Why: the agent's hook reports only the pointer; Orca shows the prompt it points at instead.
+  it('shows the prompt a launch file carries for its pane, never a sensitive one', async () => {
     await spawnOptionsFor(
       { cols: 80, rows: 24, command, launchFile: LAUNCH_FILE },
       command,
@@ -76,7 +77,9 @@ describe('renderer pty spawn: launch file', () => {
       command,
       'tab-b:leaf-b'
     )
-    expect(launchFilePromptForPane('tab-a:leaf-a')).toBe('the whole task')
-    expect(launchFilePromptForPane('tab-b:leaf-b')).toBeUndefined()
+    const pointer = buildLaunchFilePointer('/tmp/orca-launch-file-a1/task-context.md')
+    expect(launchPromptShownForPane('tab-a:leaf-a', pointer)).toBe('the whole task')
+    expect(launchPromptShownForPane('tab-b:leaf-b', pointer)).toBe(pointer)
+    expect(launchPromptShownForPane('tab-a:leaf-a', 'a later prompt')).toBe('a later prompt')
   })
 })
diff --git a/src/main/ipc/pty/ipc/spawn.ts b/src/main/ipc/pty/ipc/spawn.ts
index 6fd5183a36d..e35b40ee637 100644
--- a/src/main/ipc/pty/ipc/spawn.ts
+++ b/src/main/ipc/pty/ipc/spawn.ts
@@ -1,6 +1,8 @@
 import { getPtyIpc } from '../../pty-host-bindings'
 import { runPtyIpcSpawn } from './spawn-run'
 import type { PtySpawnIpcArgs, PtySpawnIpcDeps } from './spawn-types'
+import { isLaunchFileUnavailableMessage } from '../../../../shared/launch-prompt-file'
+import { recordTerminalLaunchRefusal } from '../../../runtime/terminal-launch-refusals'
 
 export function installPtySpawnIpcHandler(deps: PtySpawnIpcDeps): void {
   const ipcMain = getPtyIpc()
@@ -11,6 +13,15 @@ export function installPtySpawnIpcHandler(deps: PtySpawnIpcDeps): void {
     if (startupPromise) {
       await startupPromise
     }
-    return runPtyIpcSpawn(deps, args)
+    try {
+      return await runPtyIpcSpawn(deps, args)
+    } catch (error) {
+      const message = error instanceof Error ? error.message : String(error)
+      // Why: a runtime create waits on this tab's handle; the refusal is the answer it waits for.
+      if (args.tabId && isLaunchFileUnavailableMessage(message)) {
+        recordTerminalLaunchRefusal(args.tabId, message)
+      }
+      throw error
+    }
   })
 }
diff --git a/src/main/providers/local-pty-provider-shell-readiness.test.ts b/src/main/providers/local-pty-provider-shell-readiness.test.ts
index 6b1c0b64614..141dde0ff94 100644
--- a/src/main/providers/local-pty-provider-shell-readiness.test.ts
+++ b/src/main/providers/local-pty-provider-shell-readiness.test.ts
@@ -118,7 +118,7 @@ vi.mock('../shell-prompt-readiness-probe', () => ({
 
 import { LocalPtyProvider } from './local-pty-provider'
 import { POSIX_SHELL_STARTUP_COMMAND_ENV } from '../pty/posix-shell-startup-command'
-import { buildLaunchFilePointer, planLaunchPrompt } from '../../shared/launch-prompt-file'
+import { buildLaunchFilePointer, carryInLaunchFile } from '../../shared/launch-prompt-file'
 import {
   applyLocalPtyProviderMockDefaults,
   createLocalPtyMockProcess,
@@ -273,7 +273,7 @@ describe('LocalPtyProvider', () => {
       vi.useFakeTimers()
       try {
         process.env.SHELL = '/bin/sh'
-        const { prompt, launchFile } = planLaunchPrompt('secret brief', { sensitive: true })
+        const { prompt, launchFile } = carryInLaunchFile('secret brief', true)
 
         await provider.spawn({ cols: 80, rows: 24, command: `claude '${prompt}'`, launchFile })
 
diff --git a/src/main/providers/local-pty-spawn-wsl-shutdown-order.test.ts b/src/main/providers/local-pty-spawn-wsl-shutdown-order.test.ts
index dab27b9967e..528d6d6ab2f 100644
--- a/src/main/providers/local-pty-spawn-wsl-shutdown-order.test.ts
+++ b/src/main/providers/local-pty-spawn-wsl-shutdown-order.test.ts
@@ -4,11 +4,11 @@ import type * as SpawnStateModule from './local-pty-spawn-state'
 import { ptyShutdownOperations, type PtyShutdownOperation } from './local-pty-provider-state'
 import type { WslLaunchDirectory } from '../../shared/wsl-launch-directory'
 
-const { resolveWslLaunchDirectory, reattachLocalPty } = vi.hoisted(() => ({
-  resolveWslLaunchDirectory: vi.fn(),
+const { resolveSpawnWslLaunchDirectory, reattachLocalPty } = vi.hoisted(() => ({
+  resolveSpawnWslLaunchDirectory: vi.fn(),
   reattachLocalPty: vi.fn()
 }))
-vi.mock('./wsl-launch-directory-resolution', () => ({ resolveWslLaunchDirectory }))
+vi.mock('./wsl-launch-directory-resolution', () => ({ resolveSpawnWslLaunchDirectory }))
 vi.mock('./local-pty-launch-plan', async (importOriginal) => ({
   ...(await importOriginal<typeof LaunchPlanModule>()),
   resolveLocalPtyWslDistro: () => 'Ubuntu'
@@ -30,7 +30,7 @@ describe('a WSL respawn racing a shutdown of the same session id', () => {
   // starts during it would otherwise be missed and race the new PTY.
   it('waits for a shutdown that began while the distro directory was being resolved', async () => {
     let resolveDirectory!: (directory: WslLaunchDirectory | undefined) => void
-    resolveWslLaunchDirectory.mockReturnValue(
+    resolveSpawnWslLaunchDirectory.mockReturnValue(
       new Promise((resolve) => {
         resolveDirectory = resolve
       })
@@ -40,7 +40,9 @@ describe('a WSL respawn racing a shutdown of the same session id', () => {
       { cols: 80, rows: 24, sessionId: 'wsl-session', command: 'claude' },
       () => ({})
     )
-    await vi.waitFor(() => expect(resolveWslLaunchDirectory).toHaveBeenCalledWith('Ubuntu'))
+    await vi.waitFor(() =>
+      expect(resolveSpawnWslLaunchDirectory).toHaveBeenCalledWith('Ubuntu', expect.anything())
+    )
 
     let finishShutdown!: () => void
     const shutdown: PtyShutdownOperation = {
diff --git a/src/main/providers/local-pty-spawn.ts b/src/main/providers/local-pty-spawn.ts
index 48739c4ba3f..d7604864075 100644
--- a/src/main/providers/local-pty-spawn.ts
+++ b/src/main/providers/local-pty-spawn.ts
@@ -30,7 +30,7 @@ import {
   writeSpawnLaunchFile,
   type WrittenLaunchFile
 } from '../../shared/launch-file-writing'
-import { resolveWslLaunchDirectory } from './wsl-launch-directory-resolution'
+import { resolveSpawnWslLaunchDirectory } from './wsl-launch-directory-resolution'
 
 export async function spawnLocalPty(
   args: PtySpawnOptions,
@@ -40,7 +40,8 @@ export async function spawnLocalPty(
   const wslDistro =
     args.command && !args.attachOnly ? resolveLocalPtyWslDistro(args, getOptions) : undefined
   // Why before the shutdown check: no await may sit between it and this spawn's registration.
-  const wslLaunchDirectory = wslDistro ? await resolveWslLaunchDirectory(wslDistro) : undefined
+  const wslProbe = resolveSpawnWslLaunchDirectory(wslDistro, args)
+  const wslLaunchDirectory = wslProbe ? await wslProbe : undefined
   if (reattachId) {
     const pendingShutdown = ptyShutdownOperations.get(reattachId)
     if (pendingShutdown) {
diff --git a/src/main/providers/ssh-plain-shell-pty-provider.test.ts b/src/main/providers/ssh-plain-shell-pty-provider.test.ts
index 064f37a5cef..72a73c8db46 100644
--- a/src/main/providers/ssh-plain-shell-pty-provider.test.ts
+++ b/src/main/providers/ssh-plain-shell-pty-provider.test.ts
@@ -2,6 +2,7 @@ import { EventEmitter } from 'node:events'
 import { describe, expect, it, vi } from 'vitest'
 import type { ClientChannel, PseudoTtyOptions } from 'ssh2'
 import { SshPlainShellPtyProvider } from './ssh-plain-shell-pty-provider'
+import { carryInLaunchFile } from '../../shared/launch-prompt-file'
 
 class FakeShellChannel extends EventEmitter {
   readonly stderr = new EventEmitter()
@@ -147,6 +148,11 @@ describe('SshPlainShellPtyProvider', () => {
     await expect(provider.spawn({ cols: 80, rows: 24, attachOnly: true })).rejects.toThrow(
       '(no_runtime)'
     )
+    // Why: nothing here writes a launch file, so its line would name a file that is not there.
+    const { prompt, launchFile } = carryInLaunchFile('the whole task', false)
+    await expect(
+      provider.spawn({ cols: 80, rows: 24, command: `claude '${prompt}'`, launchFile })
+    ).rejects.toThrow('needs the Orca remote server')
     expect(opened).toEqual([])
   })
 })
diff --git a/src/main/providers/ssh-plain-shell-pty-provider.ts b/src/main/providers/ssh-plain-shell-pty-provider.ts
index c0fbe34372a..5c4af1da90f 100644
--- a/src/main/providers/ssh-plain-shell-pty-provider.ts
+++ b/src/main/providers/ssh-plain-shell-pty-provider.ts
@@ -72,7 +72,13 @@ export class SshPlainShellPtyProvider implements IPtyProvider {
     if (this.disposed) {
       throw new Error('SSH connection is not active')
     }
-    if (opts.agentSessionEnsure || opts.agentSessionCreateOperationId || opts.launchAgent) {
+    // Why launchFile too: no host here writes it, so the line would name a file that is not there.
+    if (
+      opts.agentSessionEnsure ||
+      opts.agentSessionCreateOperationId ||
+      opts.launchAgent ||
+      opts.launchFile
+    ) {
       throw new PlainSshUnsupportedError('Launching agents', this.mode)
     }
     if (opts.attachOnly) {
diff --git a/src/main/providers/wsl-launch-directory-resolution.test.ts b/src/main/providers/wsl-launch-directory-resolution.test.ts
index be4fb992a50..78d4c18e510 100644
--- a/src/main/providers/wsl-launch-directory-resolution.test.ts
+++ b/src/main/providers/wsl-launch-directory-resolution.test.ts
@@ -2,7 +2,11 @@ import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
 import { createLocalPtyLaunchPlan, resolveLocalPtyWslDistro } from './local-pty-launch-plan'
 import type { LocalPtyProviderOptions } from './local-pty-provider-types'
 import type { PtySpawnOptions } from './types'
-import { resolveWslLaunchDirectory } from './wsl-launch-directory-resolution'
+import {
+  resolveSpawnWslLaunchDirectory,
+  resolveWslLaunchDirectory,
+  WSL_LAUNCH_DIRECTORY_FAILURE_TTL_MS
+} from './wsl-launch-directory-resolution'
 import type * as WslModule from '../wsl'
 
 const { runWslProcess } = vi.hoisted(() => ({ runWslProcess: vi.fn() }))
@@ -57,13 +61,41 @@ describe('resolveWslLaunchDirectory', () => {
     )
   })
 
-  it('probes a distro once, and asks again only after a failure', async () => {
-    runWslProcess.mockResolvedValueOnce(answers('', 1))
-    await expect(resolveWslLaunchDirectory('Arch')).resolves.toBeUndefined()
+  it('probes a distro once, and remembers a failure only for its TTL', async () => {
+    vi.useFakeTimers()
+    try {
+      runWslProcess.mockResolvedValueOnce(answers('', 1))
+      await expect(resolveWslLaunchDirectory('Arch')).resolves.toBeUndefined()
+      runWslProcess.mockResolvedValue(answers('/home/ada\n/bin/bash\n'))
+      // Why: a broken distro must not cost every spawn a wsl.exe round trip.
+      await expect(resolveWslLaunchDirectory('Arch')).resolves.toBeUndefined()
+      expect(runWslProcess).toHaveBeenCalledTimes(1)
+      await vi.advanceTimersByTimeAsync(WSL_LAUNCH_DIRECTORY_FAILURE_TTL_MS)
+      await expect(resolveWslLaunchDirectory('Arch')).resolves.toMatchObject({
+        shell: '/bin/bash'
+      })
+      await resolveWslLaunchDirectory('Arch')
+      expect(runWslProcess).toHaveBeenCalledTimes(2)
+    } finally {
+      vi.useRealTimers()
+    }
+  })
+
+  it('probes only for a spawn that needs the directory', async () => {
     runWslProcess.mockResolvedValue(answers('/home/ada\n/bin/bash\n'))
-    await expect(resolveWslLaunchDirectory('Arch')).resolves.toMatchObject({ shell: '/bin/bash' })
-    await resolveWslLaunchDirectory('Arch')
-    expect(runWslProcess).toHaveBeenCalledTimes(2)
+    expect(
+      resolveSpawnWslLaunchDirectory('Gentoo', { command: 'claude', launchAgent: 'claude' })
+    ).toBeUndefined()
+    expect(
+      resolveSpawnWslLaunchDirectory(undefined, { command: 'x'.repeat(600), launchAgent: 'claude' })
+    ).toBeUndefined()
+    await expect(
+      resolveSpawnWslLaunchDirectory('Gentoo', {
+        command: `claude '${'x'.repeat(600)}'`,
+        launchAgent: 'claude'
+      })
+    ).resolves.toMatchObject({ distro: 'Gentoo' })
+    expect(runWslProcess).toHaveBeenCalledTimes(1)
   })
 
   it('has none for a spawn outside WSL or a distro whose home cannot be read', async () => {
diff --git a/src/main/providers/wsl-launch-directory-resolution.ts b/src/main/providers/wsl-launch-directory-resolution.ts
index f357a627d72..87cf5c10cac 100644
--- a/src/main/providers/wsl-launch-directory-resolution.ts
+++ b/src/main/providers/wsl-launch-directory-resolution.ts
@@ -1,6 +1,9 @@
 import { posix } from 'node:path'
 import { quotePosixShell, RESOLVE_WSL_LOGIN_SHELL } from '../../shared/wsl-login-shell-command'
 import type { WslLaunchDirectory } from '../../shared/wsl-launch-directory'
+import type { LaunchFile } from '../../shared/launch-prompt-file'
+import { spawnNeedsWslLaunchDirectory } from '../../shared/launch-file-writing'
+import type { TuiAgent } from '../../shared/tui-agent'
 import { toWindowsWslUncPath } from '../../shared/wsl-paths'
 import { runWslProcess } from '../wsl/wsl-runner'
 
@@ -8,11 +11,36 @@ const ORCA_CACHE_RELATIVE = '.cache/orca'
 
 const probes = new Map<string, Promise<WslLaunchDirectory | undefined>>()
 
+/** A failed probe is answered from memory this long, so a broken distro does not cost every spawn
+ *  a `wsl.exe` round trip, yet a distro that comes back is found again. */
+export const WSL_LAUNCH_DIRECTORY_FAILURE_TTL_MS = 30_000
+
+/** The directory for a spawn that needs one (`spawnNeedsWslLaunchDirectory`); no probe, and no
+ *  await for the caller, otherwise. */
+export function resolveSpawnWslLaunchDirectory(
+  distro: string | null | undefined,
+  spawn: {
+    command?: string
+    launchFile?: LaunchFile
+    launchAgent?: TuiAgent
+    attachOnly?: boolean
+  }
+): Promise<WslLaunchDirectory | undefined> | undefined {
+  const needed =
+    Boolean(distro) &&
+    spawn.attachOnly !== true &&
+    spawnNeedsWslLaunchDirectory({
+      ...spawn,
+      orcaBuiltLine: spawn.launchAgent !== undefined
+    })
+  return needed ? resolveWslLaunchDirectory(distro) : undefined
+}
+
 /**
  * The distro directory a WSL spawn's staged line and launch file are written to, and the login
  * shell the pane runs, which decides how a staged line is sourced; undefined when the distro cannot
- * be asked, and the write site then refuses a launch that needs one. Probed once per distro, and
- * only a success is kept.
+ * be asked, and the write site then refuses a launch that needs one. Probed once per distro; a
+ * success is kept, a failure for `WSL_LAUNCH_DIRECTORY_FAILURE_TTL_MS`.
  */
 export async function resolveWslLaunchDirectory(
   distro: string | null | undefined
@@ -24,7 +52,16 @@ export async function resolveWslLaunchDirectory(
   if (!probe) {
     probe = probeWslLaunchDirectory(distro)
     probes.set(distro, probe)
-    void probe.then((found) => found ?? probes.delete(distro))
+    const pending = probe
+    void pending.then((found) => {
+      if (!found) {
+        setTimeout(() => {
+          if (probes.get(distro) === pending) {
+            probes.delete(distro)
+          }
+        }, WSL_LAUNCH_DIRECTORY_FAILURE_TTL_MS).unref?.()
+      }
+    })
   }
   return await probe
 }
diff --git a/src/main/runtime/agent-terminal-startup-prompt.test.ts b/src/main/runtime/agent-terminal-startup-prompt.test.ts
index 065db00d3fc..222f9c4bd38 100644
--- a/src/main/runtime/agent-terminal-startup-prompt.test.ts
+++ b/src/main/runtime/agent-terminal-startup-prompt.test.ts
@@ -2,9 +2,9 @@
  * The launch command a terminal create builds when the launch hands it a prompt.
  *
  * This is the argv half of terminal prompt delivery, and the reason it is worth pinning is that
- * its failure mode is silent: `buildAgentStartupPlan` answers a prompt it cannot fold by returning
- * a bare command plus a `followupPrompt`, and this resolver returns options, not a live PTY, so a
- * dropped `followupPrompt` would spawn the agent with no prompt and no error anywhere.
+ * its failure mode is silent: `planLaunchPrompt` answers a prompt it cannot fold with a clean plan
+ * and `paste-after-ready`, and this resolver returns options, not a live PTY, so a dropped paste
+ * would spawn the agent with no prompt and no error anywhere.
  */
 
 import { describe, expect, it, vi } from 'vitest'
diff --git a/src/main/runtime/orca-runtime-create-agent-session.ts b/src/main/runtime/orca-runtime-create-agent-session.ts
index 91fa61c85b9..4523e483d50 100644
--- a/src/main/runtime/orca-runtime-create-agent-session.ts
+++ b/src/main/runtime/orca-runtime-create-agent-session.ts
@@ -20,9 +20,12 @@ import { resolveAgentStartupPlanInputs } from '../../shared/agent-startup-plan-i
 import {
   agentPromptRidesLaunchCommand,
   buildAgentDraftLaunchPlan,
-  buildAgentStartupPlan
+  planLaunchPrompt
 } from '../../shared/tui-agent-startup'
-import { windowsDraftRefusal } from '../../shared/startup-plan-launch-file'
+import {
+  launchPromptNeedsPasteRefusal,
+  windowsDraftRefusal
+} from '../../shared/launch-prompt-carry'
 import { resolveStartupShell } from '../../shared/tui-agent-startup-shell'
 import type { RuntimeTerminalCreate } from '../../shared/runtime-types'
 import type {
@@ -165,31 +168,39 @@ export class OrcaRuntimeWithCreateAgentSession extends OrcaRuntimeWithGetAgentSe
         ...(request.agentArgs !== undefined ? { agentArgs: request.agentArgs } : {}),
         sessionOptions: this.toAgentSessionOptions(request.launchPreferences)
       })
-      const startup =
-        request.promptDelivery === 'draft'
-          ? buildAgentDraftLaunchPlan({ ...startupArgs, draft: request.prompt ?? '' })
-          : buildAgentStartupPlan({
-              ...startupArgs,
-              prompt: request.prompt ?? '',
-              allowEmptyPromptLaunch: true
-            })
-      if (!startup) {
-        const refusal =
-          request.promptDelivery === 'draft'
-            ? windowsDraftRefusal(
-                request.prompt ?? '',
-                resolveStartupShell(startupArgs.platform, startupArgs.shell)
-              )
-            : null
-        throw new Error(refusal ?? 'agent_session_identity_required')
-      }
-      if (startup.followupPrompt && agentPromptRidesLaunchCommand(request.agent)) {
-        // Why: this create has no paste after ready, so a prompt left for one would be dropped.
-        throw new Error(
-          `${request.agent} cannot take this prompt on its command line here (too long, private, or ` +
-            "broken by the host's Windows shell), so the session was not started. Start it without " +
-            'the prompt and paste the prompt once it opens.'
-        )
+      let startup
+      let launchFile
+      if (request.promptDelivery === 'draft') {
+        startup = buildAgentDraftLaunchPlan({ ...startupArgs, draft: request.prompt ?? '' })
+        if (!startup) {
+          const refusal = windowsDraftRefusal(
+            request.prompt ?? '',
+            resolveStartupShell(startupArgs.platform, startupArgs.shell)
+          )
+          throw new Error(refusal ?? 'agent_session_identity_required')
+        }
+      } else {
+        const planned = planLaunchPrompt({ ...startupArgs, prompt: request.prompt ?? '' })
+        if (!planned) {
+          throw new Error('agent_session_identity_required')
+        }
+        switch (planned.carry) {
+          case 'none':
+          case 'on-line':
+            startup = planned.plan
+            break
+          case 'launch-file':
+            startup = planned.plan
+            launchFile = planned.launchFile
+            break
+          case 'paste-after-ready':
+            // Why: a stdin agent's prompt is the session's to submit; an argv agent's would be dropped.
+            if (agentPromptRidesLaunchCommand(request.agent)) {
+              throw new Error(launchPromptNeedsPasteRefusal(request.agent, 'session'))
+            }
+            startup = planned.cleanPlan
+            break
+        }
       }
       if (caller.signal?.aborted) {
         throw new Error('client_disconnected')
@@ -222,9 +233,7 @@ export class OrcaRuntimeWithCreateAgentSession extends OrcaRuntimeWithGetAgentSe
           launchAgent: request.agent,
           terminalKittyKeyboardProtocol: request.terminalKittyKeyboardProtocol,
           startupCommandDelivery: startup.startupCommandDelivery,
-          ...('launchFile' in startup && startup.launchFile
-            ? { launchFile: startup.launchFile }
-            : {}),
+          ...(launchFile ? { launchFile } : {}),
           // A fresh agent this host built; the request has no surface field, so it counts as `unknown`.
           telemetry: agentStartedTelemetry(request.agent, undefined),
           cwd: startupCwd,
diff --git a/src/main/runtime/orca-runtime-create-terminal-desktop.ts b/src/main/runtime/orca-runtime-create-terminal-desktop.ts
index 54f5104f0ac..1bcf2d346bd 100644
--- a/src/main/runtime/orca-runtime-create-terminal-desktop.ts
+++ b/src/main/runtime/orca-runtime-create-terminal-desktop.ts
@@ -2,7 +2,6 @@
 import * as dependencies from './orca-runtime-create-terminal-dependencies'
 import type { OrcaRuntimeWithCreateTerminal } from './orca-runtime-create-terminal'
 import type { RuntimeTerminalPresentation } from '../../shared/runtime-types'
-import { removeLaunchFile, writeLaunchFile } from '../../shared/launch-file-writing'
 
 export async function createDesktopTerminal(
   runtime: OrcaRuntimeWithCreateTerminal,
@@ -26,11 +25,6 @@ export async function createDesktopTerminal(
       `--shell ${opts.shellOverride} needs a workspace, because the shell is resolved on the workspace's execution host. No terminal was created.`
     )
   }
-  if (launchOpts.launchFile && !workspace?.connectionId) {
-    // Why: a pane whose host refuses the launch-file write never registers a handle, so this create
-    // would only time out; failing the same local write here reports the refusal at once.
-    removeLaunchFile(writeLaunchFile({ launchFile: launchOpts.launchFile }))
-  }
   const worktreeId = workspace?.id
   const cwd = workspace
     ? runtime.resolveWorkspaceTerminalStartupCwd(workspace, launchOpts.cwd)
diff --git a/src/main/runtime/orca-runtime-resolve-mobile-session-terminal-command.ts b/src/main/runtime/orca-runtime-resolve-mobile-session-terminal-command.ts
index 54ad4104fd6..26ff9f9d0f9 100644
--- a/src/main/runtime/orca-runtime-resolve-mobile-session-terminal-command.ts
+++ b/src/main/runtime/orca-runtime-resolve-mobile-session-terminal-command.ts
@@ -5,7 +5,8 @@ import type { WorktreeStartupLaunch } from '../../shared/worktree/launch-types'
 import type { TuiAgent } from '../../shared/tui-agent'
 import type { SleepingAgentLaunchConfig } from '../../shared/agent-session-resume'
 import { isTuiAgentEnabled } from '../../shared/tui-agent-selection'
-import { buildAgentStartupPlan } from '../../shared/tui-agent-startup'
+import { planLaunchPrompt } from '../../shared/tui-agent-startup'
+import { launchPromptNeedsPasteRefusal } from '../../shared/launch-prompt-carry'
 import { resolveAgentStartupPlanInputs } from '../../shared/agent-startup-plan-inputs'
 
 export class OrcaRuntimeWithResolveMobileSessionTerminalCommand extends OrcaRuntimeWithRunCreateMobileSessionTerminal {
@@ -47,7 +48,7 @@ export class OrcaRuntimeWithResolveMobileSessionTerminalCommand extends OrcaRunt
     if (!isTuiAgentEnabled(opts.agent, settings.disabledTuiAgents)) {
       throw new Error('Selected agent is disabled. Choose an enabled agent before creating.')
     }
-    const startupPlan = buildAgentStartupPlan({
+    const planned = planLaunchPrompt({
       ...resolveAgentStartupPlanInputs({
         agent: opts.agent,
         settings,
@@ -56,14 +57,25 @@ export class OrcaRuntimeWithResolveMobileSessionTerminalCommand extends OrcaRunt
         // Why: SSH runs the CLI through the relay shim (plain `orca`), so the Linux-only `orca-ide` rename must not apply.
         isRemote: Boolean(workspace.connectionId)
       }),
-      prompt: opts.agentPrompt ?? '',
-      allowEmptyPromptLaunch: true
+      prompt: opts.agentPrompt ?? ''
     })
-    if (!startupPlan) {
+    if (!planned) {
       throw new Error(`Could not build launch command for ${opts.agent}.`)
     }
-    if (opts.agentPrompt && startupPlan.followupPrompt) {
-      throw new Error(`Agent ${opts.agent} does not support startup prompt quick commands.`)
+    let startupPlan
+    let launchFile
+    switch (planned.carry) {
+      case 'none':
+      case 'on-line':
+        startupPlan = planned.plan
+        break
+      case 'launch-file':
+        startupPlan = planned.plan
+        launchFile = planned.launchFile
+        break
+      case 'paste-after-ready':
+        // Why: a quick command has no paste after ready, so the prompt would be dropped.
+        throw new Error(launchPromptNeedsPasteRefusal(opts.agent, 'terminal'))
     }
     return {
       command: startupPlan.launchCommand,
@@ -74,7 +86,7 @@ export class OrcaRuntimeWithResolveMobileSessionTerminalCommand extends OrcaRunt
       launchConfig: startupPlan.launchConfig,
       launchAgent: opts.agent,
       startupCommandDelivery: startupPlan.startupCommandDelivery,
-      ...(startupPlan.launchFile ? { launchFile: startupPlan.launchFile } : {})
+      ...(launchFile ? { launchFile } : {})
     }
   }
 }
diff --git a/src/main/runtime/orca-runtime-resolve-worktree-removal-target.ts b/src/main/runtime/orca-runtime-resolve-worktree-removal-target.ts
index e6ac7d11748..01f5c061420 100644
--- a/src/main/runtime/orca-runtime-resolve-worktree-removal-target.ts
+++ b/src/main/runtime/orca-runtime-resolve-worktree-removal-target.ts
@@ -19,14 +19,8 @@ import { terminalShellOverrideRefusal } from './terminal-shell-override-host-sup
 import { resolveTerminalStartupCwd } from '../../shared/terminal-startup-cwd'
 import { resolveLocalProjectRuntimeForWorktreeId } from '../local-project-runtime-resolution'
 import { resolveBareAgentLaunchCommand } from './runtime-agent-launch-resolution'
-import {
-  agentPromptRidesLaunchCommand,
-  buildAgentStartupPlan
-} from '../../shared/tui-agent-startup'
-import {
-  planStartupWithLaunchPrompt,
-  type LaunchPromptStartupPlan
-} from '../../shared/startup-line-prompt-carry'
+import { agentPromptRidesLaunchCommand, planLaunchPrompt } from '../../shared/tui-agent-startup'
+import { launchPromptNeedsPasteRefusal } from '../../shared/launch-prompt-carry'
 import { resolveAgentStartupPlanInputs } from '../../shared/agent-startup-plan-inputs'
 import { agentStartedTelemetry } from '../agent-launch/agent-started-telemetry'
 import { LOCAL_EXECUTION_HOST_ID } from '../../shared/execution-host'
@@ -277,21 +271,12 @@ export class OrcaRuntimeWithResolveWorktreeRemovalTarget extends OrcaRuntimeWith
       sessionOptions: this.toAgentSessionOptions(opts.launchPreferences)
     })
     // A caller that wrote its own launch file already passes the pointer to it as the prompt.
-    const prompt = opts.startupPrompt ?? ''
-    let planned: LaunchPromptStartupPlan
-    if (opts.launchFile) {
-      const plan = buildAgentStartupPlan({
-        ...planInputs,
-        prompt,
-        allowEmptyPromptLaunch: true,
-        launchFile: opts.launchFile
-      })
-      planned = { plan, launchFile: plan?.launchFile ?? opts.launchFile }
-    } else {
-      planned = planStartupWithLaunchPrompt(planInputs, prompt)
-    }
-    const { plan: startupPlan, launchFile } = planned
-    if (!startupPlan) {
+    const planned = planLaunchPrompt({
+      ...planInputs,
+      prompt: opts.startupPrompt ?? '',
+      ...(opts.launchFile ? { launchFile: opts.launchFile } : {})
+    })
+    if (!planned) {
       // Why: an explicit agent that yields no plan would otherwise spawn a bare
       // shell that never reaches agent readiness.
       if (opts.startupAgent) {
@@ -299,16 +284,29 @@ export class OrcaRuntimeWithResolveWorktreeRemovalTarget extends OrcaRuntimeWith
       }
       return opts
     }
-    if (opts.startupPrompt?.trim()) {
-      const carried = startupPlan.followupPrompt === null
-      if (!carried && !opts.onStartupPromptCarry) {
-        // Why: this create returns options, not a live PTY, so the prompt would be dropped.
-        throw new Error(
-          `${agent} cannot take this prompt on its command line here, so no terminal was created. ` +
-            'Start the agent without it and paste the prompt once it opens.'
-        )
-      }
-      opts.onStartupPromptCarry?.(carried)
+    let startupPlan
+    let launchFile
+    switch (planned.carry) {
+      case 'none':
+        startupPlan = planned.plan
+        break
+      case 'on-line':
+        startupPlan = planned.plan
+        opts.onStartupPromptCarry?.(true)
+        break
+      case 'launch-file':
+        startupPlan = planned.plan
+        launchFile = planned.launchFile
+        opts.onStartupPromptCarry?.(true)
+        break
+      case 'paste-after-ready':
+        if (!opts.onStartupPromptCarry) {
+          // Why: this create returns options, not a live PTY, so the prompt would be dropped.
+          throw new Error(launchPromptNeedsPasteRefusal(agent, 'terminal'))
+        }
+        startupPlan = planned.cleanPlan
+        opts.onStartupPromptCarry(false)
+        break
     }
 
     return {
diff --git a/src/main/runtime/orca-runtime-restore-live-paired-renderer-session-owned-mobile-terminals.ts b/src/main/runtime/orca-runtime-restore-live-paired-renderer-session-owned-mobile-terminals.ts
index 8e9ad65dc32..d540e6d497c 100644
--- a/src/main/runtime/orca-runtime-restore-live-paired-renderer-session-owned-mobile-terminals.ts
+++ b/src/main/runtime/orca-runtime-restore-live-paired-renderer-session-owned-mobile-terminals.ts
@@ -2,6 +2,7 @@
 import { OrcaRuntimeWithWaitForMobileTerminalSurface } from './orca-runtime-wait-for-mobile-terminal-surface'
 import { runtimeWorktreeIdsEqual } from './runtime-worktree-path-identity'
 import { parsePaneKey } from '../../shared/stable-pane-id'
+import { onTerminalLaunchRefusal, takeTerminalLaunchRefusal } from './terminal-launch-refusals'
 import type { RuntimePtyWorktreeRecord } from './runtime-terminal-state-records'
 import type { RuntimeMobileSessionCreateTerminalResult } from '../../shared/runtime-types'
 
@@ -137,26 +138,43 @@ export class OrcaRuntimeWithRestoreLivePairedRendererSessionOwnedMobileTerminals
     }
 
     return new Promise<string>((resolve, reject) => {
-      const timer = setTimeout(() => {
+      const stop = (): void => {
+        clearTimeout(timer)
+        stopRefusal()
         const idx = this.graphSyncCallbacks.indexOf(check)
         if (idx !== -1) {
           this.graphSyncCallbacks.splice(idx, 1)
         }
+      }
+      const timer = setTimeout(() => {
+        stop()
         reject(new Error('Timed out waiting for terminal handle after creation'))
       }, timeoutMs)
+      // Why: a host that refused the spawn's launch file never registers a handle.
+      const refuse = (message: string): void => {
+        stop()
+        reject(new Error(message))
+      }
+      const stopRefusal = onTerminalLaunchRefusal((refusedTabId) => {
+        const message = refusedTabId === tabId ? takeTerminalLaunchRefusal(tabId) : undefined
+        if (message) {
+          refuse(message)
+        }
+      })
 
       const check = (): void => {
         const handle = this.resolveHandleForTab(tabId)
         if (handle) {
-          clearTimeout(timer)
-          const idx = this.graphSyncCallbacks.indexOf(check)
-          if (idx !== -1) {
-            this.graphSyncCallbacks.splice(idx, 1)
-          }
+          stop()
           resolve(handle)
         }
       }
       this.graphSyncCallbacks.push(check)
+      const recorded = takeTerminalLaunchRefusal(tabId)
+      if (recorded) {
+        refuse(recorded)
+        return
+      }
       // Why: graph sync may have fired between the initial check and registration; re-check to avoid a missed wake-up.
       check()
     })
diff --git a/src/main/runtime/orca-runtime-tests/mobile-session-tabs-part-12.spec.ts b/src/main/runtime/orca-runtime-tests/mobile-session-tabs-part-12.spec.ts
index 92b29a8b449..28c9cc0dd2e 100644
--- a/src/main/runtime/orca-runtime-tests/mobile-session-tabs-part-12.spec.ts
+++ b/src/main/runtime/orca-runtime-tests/mobile-session-tabs-part-12.spec.ts
@@ -34,7 +34,7 @@ describe('OrcaRuntimeService', () => {
         agent: 'aider',
         agentPrompt: 'Review this diff'
       })
-    ).rejects.toThrow('does not support startup prompt quick commands')
+    ).rejects.toThrow('aider takes its prompt only after it starts')
     expect(spawn).not.toHaveBeenCalled()
   })
 
diff --git a/src/main/runtime/runtime-terminal-contracts.ts b/src/main/runtime/runtime-terminal-contracts.ts
index 1c4d22826b6..1af1b767335 100644
--- a/src/main/runtime/runtime-terminal-contracts.ts
+++ b/src/main/runtime/runtime-terminal-contracts.ts
@@ -46,10 +46,10 @@ export type TerminalCreateOptions = {
    * Initial text for `startupAgent`'s launch command, for an agent whose CLI takes a prompt
    * argument. Not a general prompt channel: an agent that takes its text only after start has no
    * launch command to carry it, and a caller that sets this for one is refused rather than having
-   * the prompt silently dropped. It rides that command, directly or as a pointer to a launch file
-   * (`startup-line-prompt-carry`), unless it needs a launch file the agent is not known to read:
-   * then the agent starts clean, `onStartupPromptCarry` says so, and the caller pastes it once
-   * ready. A caller with no such callback is refused instead.
+   * the prompt silently dropped. It rides that command, directly or as a pointer to a launch file,
+   * unless `carryLaunchPrompt` says it must be pasted: then the agent starts clean,
+   * `onStartupPromptCarry` says so, and the caller pastes it once ready. A caller with no such
+   * callback is refused instead.
    */
   startupPrompt?: string
   /** Main-internal: whether `startupPrompt` rode the launch command. Called once the plan is built. */
diff --git a/src/main/runtime/runtime-worktree-agent-startup.ts b/src/main/runtime/runtime-worktree-agent-startup.ts
index a8565e594e9..0e859f1641c 100644
--- a/src/main/runtime/runtime-worktree-agent-startup.ts
+++ b/src/main/runtime/runtime-worktree-agent-startup.ts
@@ -8,8 +8,13 @@ import { getRepoSshConnectionId } from '../../shared/execution-host'
 import { isTuiAgent } from '../../shared/tui-agent-config'
 import { isTuiAgentEnabled, pickTuiAgent } from '../../shared/tui-agent-selection'
 import { resolveAgentStartupPlanInputs } from '../../shared/agent-startup-plan-inputs'
-import { buildAgentDraftLaunchPlan, buildAgentStartupPlan } from '../../shared/tui-agent-startup'
-import { planStartupWithLaunchPrompt } from '../../shared/startup-line-prompt-carry'
+import {
+  buildAgentDraftLaunchPlan,
+  buildAgentStartupPlan,
+  planLaunchPrompt,
+  type AgentStartupPlan
+} from '../../shared/tui-agent-startup'
+import type { LaunchFile } from '../../shared/launch-prompt-file'
 import {
   detectInstalledAgentsWithShellPathHydration,
   detectRemoteAgents
@@ -149,15 +154,31 @@ export function buildWorktreeStartupForAgent(
     ...(environment.agentArgs !== undefined ? { agentArgs: environment.agentArgs } : {}),
     sessionOptions: environment.toSessionOptions(environment.launchPreferences)
   })
-  const { plan: startupPlan, launchFile } = planStartupWithLaunchPrompt(
-    planInputs,
-    environment.prompt ?? ''
-  )
-  if (!startupPlan) {
+  const planned = planLaunchPrompt({ ...planInputs, prompt: environment.prompt ?? '' })
+  if (!planned) {
     throw new Error(`Could not build launch command for ${agent}.`)
   }
-  if (environment.onPromptCarry && environment.prompt?.trim()) {
-    environment.onPromptCarry(startupPlan.followupPrompt === null)
+  let startupPlan: AgentStartupPlan
+  let launchFile: LaunchFile | undefined
+  let followup: WorktreeStartupFollowup | undefined
+  switch (planned.carry) {
+    case 'none':
+      startupPlan = planned.plan
+      break
+    case 'on-line':
+    case 'launch-file':
+      startupPlan = planned.plan
+      launchFile = planned.carry === 'launch-file' ? planned.launchFile : undefined
+      environment.onPromptCarry?.(true)
+      break
+    case 'paste-after-ready':
+      startupPlan = planned.cleanPlan
+      if (environment.onPromptCarry) {
+        environment.onPromptCarry(false)
+      } else {
+        followup = { expectedProcess: startupPlan.expectedProcess, prompt: planned.text }
+      }
+      break
   }
   return {
     agent,
@@ -171,13 +192,6 @@ export function buildWorktreeStartupForAgent(
       ...(launchFile ? { launchFile } : {}),
       telemetry: agentStartedTelemetry(agent, environment.launchSource)
     },
-    ...(startupPlan.followupPrompt && !environment.onPromptCarry
-      ? {
-          followup: {
-            expectedProcess: startupPlan.expectedProcess,
-            prompt: startupPlan.followupPrompt
-          }
-        }
-      : {})
+    ...(followup ? { followup } : {})
   }
 }
diff --git a/src/main/runtime/terminal-launch-refusals.test.ts b/src/main/runtime/terminal-launch-refusals.test.ts
new file mode 100644
index 00000000000..07c10e96e7b
--- /dev/null
+++ b/src/main/runtime/terminal-launch-refusals.test.ts
@@ -0,0 +1,46 @@
+import { describe, expect, it, vi } from 'vitest'
+import { OrcaRuntimeService } from './orca-runtime'
+import { recordTerminalLaunchRefusal, takeTerminalLaunchRefusal } from './terminal-launch-refusals'
+import { describeLaunchFileUnavailable } from '../../shared/launch-prompt-file'
+
+vi.mock('electron', () => ({
+  BrowserWindow: { fromId: vi.fn(() => null) },
+  webContents: { fromId: vi.fn(() => null) },
+  ipcMain: { on: vi.fn(), removeListener: vi.fn() },
+  app: { getPath: vi.fn(() => '/tmp') }
+}))
+
+function waitForHandle(runtime: OrcaRuntimeService, tabId: string): Promise<string> {
+  // oxlint-disable-next-line typescript/consistent-type-assertions -- SAFETY: the wait under test is a protected member of the runtime; the assertion names only that method.
+  const internal = runtime as unknown as {
+    waitForTerminalHandle: (tabId: string, timeoutMs?: number) => Promise<string>
+  }
+  return internal.waitForTerminalHandle(tabId, 5_000)
+}
+
+describe('a pane spawn the host refused for its launch file', () => {
+  const refusal = describeLaunchFileUnavailable(
+    "the WSL distro's home directory could not be reached"
+  )
+
+  it('is taken once, by the tab it refused', () => {
+    recordTerminalLaunchRefusal('tab-taken', refusal)
+    expect(takeTerminalLaunchRefusal('tab-other')).toBeUndefined()
+    expect(takeTerminalLaunchRefusal('tab-taken')).toBe(refusal)
+    expect(takeTerminalLaunchRefusal('tab-taken')).toBeUndefined()
+  })
+
+  // Why: agent.launch waits on this handle; the refusal is its answer, not a 10 s timeout.
+  it('ends a create waiting on that tab’s handle with the reason', async () => {
+    const runtime = new OrcaRuntimeService()
+    const waiting = waitForHandle(runtime, 'tab-waiting')
+    recordTerminalLaunchRefusal('tab-waiting', refusal)
+    await expect(waiting).rejects.toThrow(refusal)
+  })
+
+  it('ends a create that starts waiting after the refusal arrived', async () => {
+    const runtime = new OrcaRuntimeService()
+    recordTerminalLaunchRefusal('tab-early', refusal)
+    await expect(waitForHandle(runtime, 'tab-early')).rejects.toThrow(refusal)
+  })
+})
diff --git a/src/main/runtime/terminal-launch-refusals.ts b/src/main/runtime/terminal-launch-refusals.ts
new file mode 100644
index 00000000000..9ec35d90534
--- /dev/null
+++ b/src/main/runtime/terminal-launch-refusals.ts
@@ -0,0 +1,37 @@
+/**
+ * A pane spawn its host refused because what carries the agent's prompt (a launch file, or the
+ * staged script of a long line) could not be written. Kept by tab so a create waiting on that tab's
+ * handle reports the reason at once instead of timing out. Entries expire; nothing waits forever.
+ */
+const REFUSAL_TTL_MS = 60_000
+const MAX_REFUSALS = 64
+
+const refusals = new Map<string, { message: string; at: number }>()
+const listeners = new Set<(tabId: string, message: string) => void>()
+
+export function recordTerminalLaunchRefusal(tabId: string, message: string): void {
+  const now = Date.now()
+  for (const [key, entry] of refusals) {
+    if (now - entry.at > REFUSAL_TTL_MS || refusals.size >= MAX_REFUSALS) {
+      refusals.delete(key)
+    }
+  }
+  refusals.set(tabId, { message, at: now })
+  for (const listener of listeners) {
+    listener(tabId, message)
+  }
+}
+
+/** The refusal recorded for `tabId`, consumed. */
+export function takeTerminalLaunchRefusal(tabId: string): string | undefined {
+  const entry = refusals.get(tabId)
+  refusals.delete(tabId)
+  return entry && Date.now() - entry.at <= REFUSAL_TTL_MS ? entry.message : undefined
+}
+
+export function onTerminalLaunchRefusal(
+  listener: (tabId: string, message: string) => void
+): () => void {
+  listeners.add(listener)
+  return () => listeners.delete(listener)
+}
diff --git a/src/relay/pty-handler-startup-command-staging.test.ts b/src/relay/pty-handler-startup-command-staging.test.ts
index 0d750d5a69f..a3ab9150bcb 100644
--- a/src/relay/pty-handler-startup-command-staging.test.ts
+++ b/src/relay/pty-handler-startup-command-staging.test.ts
@@ -34,7 +34,7 @@ vi.mock('../main/shell-prompt-readiness-probe', () => ({
 
 import type { PtyHandler } from './pty-handler'
 import { beginPtyHandlerTest, endPtyHandlerTest } from './pty-handler-test-harness'
-import { buildLaunchFilePointer, planLaunchPrompt } from '../shared/launch-prompt-file'
+import { buildLaunchFilePointer, carryInLaunchFile } from '../shared/launch-prompt-file'
 import type { MockDispatcher } from './pty-handler-test-harness'
 
 const describePosix = process.platform === 'win32' ? describe.skip : describe
@@ -114,7 +114,7 @@ describePosix('relay startup command staging', () => {
   })
 
   it('writes a launch file before typing the line that names it, and removes it on exit', async () => {
-    const { prompt, launchFile } = planLaunchPrompt('secret brief', { sensitive: true })
+    const { prompt, launchFile } = carryInLaunchFile('secret brief', true)
     await dispatcher.callRequest('pty.spawn', {
       command: `claude '${prompt}'`,
       commandDelivery: 'provider',
diff --git a/src/renderer/src/components/right-sidebar/buildSourceControlAgentDeliveryPlan.ts b/src/renderer/src/components/right-sidebar/buildSourceControlAgentDeliveryPlan.ts
index d8320dd1075..c5307266a25 100644
--- a/src/renderer/src/components/right-sidebar/buildSourceControlAgentDeliveryPlan.ts
+++ b/src/renderer/src/components/right-sidebar/buildSourceControlAgentDeliveryPlan.ts
@@ -16,6 +16,7 @@ type BuildSourceControlAgentDeliveryPlanArgs = {
   /** Why: keep the previewed command label in sync with the real remote launch,
    * which omits the Linux-only `orca-ide` rename for SSH hosts. */
   isRemote?: boolean
+  launchHostIsPaired?: boolean
 }
 
 export function buildSourceControlAgentDeliveryPlan({
@@ -26,7 +27,8 @@ export function buildSourceControlAgentDeliveryPlan({
   detectedAgents,
   connectionUnavailable,
   launchPlatform,
-  isRemote
+  isRemote,
+  launchHostIsPaired
 }: BuildSourceControlAgentDeliveryPlanArgs): SourceControlAgentActionDeliveryPlanState {
   if (connectionUnavailable) {
     return buildSourceControlAgentConnectionErrorPlan()
@@ -50,7 +52,8 @@ export function buildSourceControlAgentDeliveryPlan({
     cmdOverrides: settings?.agentCmdOverrides,
     terminalWindowsShell: settings?.terminalWindowsShell,
     platform: launchPlatform,
-    isRemote
+    isRemote,
+    launchHostIsPaired
   })
   if (!result.ok) {
     return { status: 'error', error: result.error }
diff --git a/src/renderer/src/components/right-sidebar/useSourceControlAgentActionStart.ts b/src/renderer/src/components/right-sidebar/useSourceControlAgentActionStart.ts
index 2209c1621fe..85fe301b820 100644
--- a/src/renderer/src/components/right-sidebar/useSourceControlAgentActionStart.ts
+++ b/src/renderer/src/components/right-sidebar/useSourceControlAgentActionStart.ts
@@ -9,6 +9,9 @@ import type { GlobalSettings } from '../../../../shared/global-settings-types'
 import type { Repo } from '../../../../shared/repo-types'
 import type { TuiAgent } from '../../../../shared/tui-agent'
 import { buildSourceControlAgentDeliveryPlan } from './buildSourceControlAgentDeliveryPlan'
+import { useAppStore } from '@/store'
+import { launchHostIsPaired } from '@/lib/launch-file-host'
+import { getRuntimeEnvironmentIdForWorktree } from '@/lib/worktree-runtime-owner'
 import type { SourceControlAgentActionDeliveryPlanState } from './SourceControlAgentActionDialogForm'
 import { runSourceControlAgentActionStart } from './runSourceControlAgentActionStart'
 import { buildSourceControlAgentConnectionErrorPlan } from './source-control-agent-action-dialog-support'
@@ -116,7 +119,10 @@ export function useSourceControlAgentActionStart({
         detectedAgents: currentDetectedAgents,
         connectionUnavailable,
         launchPlatform,
-        isRemote
+        isRemote,
+        launchHostIsPaired: launchHostIsPaired(
+          getRuntimeEnvironmentIdForWorktree(useAppStore.getState(), worktreeId)
+        )
       })
     },
     [
@@ -128,7 +134,8 @@ export function useSourceControlAgentActionStart({
       refreshDetectedAgents,
       selectedAgent,
       launchPlatform,
-      isRemote
+      isRemote,
+      worktreeId
     ]
   )
 
diff --git a/src/renderer/src/components/sidebar/folder-workspace-agent-startup.ts b/src/renderer/src/components/sidebar/folder-workspace-agent-startup.ts
index d3746a38e82..9c9240011dc 100644
--- a/src/renderer/src/components/sidebar/folder-workspace-agent-startup.ts
+++ b/src/renderer/src/components/sidebar/folder-workspace-agent-startup.ts
@@ -62,7 +62,6 @@ export function buildFolderWorkspaceLinkedStartupPlan(args: {
       agent: draftLaunchPlan.agent,
       launchCommand: draftLaunchPlan.launchCommand,
       expectedProcess: draftLaunchPlan.expectedProcess,
-      followupPrompt: null,
       launchConfig: draftLaunchPlan.launchConfig,
       ...(draftLaunchPlan.sessionOptions ? { sessionOptions: draftLaunchPlan.sessionOptions } : {}),
       ...(draftLaunchPlan.startupCommandDelivery
diff --git a/src/renderer/src/components/sidebar/folder-workspace-composer-submit.test.ts b/src/renderer/src/components/sidebar/folder-workspace-composer-submit.test.ts
index 6976647ed8b..3b9ebe5cd1e 100644
--- a/src/renderer/src/components/sidebar/folder-workspace-composer-submit.test.ts
+++ b/src/renderer/src/components/sidebar/folder-workspace-composer-submit.test.ts
@@ -311,7 +311,6 @@ describe('submitFolderWorkspaceCreate', () => {
       startup: expect.objectContaining({
         agent: 'codex',
         launchCommand: 'codex',
-        followupPrompt: null,
         draftPrompt: `Review this before starting\n\n${linkedWorkItem.url}`
       })
     })
@@ -387,7 +386,7 @@ describe('submitFolderWorkspaceCreate', () => {
       startup: expect.objectContaining({
         agent: 'aider',
         launchCommand: 'aider',
-        followupPrompt: 'Fix the failing folder prompt flow'
+        pastePromptAfterReady: 'Fix the failing folder prompt flow'
       })
     })
   })
diff --git a/src/renderer/src/components/sidebar/folder-workspace-composer-submit.ts b/src/renderer/src/components/sidebar/folder-workspace-composer-submit.ts
index 1712f42f0a7..d942af2701f 100644
--- a/src/renderer/src/components/sidebar/folder-workspace-composer-submit.ts
+++ b/src/renderer/src/components/sidebar/folder-workspace-composer-submit.ts
@@ -1,7 +1,7 @@
 import { ensureAgentStartupInTerminal, type LinkedWorkItemSummary } from '@/lib/new-workspace'
 import { seedNativeChatLaunchDraftForAgentTab } from '@/lib/agent-launch-prompt-delivery'
 import { createBrowserUuid } from '@/lib/browser-uuid'
-import { buildAgentStartupPlan } from '@/lib/tui-agent-startup'
+import { planLaunchPrompt } from '@/lib/tui-agent-startup'
 import { tuiAgentToAgentKind } from '@/lib/telemetry'
 import { activateAndRevealFolderWorkspace } from '@/lib/worktree-activation'
 import { isWorkItemLookupText } from '@/lib/work-item-lookup-text'
@@ -21,7 +21,11 @@ import { planAgentSessionLaunch } from '@/lib/agent-session-launch-plan'
 import { beginStructuredAgentSessionProvisionalLaunch } from '@/lib/structured-agent-session-provisional-tab'
 import { getNewWorkspaceProjectGroupHostId } from '@/lib/new-workspace-project-options'
 import { useAppStore } from '@/store'
-import { launchHostWritesLaunchFile } from '@/lib/launch-file-host'
+import { launchHostIsPaired } from '@/lib/launch-file-host'
+import {
+  composerAgentStartupPlan,
+  type ComposerAgentStartupPlan
+} from '@/lib/composer-agent-startup-plan'
 import {
   buildFolderWorkspaceLinkedStartupPlan,
   getFolderWorkspaceAgentLaunchPlatform,
@@ -99,7 +103,7 @@ export async function submitFolderWorkspaceCreate({
     isRemote: launchIsRemote,
     terminalWindowsShell
   })
-  const startupPlan =
+  const startupPlan: ComposerAgentStartupPlan | null =
     quickAgent && linkedWorkItem
       ? buildFolderWorkspaceLinkedStartupPlan({
           agent: quickAgent,
@@ -114,21 +118,21 @@ export async function submitFolderWorkspaceCreate({
           isRemote: launchIsRemote
         })
       : quickAgent
-        ? buildAgentStartupPlan({
-            agent: quickAgent,
-            prompt: note,
-            cmdOverrides: agentCmdOverrides ?? {},
-            agentArgs,
-            agentEnv,
-            sessionOptions,
-            platform: launchPlatform,
-            shell: launchShell,
-            isRemote: launchIsRemote,
-            hostWritesLaunchFile: launchHostWritesLaunchFile({
-              activeRuntimeEnvironmentId: runtimeEnvironmentId
+        ? composerAgentStartupPlan(
+            planLaunchPrompt({
+              agent: quickAgent,
+              prompt: note,
+              cmdOverrides: agentCmdOverrides ?? {},
+              agentArgs,
+              agentEnv,
+              sessionOptions,
+              platform: launchPlatform,
+              shell: launchShell,
+              isRemote: launchIsRemote,
+              launchHostIsPaired: launchHostIsPaired(runtimeEnvironmentId)
             }),
-            allowEmptyPromptLaunch: true
-          })
+            note
+          )
         : null
   // Why: the argv-prefill plan carries the draft inside `launchCommand`, so
   // `startupPlan.draftPrompt` alone can't tell whether this launch has one.
@@ -194,6 +198,7 @@ export async function submitFolderWorkspaceCreate({
             ? { startupCommandDelivery: startupPlan.startupCommandDelivery }
             : {}),
           ...(startupPlan.launchFile ? { launchFile: startupPlan.launchFile } : {}),
+          ...(startupPlan.launchPrompt ? { launchPrompt: startupPlan.launchPrompt } : {}),
           telemetry: {
             agent_kind: tuiAgentToAgentKind(quickAgent),
             launch_source: launchSource,
@@ -246,7 +251,7 @@ export async function submitFolderWorkspaceCreate({
     if (
       !structuredLaunchAccepted &&
       startupPlan &&
-      (startupPlan.followupPrompt || startupPlan.draftPrompt) &&
+      (startupPlan.pastePromptAfterReady || startupPlan.draftPrompt) &&
       activation !== false
     ) {
       void ensureAgentStartupInTerminal({
diff --git a/src/renderer/src/components/terminal-pane/ipc-pty-connect.ts b/src/renderer/src/components/terminal-pane/ipc-pty-connect.ts
index 15e857f239b..da09d847cfd 100644
--- a/src/renderer/src/components/terminal-pane/ipc-pty-connect.ts
+++ b/src/renderer/src/components/terminal-pane/ipc-pty-connect.ts
@@ -191,11 +191,13 @@ function handleConnectError(
   if (message.includes('was explicitly killed')) {
     return undefined
   }
-  // Why the transport's file: the spawn sends it unless the connect call named its own command.
-  const { launchFile } = context.transportOptions
-  if (launchFile && !options.command && isLaunchFileUnavailableMessage(message)) {
+  // Why the transport's prompt: the spawn carries it unless the connect call named its own command.
+  // The host refuses a launch file, or a WSL line it cannot stage, with the same code.
+  const { launchFile, launchPrompt } = context.transportOptions
+  const refusedPrompt = launchFile?.content ?? launchPrompt
+  if (refusedPrompt && !options.command && isLaunchFileUnavailableMessage(message)) {
     // The pane gets no shell; the notice holds the only copy of the prompt.
-    showAgentLaunchNotStartedNotice({ prompt: launchFile.content })
+    showAgentLaunchNotStartedNotice({ prompt: refusedPrompt })
   }
   if (connectionId && options.sessionId && message.includes(SSH_PTY_CONNECTION_MISMATCH_MARKER)) {
     // Why not `sessionExpired`: this string is minted by `toRelaySshPtyId`/`toAppSshPtyId` from a
diff --git a/src/renderer/src/components/terminal-pane/pty-connection-types.ts b/src/renderer/src/components/terminal-pane/pty-connection-types.ts
index 9fc0f528040..0f358a74cbe 100644
--- a/src/renderer/src/components/terminal-pane/pty-connection-types.ts
+++ b/src/renderer/src/components/terminal-pane/pty-connection-types.ts
@@ -27,6 +27,8 @@ export type PtyPaneStartup = {
   delivery?: 'terminal-paste'
   startupCommandDelivery?: StartupCommandDelivery
   launchFile?: LaunchFile
+  /** The submitted prompt `command` carries; handed back to copy if the host refuses the spawn. */
+  launchPrompt?: string
   env?: Record<string, string>
   envToDelete?: string[]
   launchConfig?: SleepingAgentLaunchConfig
diff --git a/src/renderer/src/components/terminal-pane/pty-connection/pane-startup-command-options.ts b/src/renderer/src/components/terminal-pane/pty-connection/pane-startup-command-options.ts
new file mode 100644
index 00000000000..d47777f7453
--- /dev/null
+++ b/src/renderer/src/components/terminal-pane/pty-connection/pane-startup-command-options.ts
@@ -0,0 +1,18 @@
+import type { ConnectPanePtySession } from './connect-pane-pty-session'
+
+/** The startup command a pane spawn sends, with the launch file and prompt it carries; none when
+ *  the startup is delivered by terminal paste instead. */
+export function paneStartupCommandOptions(session: ConnectPanePtySession): {
+  command?: string
+  launchFile?: NonNullable<ConnectPanePtySession['paneStartup']>['launchFile']
+  launchPrompt?: string
+} {
+  if (session.shouldDeliverStartupViaTerminalPaste) {
+    return {}
+  }
+  return {
+    command: session.paneStartup?.command,
+    launchFile: session.paneStartup?.launchFile,
+    launchPrompt: session.paneStartup?.launchPrompt
+  }
+}
diff --git a/src/renderer/src/components/terminal-pane/pty-connection/pty-input-recovery.ts b/src/renderer/src/components/terminal-pane/pty-connection/pty-input-recovery.ts
index 84b98317b54..1a48b5a7a2b 100644
--- a/src/renderer/src/components/terminal-pane/pty-connection/pty-input-recovery.ts
+++ b/src/renderer/src/components/terminal-pane/pty-connection/pty-input-recovery.ts
@@ -25,6 +25,7 @@ import { shouldRetainDisposedPaneSpawn } from './disposed-spawn-retention'
 import { buffersInputOnlyForSshReattach } from './ssh-reattach-input-buffering'
 
 import type { ConnectPanePtySession } from './connect-pane-pty-session'
+import { paneStartupCommandOptions } from './pane-startup-command-options'
 import { resolveTerminalInlineImagesEnabled } from '../../../../../shared/terminal-inline-images-settings'
 
 /** Transport creation, terminal capability replies, viewport claims, and undeliverable-input recovery. */
@@ -62,9 +63,7 @@ export function installPtyInputRecovery(session: ConnectPanePtySession): void {
       : {}),
     env: session.paneEnv,
     ...(session.paneStartup?.envToDelete ? { envToDelete: session.paneStartup.envToDelete } : {}),
-    ...(session.shouldDeliverStartupViaTerminalPaste
-      ? {}
-      : { command: session.paneStartup?.command, launchFile: session.paneStartup?.launchFile }),
+    ...paneStartupCommandOptions(session),
     ...(session.shouldUseProviderSshStartupDelivery
       ? { commandDelivery: 'provider' as const }
       : {}),
diff --git a/src/renderer/src/components/terminal-pane/pty-transport-launch-refusal.test.ts b/src/renderer/src/components/terminal-pane/pty-transport-launch-refusal.test.ts
new file mode 100644
index 00000000000..6bc07f350fa
--- /dev/null
+++ b/src/renderer/src/components/terminal-pane/pty-transport-launch-refusal.test.ts
@@ -0,0 +1,49 @@
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
+import { installIpcPtyWindow, restorePtySpecWindow } from './pty-transport-test-harness'
+import { describeLaunchFileUnavailable } from '../../../../shared/launch-prompt-file'
+
+const mocks = vi.hoisted(() => ({ showNotStarted: vi.fn() }))
+vi.mock('@/lib/agent-launch-prompt-not-delivered-notice', () => ({
+  showAgentLaunchNotStartedNotice: mocks.showNotStarted
+}))
+
+describe('a pane spawn the host refused for what carries its prompt', () => {
+  const originalWindow = globalThis.window
+
+  beforeEach(() => {
+    vi.resetModules()
+    mocks.showNotStarted.mockReset()
+    installIpcPtyWindow(originalWindow, { data: () => {}, exit: () => {} })
+  })
+
+  afterEach(() => {
+    restorePtySpecWindow(originalWindow)
+  })
+
+  // Why: a WSL launch with no launch file still needs its staged line written into the distro.
+  it('hands back a staged line’s prompt to copy, as it does a launch file’s', async () => {
+    const { createIpcPtyTransport } = await import('./pty-transport')
+    vi.mocked(window.api.pty.spawn).mockRejectedValueOnce(
+      new Error(describeLaunchFileUnavailable('unreachable', 'staged-line'))
+    )
+
+    await createIpcPtyTransport({
+      command: "claude 'a long prompt'",
+      launchPrompt: 'a long prompt'
+    }).connect({ url: '', callbacks: { onError: vi.fn() } })
+
+    expect(mocks.showNotStarted).toHaveBeenCalledWith({ prompt: 'a long prompt' })
+  })
+
+  it('stays quiet for any other spawn failure', async () => {
+    const { createIpcPtyTransport } = await import('./pty-transport')
+    vi.mocked(window.api.pty.spawn).mockRejectedValueOnce(new Error('spawn ENOENT'))
+
+    await createIpcPtyTransport({ command: 'claude', launchPrompt: 'fix it' }).connect({
+      url: '',
+      callbacks: { onError: vi.fn() }
+    })
+
+    expect(mocks.showNotStarted).not.toHaveBeenCalled()
+  })
+})
diff --git a/src/renderer/src/components/terminal-pane/pty-transport-types.ts b/src/renderer/src/components/terminal-pane/pty-transport-types.ts
index 887a2b6ff43..b7b3b5d6150 100644
--- a/src/renderer/src/components/terminal-pane/pty-transport-types.ts
+++ b/src/renderer/src/components/terminal-pane/pty-transport-types.ts
@@ -270,6 +270,8 @@ export type IpcPtyTransportOptions = {
   startupCommandDelivery?: StartupCommandDelivery
   /** Local spawns only: a remote runtime writes no file the renderer names. */
   launchFile?: LaunchFile
+  /** See `PtyPaneStartup.launchPrompt`. */
+  launchPrompt?: string
   connectionId?: string | null
   executionHostId?: ExecutionHostId | null
   worktreeId?: string
diff --git a/src/renderer/src/hooks/composer-state/full-creation-startup.test.ts b/src/renderer/src/hooks/composer-state/full-creation-startup.test.ts
index 75b08c0d9ec..1f48e5233e7 100644
--- a/src/renderer/src/hooks/composer-state/full-creation-startup.test.ts
+++ b/src/renderer/src/hooks/composer-state/full-creation-startup.test.ts
@@ -1,5 +1,6 @@
 import { describe, expect, it } from 'vitest'
-import { buildAgentStartupPlan } from '@/lib/tui-agent-startup'
+import { planLaunchPrompt } from '@/lib/tui-agent-startup'
+import { composerAgentStartupPlan } from '@/lib/composer-agent-startup-plan'
 import { buildFullCreationStartup } from './full-creation-startup'
 import { tuiAgentToAgentKind } from '../../../../shared/agent-kind'
 
@@ -7,13 +8,17 @@ describe('the full composer’s renderer-spawned startup', () => {
   // Why: SSH repos, folder repos, repos with default tabs and a failed backend spawn all take this
   // path; without the file the agent is pointed at nothing and the prompt is lost.
   it('carries the launch file its command points at', () => {
-    const startupPlan = buildAgentStartupPlan({
-      agent: 'claude',
-      prompt: 'x'.repeat(200_000),
-      cmdOverrides: {},
-      platform: 'linux',
-      isRemote: true
-    })
+    const prompt = 'x'.repeat(200_000)
+    const startupPlan = composerAgentStartupPlan(
+      planLaunchPrompt({
+        agent: 'claude',
+        prompt,
+        cmdOverrides: {},
+        platform: 'linux',
+        isRemote: true
+      }),
+      prompt
+    )
     const startup = buildFullCreationStartup({
       startupPlan,
       backendSpawnedStartup: false,
@@ -29,5 +34,7 @@ describe('the full composer’s renderer-spawned startup', () => {
     expect(startupPlan?.launchFile).toBeDefined()
     expect(startup?.launchFile).toEqual(startupPlan?.launchFile)
     expect(startup?.command).toContain(startupPlan?.launchFile?.placeholder)
+    // Handed back to copy if the host refuses the spawn.
+    expect(startup?.launchPrompt).toBe(prompt)
   })
 })
diff --git a/src/renderer/src/hooks/composer-state/full-creation-startup.ts b/src/renderer/src/hooks/composer-state/full-creation-startup.ts
index a00666c2112..c0ddc1bd215 100644
--- a/src/renderer/src/hooks/composer-state/full-creation-startup.ts
+++ b/src/renderer/src/hooks/composer-state/full-creation-startup.ts
@@ -1,9 +1,9 @@
 import type { TuiAgent } from '../../../../shared/tui-agent'
-import type { AgentStartupPlan } from '@/lib/tui-agent-startup'
+import type { ComposerAgentStartupPlan } from '@/lib/composer-agent-startup-plan'
 import type { WorktreeStartupPayload } from '@/lib/worktree-startup-payload'
 
 export function buildFullCreationStartup(args: {
-  startupPlan: AgentStartupPlan | null
+  startupPlan: ComposerAgentStartupPlan | null
   backendSpawnedStartup: boolean
   agent: TuiAgent
   shouldSeedInitialAgentStatus: boolean
@@ -25,6 +25,7 @@ export function buildFullCreationStartup(args: {
       : {}),
     // Why: the command points at this file; without it the agent is told to read nothing.
     ...(args.startupPlan.launchFile ? { launchFile: args.startupPlan.launchFile } : {}),
+    ...(args.startupPlan.launchPrompt ? { launchPrompt: args.startupPlan.launchPrompt } : {}),
     ...(args.shouldSeedInitialAgentStatus
       ? { initialAgentStatus: { agent: args.agent, prompt: args.prompt.trim() } }
       : {}),
diff --git a/src/renderer/src/hooks/composer-state/full-submit-preparation.ts b/src/renderer/src/hooks/composer-state/full-submit-preparation.ts
index e3136a48b39..98c273cde24 100644
--- a/src/renderer/src/hooks/composer-state/full-submit-preparation.ts
+++ b/src/renderer/src/hooks/composer-state/full-submit-preparation.ts
@@ -30,7 +30,7 @@ import { ensureHooksConfirmed, confirmRuntimeIssueCommandRead } from '@/lib/ensu
 import { useAppStore } from '@/store'
 import type { SetupDecision } from '../../../../shared/worktree/create-types'
 import { resolveComposerBranchNameOverrideForCreate } from '../composer-branch-selection'
-import { buildAgentStartupPlan } from '@/lib/tui-agent-startup'
+import { planLaunchPrompt } from '@/lib/tui-agent-startup'
 import {
   resolveTuiAgentLaunchArgs,
   resolveTuiAgentLaunchEnv
@@ -39,7 +39,8 @@ import { resolveInitialNativeChatSessionOptions } from '@/components/native-chat
 import { isNativeChatTranscriptLocalReadable } from '@/lib/native-chat-transcript-readability'
 import type { AgentStartedTelemetry } from '@/lib/worktree-startup-payload'
 import { tuiAgentToAgentKind } from '@/lib/telemetry'
-import { launchHostWritesLaunchFile } from '@/lib/launch-file-host'
+import { launchHostIsPaired } from '@/lib/launch-file-host'
+import { composerAgentStartupPlan } from '@/lib/composer-agent-startup-plan'
 import type { PendingSmartGitHubSubmitResolution } from './source-selection-decisions'
 
 export function useFullSubmitPreparation(input: FullSubmitPreparationInput) {
@@ -181,30 +182,33 @@ export function useFullSubmitPreparation(input: FullSubmitPreparationInput) {
         !effectiveBranchNameOverride &&
         !createDisplayName
 
-      const startupPlan = buildAgentStartupPlan({
-        agent: tuiAgent,
-        prompt: submitStartupPrompt,
-        cmdOverrides: settings?.agentCmdOverrides ?? {},
-        agentArgs: resolveTuiAgentLaunchArgs(tuiAgent, settings?.agentDefaultArgs),
-        agentEnv: resolveTuiAgentLaunchEnv(tuiAgent, settings?.agentDefaultEnv),
-        sessionOptions: resolveInitialNativeChatSessionOptions(
-          {
-            experimentalNativeChat: settings?.experimentalNativeChat,
-            openAgentTabsInChatByDefault: settings?.openAgentTabsInChatByDefault,
-            nativeChatSessionOptions: settings?.nativeChatSessionOptions
-          },
-          {
-            agent: tuiAgent,
-            nativeChatTranscriptIsLocalReadable: isNativeChatTranscriptLocalReadable(
-              selectedRepo?.connectionId
-            )
-          }
-        ),
-        platform: selectedRepoAgentLaunchPlatform,
-        shell: selectedRepoStartupShell,
-        isRemote: selectedRepoIsRemote,
-        hostWritesLaunchFile: launchHostWritesLaunchFile(selectedRepoSettings)
-      })
+      const startupPlan = composerAgentStartupPlan(
+        planLaunchPrompt({
+          agent: tuiAgent,
+          prompt: submitStartupPrompt,
+          cmdOverrides: settings?.agentCmdOverrides ?? {},
+          agentArgs: resolveTuiAgentLaunchArgs(tuiAgent, settings?.agentDefaultArgs),
+          agentEnv: resolveTuiAgentLaunchEnv(tuiAgent, settings?.agentDefaultEnv),
+          sessionOptions: resolveInitialNativeChatSessionOptions(
+            {
+              experimentalNativeChat: settings?.experimentalNativeChat,
+              openAgentTabsInChatByDefault: settings?.openAgentTabsInChatByDefault,
+              nativeChatSessionOptions: settings?.nativeChatSessionOptions
+            },
+            {
+              agent: tuiAgent,
+              nativeChatTranscriptIsLocalReadable: isNativeChatTranscriptLocalReadable(
+                selectedRepo?.connectionId
+              )
+            }
+          ),
+          platform: selectedRepoAgentLaunchPlatform,
+          shell: selectedRepoStartupShell,
+          isRemote: selectedRepoIsRemote,
+          launchHostIsPaired: launchHostIsPaired(selectedRepoSettings?.activeRuntimeEnvironmentId)
+        }),
+        submitStartupPrompt
+      )
 
       const shouldSeedInitialAgentStatus =
         tuiAgent === 'command-code' && submitStartupPrompt.trim().length > 0
@@ -217,7 +221,7 @@ export function useFullSubmitPreparation(input: FullSubmitPreparationInput) {
       }
 
       const backendStartup =
-        startupPlan && !startupPlan.draftPrompt && !startupPlan.followupPrompt
+        startupPlan && !startupPlan.draftPrompt && !startupPlan.pastePromptAfterReady
           ? {
               command: startupPlan.launchCommand,
               ...(startupPlan.env ? { env: startupPlan.env } : {}),
diff --git a/src/renderer/src/hooks/composer-state/quick-creation-execution.ts b/src/renderer/src/hooks/composer-state/quick-creation-execution.ts
index edca524131b..37a57f9056b 100644
--- a/src/renderer/src/hooks/composer-state/quick-creation-execution.ts
+++ b/src/renderer/src/hooks/composer-state/quick-creation-execution.ts
@@ -40,7 +40,7 @@ import { useAppStore } from '@/store'
 import { settleComposerSubmit } from '@/lib/composer-submit-cancellation'
 import { ensureHooksConfirmed } from '@/lib/ensure-hooks-confirmed'
 import { getActiveRuntimeTarget } from '@/runtime/runtime-rpc-client'
-import { launchHostWritesLaunchFile } from '@/lib/launch-file-host'
+import { launchHostIsPaired } from '@/lib/launch-file-host'
 import { runBackgroundWorktreeCreation } from '@/lib/worktree-creation-flow'
 import { translate } from '@/i18n/i18n'
 import { resolveQuickCreateLinkedWorkItemPrompt } from '@/lib/linked-work-item-context'
@@ -140,7 +140,7 @@ export function useQuickCreationExecution(input: QuickCreationExecutionInput) {
         platform: selectedRepoAgentLaunchPlatform,
         shell: selectedRepoStartupShell,
         isRemote: selectedRepoIsRemote,
-        hostWritesLaunchFile: launchHostWritesLaunchFile(selectedRepoSettings),
+        launchHostIsPaired: launchHostIsPaired(selectedRepoSettings?.activeRuntimeEnvironmentId),
         telemetrySource
       })
 
diff --git a/src/renderer/src/hooks/composer-state/quick-startup-plan.ts b/src/renderer/src/hooks/composer-state/quick-startup-plan.ts
index 30adf55de72..2b2eed7538f 100644
--- a/src/renderer/src/hooks/composer-state/quick-startup-plan.ts
+++ b/src/renderer/src/hooks/composer-state/quick-startup-plan.ts
@@ -3,8 +3,11 @@ import type { TuiAgent } from '../../../../shared/tui-agent'
 import type { AgentStartupShell } from '../../../../shared/tui-agent-startup-shell'
 import type { AgentStartedTelemetry } from '@/lib/worktree-startup-payload'
 import type { WorktreeCreationRequest } from '@/lib/pending-worktree-creation'
-import type { AgentStartupPlan } from '@/lib/tui-agent-startup'
-import { buildAgentDraftLaunchPlan, buildAgentStartupPlan } from '@/lib/tui-agent-startup'
+import { buildAgentDraftLaunchPlan, planLaunchPrompt } from '@/lib/tui-agent-startup'
+import {
+  composerAgentStartupPlan,
+  type ComposerAgentStartupPlan
+} from '@/lib/composer-agent-startup-plan'
 import {
   resolveTuiAgentLaunchArgs,
   resolveTuiAgentLaunchEnv
@@ -22,12 +25,13 @@ export type QuickComposerStartupInput = {
   platform: NodeJS.Platform
   shell: AgentStartupShell | null | undefined
   isRemote: boolean
-  hostWritesLaunchFile: boolean
+  /** See `launchHostIsPaired`. */
+  launchHostIsPaired: boolean
   telemetrySource: WorktreeCreationRequest['telemetrySource']
 }
 
 export type QuickComposerStartup = {
-  startupPlan: AgentStartupPlan | null
+  startupPlan: ComposerAgentStartupPlan | null
   backendStartup: WorktreeCreationRequest['startup']
   telemetry: AgentStartedTelemetry | null
 }
@@ -67,13 +71,12 @@ export function buildQuickComposerStartup(input: QuickComposerStartupInput): Qui
           shell: input.shell ?? undefined,
           isRemote: input.isRemote
         })
-  let startupPlan: AgentStartupPlan | null = null
+  let startupPlan: ComposerAgentStartupPlan | null = null
   if (draftLaunchPlan) {
     startupPlan = {
       agent: draftLaunchPlan.agent,
       launchCommand: draftLaunchPlan.launchCommand,
       expectedProcess: draftLaunchPlan.expectedProcess,
-      followupPrompt: null,
       launchConfig: draftLaunchPlan.launchConfig,
       ...(draftLaunchPlan.sessionOptions ? { sessionOptions: draftLaunchPlan.sessionOptions } : {}),
       ...(draftLaunchPlan.startupCommandDelivery
@@ -82,19 +85,21 @@ export function buildQuickComposerStartup(input: QuickComposerStartupInput): Qui
       ...(draftLaunchPlan.env ? { env: draftLaunchPlan.env } : {})
     }
   } else if (agent !== null) {
-    startupPlan = buildAgentStartupPlan({
-      agent,
-      prompt,
-      cmdOverrides: settings?.agentCmdOverrides ?? {},
-      agentArgs: resolveTuiAgentLaunchArgs(agent, settings?.agentDefaultArgs),
-      agentEnv: resolveTuiAgentLaunchEnv(agent, settings?.agentDefaultEnv),
-      sessionOptions,
-      platform: input.platform,
-      shell: input.shell ?? undefined,
-      isRemote: input.isRemote,
-      hostWritesLaunchFile: input.hostWritesLaunchFile,
-      allowEmptyPromptLaunch: true
-    })
+    startupPlan = composerAgentStartupPlan(
+      planLaunchPrompt({
+        agent,
+        prompt,
+        cmdOverrides: settings?.agentCmdOverrides ?? {},
+        agentArgs: resolveTuiAgentLaunchArgs(agent, settings?.agentDefaultArgs),
+        agentEnv: resolveTuiAgentLaunchEnv(agent, settings?.agentDefaultEnv),
+        sessionOptions,
+        platform: input.platform,
+        shell: input.shell ?? undefined,
+        isRemote: input.isRemote,
+        launchHostIsPaired: input.launchHostIsPaired
+      }),
+      prompt
+    )
     if (startupPlan && draftPrompt) {
       startupPlan.draftPrompt = draftPrompt
     }
@@ -109,7 +114,7 @@ export function buildQuickComposerStartup(input: QuickComposerStartupInput): Qui
           request_kind: 'new'
         }
   const backendStartup =
-    startupPlan && !startupPlan.draftPrompt && !startupPlan.followupPrompt
+    startupPlan && !startupPlan.draftPrompt && !startupPlan.pastePromptAfterReady
       ? {
           command: startupPlan.launchCommand,
           ...(startupPlan.env ? { env: startupPlan.env } : {}),
diff --git a/src/renderer/src/lib/agent-launch-caller-permissions.test.ts b/src/renderer/src/lib/agent-launch-caller-permissions.test.ts
index 7e7ffee9182..9323a9aa5a4 100644
--- a/src/renderer/src/lib/agent-launch-caller-permissions.test.ts
+++ b/src/renderer/src/lib/agent-launch-caller-permissions.test.ts
@@ -39,6 +39,9 @@ vi.mock('@/components/native-chat/native-chat-session-option-cache', () => ({
   seedNativeChatAppliedSessionOptions: vi.fn()
 }))
 vi.mock('@/lib/agent-paste-draft', () => ({ pasteDraftWhenAgentReady: vi.fn(async () => true) }))
+vi.mock('@/lib/agent-launch-prompt-receipt', () => ({
+  waitForLaunchPromptReceipt: vi.fn(async () => true)
+}))
 vi.mock('@/lib/agent-ready-wait', () => ({
   waitForAgentReady: vi.fn(async () => ({ ready: true, reason: 'foreground-match' }))
 }))
diff --git a/src/renderer/src/lib/agent-launch-caller-placement.test.ts b/src/renderer/src/lib/agent-launch-caller-placement.test.ts
index 03db8a68367..d2721aa2805 100644
--- a/src/renderer/src/lib/agent-launch-caller-placement.test.ts
+++ b/src/renderer/src/lib/agent-launch-caller-placement.test.ts
@@ -39,6 +39,9 @@ vi.mock('@/components/native-chat/native-chat-session-option-cache', () => ({
   seedNativeChatAppliedSessionOptions: vi.fn()
 }))
 vi.mock('@/lib/agent-paste-draft', () => ({ pasteDraftWhenAgentReady: vi.fn(async () => true) }))
+vi.mock('@/lib/agent-launch-prompt-receipt', () => ({
+  waitForLaunchPromptReceipt: vi.fn(async () => true)
+}))
 vi.mock('@/lib/agent-ready-wait', () => ({
   waitForAgentReady: vi.fn(async () => ({ ready: true, reason: 'foreground-match' }))
 }))
diff --git a/src/renderer/src/lib/agent-launch-caller-prompt-transport.test.ts b/src/renderer/src/lib/agent-launch-caller-prompt-transport.test.ts
index e1a3d474de9..696d3e40423 100644
--- a/src/renderer/src/lib/agent-launch-caller-prompt-transport.test.ts
+++ b/src/renderer/src/lib/agent-launch-caller-prompt-transport.test.ts
@@ -48,6 +48,10 @@ vi.mock('@/lib/agent-paste-draft', () => ({
 vi.mock('@/lib/agent-ready-wait', () => ({
   waitForAgentReady: vi.fn(async () => ({ ready: true, reason: 'foreground-match' }))
 }))
+const mockWaitForLaunchPromptReceipt = vi.fn(async () => true)
+vi.mock('@/lib/agent-launch-prompt-receipt', () => ({
+  waitForLaunchPromptReceipt: mockWaitForLaunchPromptReceipt
+}))
 vi.mock('@/runtime/local-runtime-capabilities', () => ({
   readLocalRuntimeCapabilitiesOrUnknown: () => []
 }))
@@ -159,14 +163,16 @@ describe('agent launch caller prompt transport', () => {
   )
 
   it.each(cases)(
-    'exposes no delivery promise to %s, whose Codex prompt rides argv or is a draft',
+    'exposes a delivery result to %s only for a prompt it waits to see submitted',
     async (_id, profile) => {
       const result = await launch(profile)
 
-      // Why: three call sites branch on this promise being present. A draft launch must NOT get one,
-      // because the composer owns the text until the user sends it, and a prompt the launch command
-      // carries has no paste to wait for.
-      expect(result?.promptDeliveryResult).toBeUndefined()
+      // Why: call sites branch on this result being present. A draft launch must NOT get one,
+      // because the composer owns the text until the user sends it; a submit-after-ready prompt the
+      // launch command carries still waits on the agent's receipt before the caller acts.
+      const waitsForSubmit =
+        profile.args.prompt !== undefined && profile.args.promptDelivery === 'submit-after-ready'
+      expect(result?.promptDeliveryResult !== undefined).toBe(waitsForSubmit)
     }
   )
 
@@ -215,6 +221,38 @@ describe('agent launch caller prompt transport', () => {
     }
   })
 
+  // The bug class the explicit carry outcome removes: a plan that exists but does not carry the
+  // prompt was treated as delivered, and the prompt was dropped.
+  it.each([
+    ['past the argv ceiling', 'x'.repeat(20_000), 'darwin' as const],
+    ['a Windows-damaged prompt', 'say "hi"', 'win32' as const]
+  ])(
+    'pastes %s for an agent not measured reading a launch file, and waits for that paste',
+    async (_label, prompt, launchPlatform) => {
+      const { launchAgentInNewTab } = await import('./launch-agent-in-new-tab')
+
+      const result = launchAgentInNewTab({
+        agent: 'gemini',
+        worktreeId: 'wt-1',
+        prompt,
+        promptDelivery: 'submit-after-ready',
+        launchPlatform
+      })
+
+      await expect(result?.promptDeliveryResult).resolves.toEqual({
+        delivered: true,
+        failureNotified: false
+      })
+      expect(mockPasteDraftWhenAgentReady.mock.calls[0]?.[0]).toMatchObject({
+        content: prompt,
+        submit: true
+      })
+      expect(mockWaitForLaunchPromptReceipt).not.toHaveBeenCalled()
+      expect(queuedStartupCommand(store)).not.toContain(prompt.slice(0, 8))
+      expect(queuedStartupPayload(store)?.launchFile).toBeUndefined()
+    }
+  )
+
   it('mirrors an argv-carried draft into the chat composer', async () => {
     const { launchAgentInNewTab } = await import('./launch-agent-in-new-tab')
 
diff --git a/src/renderer/src/lib/agent-launch-prompt-receipt.test.ts b/src/renderer/src/lib/agent-launch-prompt-receipt.test.ts
new file mode 100644
index 00000000000..475a77660a6
--- /dev/null
+++ b/src/renderer/src/lib/agent-launch-prompt-receipt.test.ts
@@ -0,0 +1,117 @@
+import { beforeEach, describe, expect, it, vi } from 'vitest'
+
+type ReceiptState = {
+  tabsByWorktree: Record<string, { id: string }[]>
+  ptyIdsByTabId: Record<string, string[]>
+  agentStatusByPaneKey: Record<string, { updatedAt: number }>
+  settings: Record<string, unknown>
+}
+
+const mocks = vi.hoisted(() => {
+  let state: ReceiptState = {
+    tabsByWorktree: {},
+    ptyIdsByTabId: {},
+    agentStatusByPaneKey: {},
+    settings: {}
+  }
+  const listeners = new Set<(next: ReceiptState) => void>()
+  return {
+    getState: () => state,
+    setState: (next: Partial<ReceiptState>) => {
+      state = { ...state, ...next }
+      for (const listener of listeners) {
+        listener(state)
+      }
+    },
+    subscribe: (listener: (next: ReceiptState) => void) => {
+      listeners.add(listener)
+      return () => listeners.delete(listener)
+    },
+    reset: () => {
+      listeners.clear()
+      state = { tabsByWorktree: {}, ptyIdsByTabId: {}, agentStatusByPaneKey: {}, settings: {} }
+    },
+    readiness: vi.fn(),
+    waitForAgentReady: vi.fn()
+  }
+})
+
+vi.mock('@/store', () => ({
+  useAppStore: { getState: mocks.getState, subscribe: mocks.subscribe }
+}))
+vi.mock('./agent-paste-draft', () => ({
+  PTY_SPAWN_TIMEOUT_MS: 8000,
+  getSettingsForAgentTabRuntimeOwner: () => ({}),
+  waitForAgentDraftInputReadyOnTab: mocks.readiness
+}))
+vi.mock('./agent-ready-wait', () => ({ waitForAgentReady: mocks.waitForAgentReady }))
+
+import { waitForLaunchPromptReceipt } from './agent-launch-prompt-receipt'
+
+const never = (): Promise<never> => new Promise(() => {})
+
+describe('whether a prompt that rode the launch command reached the agent', () => {
+  beforeEach(() => {
+    mocks.reset()
+    mocks.setState({ tabsByWorktree: { 'wt-1': [{ id: 'tab-1' }] } })
+    mocks.readiness.mockReset().mockImplementation(never)
+    mocks.waitForAgentReady.mockReset()
+  })
+
+  it('is delivered when the agent’s own hook reports a turn on the tab', async () => {
+    const receipt = waitForLaunchPromptReceipt({ tabId: 'tab-1', agent: 'claude', launchedAt: 100 })
+    mocks.setState({ ptyIdsByTabId: { 'tab-1': ['pty-1'] } })
+    mocks.setState({
+      agentStatusByPaneKey: { 'tab-1:11111111-1111-4111-8111-111111111111': { updatedAt: 150 } }
+    })
+    await expect(receipt).resolves.toBe(true)
+  })
+
+  it('ignores a status row older than the launch', async () => {
+    mocks.setState({
+      agentStatusByPaneKey: { 'tab-1:11111111-1111-4111-8111-111111111111': { updatedAt: 50 } }
+    })
+    mocks.readiness.mockResolvedValue(null)
+    const receipt = waitForLaunchPromptReceipt({ tabId: 'tab-1', agent: 'claude', launchedAt: 100 })
+    await expect(receipt).resolves.toBe(false)
+  })
+
+  it('is not delivered when the PTY exits before the agent shows any sign of the prompt', async () => {
+    const receipt = waitForLaunchPromptReceipt({ tabId: 'tab-1', agent: 'claude', launchedAt: 100 })
+    mocks.setState({ ptyIdsByTabId: { 'tab-1': ['pty-1'] } })
+    mocks.setState({ ptyIdsByTabId: { 'tab-1': [] } })
+    await expect(receipt).resolves.toBe(false)
+  })
+
+  it('is not delivered when the PTY never spawns, as when the host refused its launch file', async () => {
+    mocks.readiness.mockResolvedValue(null)
+    const receipt = waitForLaunchPromptReceipt({ tabId: 'tab-1', agent: 'codex', launchedAt: 100 })
+    await expect(receipt).resolves.toBe(false)
+  })
+
+  it('is delivered when the agent shows the composer a paste would have waited for', async () => {
+    mocks.readiness.mockResolvedValue({ ptyId: 'pty-1', ready: true })
+    const receipt = waitForLaunchPromptReceipt({ tabId: 'tab-1', agent: 'codex', launchedAt: 100 })
+    await expect(receipt).resolves.toBe(true)
+  })
+
+  it('reports an unconfirmed delivery when only the agent process was seen', async () => {
+    mocks.readiness.mockResolvedValue({ ptyId: 'pty-1', ready: false })
+    mocks.waitForAgentReady.mockResolvedValue({ ready: true, reason: 'foreground-match' })
+    const onUnconfirmedDelivery = vi.fn()
+    const receipt = waitForLaunchPromptReceipt({
+      tabId: 'tab-1',
+      agent: 'codex',
+      launchedAt: 100,
+      onUnconfirmedDelivery
+    })
+    await expect(receipt).resolves.toBe(true)
+    expect(onUnconfirmedDelivery).toHaveBeenCalledOnce()
+  })
+
+  it('is not delivered when the tab closes first', async () => {
+    const receipt = waitForLaunchPromptReceipt({ tabId: 'tab-1', agent: 'claude', launchedAt: 100 })
+    mocks.setState({ tabsByWorktree: { 'wt-1': [] } })
+    await expect(receipt).resolves.toBe(false)
+  })
+})
diff --git a/src/renderer/src/lib/agent-launch-prompt-receipt.ts b/src/renderer/src/lib/agent-launch-prompt-receipt.ts
new file mode 100644
index 00000000000..e2ef5c2c00a
--- /dev/null
+++ b/src/renderer/src/lib/agent-launch-prompt-receipt.ts
@@ -0,0 +1,98 @@
+import type { TuiAgent } from '../../../shared/tui-agent'
+import { TUI_AGENT_CONFIG } from '../../../shared/tui-agent-config'
+import { resolveDraftPasteReadyTimeoutMs } from '../../../shared/draft-paste-ready-timeout'
+import { parsePaneKey } from '../../../shared/stable-pane-id'
+import { useAppStore } from '@/store'
+import { waitForAgentReady } from './agent-ready-wait'
+import {
+  getSettingsForAgentTabRuntimeOwner,
+  PTY_SPAWN_TIMEOUT_MS,
+  waitForAgentDraftInputReadyOnTab
+} from './agent-paste-draft'
+
+type AppState = ReturnType<typeof useAppStore.getState>
+
+function hookReportedTurnOnTab(state: AppState, tabId: string, launchedAt: number): boolean {
+  return Object.entries(state.agentStatusByPaneKey).some(
+    ([paneKey, entry]) => parsePaneKey(paneKey)?.tabId === tabId && entry.updatedAt >= launchedAt
+  )
+}
+
+function tabExists(state: AppState, tabId: string): boolean {
+  return Object.values(state.tabsByWorktree).some((tabs) => tabs?.some((tab) => tab.id === tabId))
+}
+
+/**
+ * Whether a prompt that rode an agent's launch command reached the agent. True at the first real
+ * signal: the agent's own hook reports a turn on the tab, or the agent shows the composer a paste
+ * would have waited for. False when the PTY never spawns (a refused launch file included), exits
+ * first, or the tab closes. Never "the tab exists".
+ */
+export function waitForLaunchPromptReceipt(args: {
+  tabId: string
+  agent: TuiAgent
+  launchedAt: number
+  /** The agent process was seen, but neither its hook nor its composer. */
+  onUnconfirmedDelivery?: () => void
+}): Promise<boolean> {
+  const { tabId, agent, launchedAt } = args
+  return new Promise((resolve) => {
+    let settled = false
+    let sawPty = false
+    let unsubscribe: (() => void) | null = null
+    const finish = (delivered: boolean): void => {
+      if (settled) {
+        return
+      }
+      settled = true
+      unsubscribe?.()
+      resolve(delivered)
+    }
+    let scannedStatus: AppState['agentStatusByPaneKey'] | null = null
+    const observe = (state: AppState): void => {
+      // Why the identity check: the store writes often and replaces this map only when it changes.
+      if (state.agentStatusByPaneKey !== scannedStatus) {
+        scannedStatus = state.agentStatusByPaneKey
+        if (hookReportedTurnOnTab(state, tabId, launchedAt)) {
+          finish(true)
+          return
+        }
+      }
+      if ((state.ptyIdsByTabId[tabId]?.length ?? 0) > 0) {
+        sawPty = true
+      } else if (sawPty || !tabExists(state, tabId)) {
+        finish(false)
+      }
+    }
+    unsubscribe = useAppStore.subscribe(observe)
+    observe(useAppStore.getState())
+    const config = TUI_AGENT_CONFIG[agent]
+    void waitForAgentDraftInputReadyOnTab({
+      tabId,
+      spawnTimeoutMs: PTY_SPAWN_TIMEOUT_MS,
+      readinessTimeoutMs: resolveDraftPasteReadyTimeoutMs(agent),
+      readySignal: config.draftPasteReadySignal ?? 'render-quiet-after-bracketed-paste',
+      settings: getSettingsForAgentTabRuntimeOwner(tabId)
+    }).then(
+      async (readiness) => {
+        if (settled) {
+          return
+        }
+        if (!readiness) {
+          finish(false)
+          return
+        }
+        if (readiness.ready) {
+          finish(true)
+          return
+        }
+        const fallback = await waitForAgentReady(tabId, config.expectedProcess, { timeoutMs: 1000 })
+        if (fallback.ready && !settled) {
+          args.onUnconfirmedDelivery?.()
+        }
+        finish(fallback.ready)
+      },
+      () => finish(false)
+    )
+  })
+}
diff --git a/src/renderer/src/lib/agent-paste-draft.ts b/src/renderer/src/lib/agent-paste-draft.ts
index f0df326b316..510ebb1aae0 100644
--- a/src/renderer/src/lib/agent-paste-draft.ts
+++ b/src/renderer/src/lib/agent-paste-draft.ts
@@ -42,7 +42,7 @@ export const POST_PASTE_SUBMIT_DELAY_MS = AGENT_PROMPT_POST_PASTE_SUBMIT_DELAY_M
 // hasn't appeared in 8s means the launch itself failed — waiting the (longer)
 // composer budget on top would only delay that verdict. Keeping them distinct
 // also stops one slow step from spending the other's budget (STA-3367).
-const PTY_SPAWN_TIMEOUT_MS = 8000
+export const PTY_SPAWN_TIMEOUT_MS = 8000
 
 export function getSettingsForAgentTabRuntimeOwner(
   tabId: string
@@ -273,7 +273,7 @@ async function sendBracketedPasteToAgent(args: {
   }
 }
 
-function waitForAgentDraftInputReadyOnTab(args: {
+export function waitForAgentDraftInputReadyOnTab(args: {
   tabId: string
   spawnTimeoutMs: number
   readinessTimeoutMs: number
diff --git a/src/renderer/src/lib/agent-startup-delayed-delivery.ts b/src/renderer/src/lib/agent-startup-delayed-delivery.ts
index 2e38fad41d1..3a1cde634f3 100644
--- a/src/renderer/src/lib/agent-startup-delayed-delivery.ts
+++ b/src/renderer/src/lib/agent-startup-delayed-delivery.ts
@@ -1,5 +1,5 @@
 import { useAppStore } from '@/store'
-import type { AgentStartupPlan } from '@/lib/tui-agent-startup'
+import type { ComposerAgentStartupPlan } from '@/lib/composer-agent-startup-plan'
 import { parsePaneKey } from '../../../shared/stable-pane-id'
 import {
   agentStartupDeliveryKey as deliveryKey,
@@ -15,8 +15,8 @@ type PendingAgentStartupDelivery = {
   worktreeId: string
   tabId: string
   launchToken: string
-  startup: AgentStartupPlan
-  deliver: (tabId: string, ptyId: string, startup: AgentStartupPlan) => Promise<void>
+  startup: ComposerAgentStartupPlan
+  deliver: (tabId: string, ptyId: string, startup: ComposerAgentStartupPlan) => Promise<void>
 }
 
 const pendingAgentStartupDeliveries = new Map<string, PendingAgentStartupDelivery>()
diff --git a/src/renderer/src/lib/composer-agent-startup-plan.ts b/src/renderer/src/lib/composer-agent-startup-plan.ts
new file mode 100644
index 00000000000..18cfaa41df6
--- /dev/null
+++ b/src/renderer/src/lib/composer-agent-startup-plan.ts
@@ -0,0 +1,31 @@
+import type { AgentStartupPlan } from '@/lib/tui-agent-startup'
+import type { LaunchPromptPlan } from '../../../shared/launch-prompt-carry'
+import type { LaunchFile } from '../../../shared/launch-prompt-file'
+
+/** What a created workspace launches, and the submitted prompt it still owes the agent, if any. */
+export type ComposerAgentStartupPlan = AgentStartupPlan & {
+  launchFile?: LaunchFile
+  /** The submitted prompt the launch carries; handed back to copy if the host refuses the spawn. */
+  launchPrompt?: string
+  /** A submitted prompt the launch could not carry; pasted once the agent is ready. */
+  pastePromptAfterReady?: string
+}
+
+export function composerAgentStartupPlan(
+  planned: LaunchPromptPlan<AgentStartupPlan> | null,
+  prompt: string
+): ComposerAgentStartupPlan | null {
+  if (!planned) {
+    return null
+  }
+  switch (planned.carry) {
+    case 'none':
+      return planned.plan
+    case 'on-line':
+      return { ...planned.plan, launchPrompt: prompt.trim() }
+    case 'launch-file':
+      return { ...planned.plan, launchFile: planned.launchFile, launchPrompt: prompt.trim() }
+    case 'paste-after-ready':
+      return { ...planned.cleanPlan, pastePromptAfterReady: planned.text }
+  }
+}
diff --git a/src/renderer/src/lib/launch-agent-background-session.ts b/src/renderer/src/lib/launch-agent-background-session.ts
index 76e776c6be5..8ef6c3fe556 100644
--- a/src/renderer/src/lib/launch-agent-background-session.ts
+++ b/src/renderer/src/lib/launch-agent-background-session.ts
@@ -1,6 +1,7 @@
 import { useAppStore } from '@/store'
-import { launchHostWritesLaunchFile } from '@/lib/launch-file-host'
-import { buildAgentStartupPlan } from '@/lib/tui-agent-startup'
+import { launchHostIsPaired } from '@/lib/launch-file-host'
+import { planLaunchPrompt, type AgentStartupPlan } from '@/lib/tui-agent-startup'
+import type { LaunchFile } from '../../../shared/launch-prompt-file'
 import type {
   LaunchAgentBackgroundSessionArgs,
   LaunchAgentBackgroundSessionResult
@@ -75,23 +76,38 @@ export async function launchAgentBackgroundSession(
   // Route by the worktree's owner host, not the focused runtime.
   const ownerSettings = getSettingsForWorktreeRuntimeOwner(store, worktreeId)
   const runtimeTarget = getActiveRuntimeTarget(ownerSettings)
-  const startupPlan = buildAgentStartupPlan({
+  const planned = planLaunchPrompt({
     agent,
-    prompt: hasPrompt && !isFollowupPath ? trimmedPrompt : '',
+    prompt: trimmedPrompt,
     cmdOverrides,
     agentArgs,
     agentEnv,
     platform: launchPlatform,
     shell: startupShell,
     isRemote,
-    hostWritesLaunchFile: launchHostWritesLaunchFile(ownerSettings),
-    allowEmptyPromptLaunch: !hasPrompt || isFollowupPath
+    launchHostIsPaired: launchHostIsPaired(ownerSettings?.activeRuntimeEnvironmentId)
   })
-  if (!startupPlan) {
+  if (!planned) {
     return null
   }
-  // A prompt the host could not be sent in a launch file waits for the paste, as a stdin agent's does.
-  const promptLeftForPaste = hasPrompt && !isFollowupPath ? startupPlan.followupPrompt : null
+  let startupPlan: AgentStartupPlan
+  let launchFile: LaunchFile | undefined
+  // An argv agent's prompt the line could not carry waits for the paste, as a stdin agent's does.
+  let promptLeftForPaste: string | null = null
+  switch (planned.carry) {
+    case 'none':
+    case 'on-line':
+      startupPlan = planned.plan
+      break
+    case 'launch-file':
+      startupPlan = planned.plan
+      launchFile = planned.launchFile
+      break
+    case 'paste-after-ready':
+      startupPlan = planned.cleanPlan
+      promptLeftForPaste = isFollowupPath ? null : planned.text
+      break
+  }
   let pasteDraftAfterLaunch = hasPrompt && isFollowupPath ? trimmedPrompt : promptLeftForPaste
 
   // A hidden run tab must never be store-visible without its PTY (#2989).
@@ -192,7 +208,7 @@ export async function launchAgentBackgroundSession(
             ? { startupCommandDelivery: startupPlan.startupCommandDelivery }
             : {}),
         env: paneEnv,
-        ...(startupPlan.launchFile ? { launchFile: startupPlan.launchFile } : {}),
+        ...(launchFile ? { launchFile } : {}),
         launchConfig: startupPlan.launchConfig,
         launchToken,
         launchAgent: agent,
diff --git a/src/renderer/src/lib/launch-agent-in-new-tab.test.ts b/src/renderer/src/lib/launch-agent-in-new-tab.test.ts
index 1352314224c..d44a7bb6739 100644
--- a/src/renderer/src/lib/launch-agent-in-new-tab.test.ts
+++ b/src/renderer/src/lib/launch-agent-in-new-tab.test.ts
@@ -14,6 +14,7 @@ const mockMarkNativeChatLaunchPromptFailed = vi.fn()
 const mockTrack = vi.fn()
 const mockToastMessage = vi.fn()
 const mockWaitForAgentReady = vi.fn()
+const mockWaitForLaunchPromptReceipt = vi.fn()
 
 const store = {
   activeRepoId: 'repo-1',
@@ -117,6 +118,10 @@ vi.mock('@/lib/agent-ready-wait', () => ({
   waitForAgentReady: mockWaitForAgentReady
 }))
 
+vi.mock('@/lib/agent-launch-prompt-receipt', () => ({
+  waitForLaunchPromptReceipt: mockWaitForLaunchPromptReceipt
+}))
+
 vi.mock('@/lib/telemetry', () => ({
   track: mockTrack,
   tuiAgentToAgentKind: (agent: string) => agent
@@ -137,6 +142,7 @@ vi.mock('@/runtime/web-runtime-session', () => ({
 describe('launchAgentInNewTab', () => {
   beforeEach(() => {
     vi.clearAllMocks()
+    mockWaitForLaunchPromptReceipt.mockResolvedValue(true)
     mockIsWebRuntimeSessionActive.mockReturnValue(false)
     mockCreateWebRuntimeSessionTerminal.mockResolvedValue({ status: 'created' })
     mockCreateWebRuntimeAgentSessionTerminalWithLaunchDraft.mockResolvedValue({ status: 'created' })
@@ -699,7 +705,14 @@ describe('launchAgentInNewTab', () => {
       promptDelivery: 'submit-after-ready'
     })
 
-    expect(result?.promptDeliveryResult).toBeUndefined()
+    // Why: the launch line carried it, so delivery is the agent's receipt, never the tab existing.
+    await expect(result?.promptDeliveryResult).resolves.toEqual({
+      delivered: true,
+      failureNotified: false
+    })
+    expect(mockWaitForLaunchPromptReceipt).toHaveBeenCalledWith(
+      expect.objectContaining({ tabId: 'tab-1', agent: 'command-code' })
+    )
     expect(mockPasteDraftWhenAgentReady).not.toHaveBeenCalled()
     // Why: Command Code has no prompt-submit hook, so the spawn seeds working from this prompt.
     expect(mockQueueTabStartupCommand).toHaveBeenCalledWith(
@@ -839,14 +852,36 @@ describe('launchAgentInNewTab', () => {
       promptDelivery: 'submit-after-ready'
     })
 
-    expect(result?.promptDeliveryResult).toBeUndefined()
+    await expect(result?.promptDeliveryResult).resolves.toEqual({
+      delivered: true,
+      failureNotified: false
+    })
     expect(mockPasteDraftWhenAgentReady).not.toHaveBeenCalled()
     const queued = mockQueueTabStartupCommand.mock.calls[0]?.[1]
+    // Handed back to copy if the host refuses to write the file.
+    expect(queued?.launchPrompt).toBe(prompt)
     expect(queued?.launchFile).toMatchObject({ content: prompt, sensitive: false })
     expect(queued?.command).toContain(queued?.launchFile?.placeholder)
     expect(queued?.command).not.toContain('xxxx')
   })
 
+  it('reports a carried prompt undelivered when the agent never received it', async () => {
+    const { launchAgentInNewTab } = await import('./launch-agent-in-new-tab')
+    mockWaitForLaunchPromptReceipt.mockResolvedValue(false)
+    const onPromptDelivered = vi.fn()
+
+    const result = launchAgentInNewTab({
+      agent: 'claude',
+      worktreeId: 'wt-1',
+      prompt: 'resolve these threads',
+      promptDelivery: 'submit-after-ready',
+      onPromptDelivered
+    })
+
+    await expect(result?.promptDeliveryResult).resolves.toMatchObject({ delivered: false })
+    expect(onPromptDelivered).not.toHaveBeenCalled()
+  })
+
   it('queues per-launch CLI arguments ahead of the generated prompt on argv', async () => {
     const { launchAgentInNewTab } = await import('./launch-agent-in-new-tab')
 
diff --git a/src/renderer/src/lib/launch-agent-in-new-tab.ts b/src/renderer/src/lib/launch-agent-in-new-tab.ts
index 208c51a9e45..096a2241d5c 100644
--- a/src/renderer/src/lib/launch-agent-in-new-tab.ts
+++ b/src/renderer/src/lib/launch-agent-in-new-tab.ts
@@ -3,12 +3,8 @@ import type { AgentStartupPlan } from '@/lib/tui-agent-startup'
 import { planLaunchAgentStartupPrompt } from '@/lib/launch-agent-startup-prompt-plan'
 import { persistAgentLaunchTabOrder } from '@/lib/launch-agent-tab-order'
 import { tuiAgentToAgentKind } from '@/lib/telemetry'
-import { createPasteReadinessTimeoutNotice } from '@/lib/launch-agent-paste-timeout-notice'
-import {
-  deliverLaunchPromptToAgentTab,
-  seedNativeChatLaunchDraftForAgentTab,
-  seedNativeChatLaunchPromptForAgentTab
-} from '@/lib/agent-launch-prompt-delivery'
+import { launchHostIsPaired } from '@/lib/launch-file-host'
+import { deliverNewTabLaunchPrompt } from '@/lib/launch-agent-new-tab-prompt-delivery'
 import { initialAgentTabViewModeProps } from '@/lib/native-chat-initial-view-mode'
 import { isNativeChatTranscriptLocalReadable } from '@/lib/native-chat-transcript-readability'
 import { getRuntimeEnvironmentIdForWorktree } from '@/lib/worktree-runtime-owner'
@@ -157,14 +153,14 @@ function launchAgentInNewTabInternal(args: LaunchAgentInNewTabArgs): LaunchAgent
     sessionOptions: resolveInitialNativeChatSessionOptions(store.settings, initialViewModeOptions)
   }
   const runtimeEnvironmentId = getRuntimeEnvironmentIdForWorktree(store, worktreeId)
-  const launchesOnPairedHost = isWebRuntimeSessionActive(runtimeEnvironmentId)
+  const hostPublishesTab = isWebRuntimeSessionActive(runtimeEnvironmentId)
   const { startupPlan, launchFile, pasteDraftAfterLaunch, submitPastedPrompt } =
     planLaunchAgentStartupPrompt({
       base: startupPlanBase,
       prompt: trimmedPrompt,
       promptDelivery,
       isFollowupPath,
-      launchesOnPairedHost
+      launchHostIsPaired: launchHostIsPaired(runtimeEnvironmentId)
     })
   let promptDeliveryResult: Promise<{ delivered: boolean; failureNotified: boolean }> | undefined
 
@@ -172,7 +168,7 @@ function launchAgentInNewTabInternal(args: LaunchAgentInNewTabArgs): LaunchAgent
     return null
   }
 
-  if (launchesOnPairedHost) {
+  if (hostPublishesTab) {
     if (beforeSurfaceOpen?.({ kind: 'host-published' }) === false) {
       return null
     }
@@ -246,6 +242,7 @@ function launchAgentInNewTabInternal(args: LaunchAgentInNewTabArgs): LaunchAgent
   if (beforeSurfaceOpen?.({ kind: 'local-terminal' }) === false) {
     return null
   }
+  const launchedAt = Date.now()
   // Why: queue startup BEFORE TerminalPane mounts — it snapshots pendingStartupByTabId in useState on first render.
   // Why: followup path pastes an unsubmitted draft, so gate the initial chat view like a draft launch, not auto-submit.
   const tab = store.createTab(worktreeId, groupId, undefined, {
@@ -270,6 +267,9 @@ function launchAgentInNewTabInternal(args: LaunchAgentInNewTabArgs): LaunchAgent
       ? { startupCommandDelivery: startupPlan.startupCommandDelivery }
       : {}),
     ...(launchFile ? { launchFile } : {}),
+    ...(hasPrompt && pasteDraftAfterLaunch === null && promptDelivery !== 'draft'
+      ? { launchPrompt: trimmedPrompt }
+      : {}),
     // Why: Command Code has no prompt-submit hook, so a prompt its launch line submits seeds working.
     ...(agent === 'command-code' &&
     hasPrompt &&
@@ -283,45 +283,20 @@ function launchAgentInNewTabInternal(args: LaunchAgentInNewTabArgs): LaunchAgent
       request_kind: 'new'
     }
   })
-  // Why: fire-and-forget the paste-after-ready delivery so callers keep the synchronous { tabId, startupPlan } signature.
-  // Why: safe to call unconditionally — the helper short-circuits (no paste) for native-prefill agents already holding the draft.
-  // Why: no paste runs, so no paste seeds the chat's copy: a draft rode in on argv (Claude --prefill
-  // etc.), and a submitted prompt rode the launch line. Not with a launch file: the transcript then
-  // shows the pointer sentence, which would never prune this copy.
-  if (hasPrompt && pasteDraftAfterLaunch === null && !launchFile) {
-    if (promptDelivery === 'draft') {
-      seedNativeChatLaunchDraftForAgentTab({ tabId: tab.id, agent, text: trimmedPrompt })
-    } else if (promptDelivery === 'submit-after-ready') {
-      seedNativeChatLaunchPromptForAgentTab({ tabId: tab.id, agent, text: trimmedPrompt })
-    }
-  }
-  if (pasteDraftAfterLaunch !== null) {
-    const timeoutNotice = createPasteReadinessTimeoutNotice({
-      worktreeId,
-      tabId: tab.id,
-      agent,
-      submitted: submitPastedPrompt,
-      content: pasteDraftAfterLaunch
-    })
-    const deliveryPromise = deliverLaunchPromptToAgentTab({
-      tabId: tab.id,
-      content: pasteDraftAfterLaunch,
-      agent,
-      submit: submitPastedPrompt,
-      forcePaste: true,
-      onTimeout: timeoutNotice.onTimeout,
-      ...(onPromptDeliveryUnconfirmed ? { onUnconfirmedDelivery: onPromptDeliveryUnconfirmed } : {})
-    }).then((delivered) => timeoutNotice.settle(delivered, onPromptDelivered))
-    if (promptDelivery === 'submit-after-ready') {
-      promptDeliveryResult = deliveryPromise
-    } else {
-      void deliveryPromise.catch((error) =>
-        console.error('Prompt delivery failed after launch', error)
-      )
-    }
-  } else if (hasPrompt) {
-    onPromptDelivered?.()
-  }
+  // Why: fire-and-forget the delivery so callers keep the synchronous { tabId, startupPlan } signature.
+  promptDeliveryResult = deliverNewTabLaunchPrompt({
+    worktreeId,
+    tabId: tab.id,
+    agent,
+    prompt: trimmedPrompt,
+    promptDelivery,
+    pasteDraftAfterLaunch,
+    submitPastedPrompt,
+    promptInLaunchFile: launchFile !== undefined,
+    launchedAt,
+    ...(onPromptDelivered ? { onPromptDelivered } : {}),
+    ...(onPromptDeliveryUnconfirmed ? { onPromptDeliveryUnconfirmed } : {})
+  })
 
   // Why: without setActiveTabType('terminal') an activated launch can stay hidden behind an editor.
   // Scoped to the launch's worktree so a floating or background launch leaves the main window's tab alone.
diff --git a/src/renderer/src/lib/launch-agent-new-tab-prompt-delivery.ts b/src/renderer/src/lib/launch-agent-new-tab-prompt-delivery.ts
new file mode 100644
index 00000000000..4bb2e43ab66
--- /dev/null
+++ b/src/renderer/src/lib/launch-agent-new-tab-prompt-delivery.ts
@@ -0,0 +1,83 @@
+import { createPasteReadinessTimeoutNotice } from '@/lib/launch-agent-paste-timeout-notice'
+import { waitForLaunchPromptReceipt } from '@/lib/agent-launch-prompt-receipt'
+import {
+  deliverLaunchPromptToAgentTab,
+  seedNativeChatLaunchDraftForAgentTab,
+  seedNativeChatLaunchPromptForAgentTab
+} from '@/lib/agent-launch-prompt-delivery'
+import type { TuiAgent } from '../../../shared/tui-agent'
+
+export type NewTabPromptDeliveryResult = Promise<{ delivered: boolean; failureNotified: boolean }>
+
+/**
+ * How a new agent tab's prompt is delivered, and when it counts as delivered: a paste once the agent
+ * is ready, or for a prompt the launch carried, the agent's receipt of it. Never the tab existing.
+ * Returns the result a `submit-after-ready` caller waits on; other callers fire and forget.
+ */
+export function deliverNewTabLaunchPrompt(args: {
+  worktreeId: string
+  tabId: string
+  agent: TuiAgent
+  /** Trimmed; empty for a launch with no prompt. */
+  prompt: string
+  promptDelivery: 'auto-submit' | 'draft' | 'submit-after-ready'
+  pasteDraftAfterLaunch: string | null
+  submitPastedPrompt: boolean
+  promptInLaunchFile: boolean
+  launchedAt: number
+  onPromptDelivered?: () => void
+  onPromptDeliveryUnconfirmed?: () => void
+}): NewTabPromptDeliveryResult | undefined {
+  const { tabId, agent, prompt, promptDelivery, pasteDraftAfterLaunch } = args
+  if (!prompt) {
+    return undefined
+  }
+  const unconfirmed = args.onPromptDeliveryUnconfirmed
+    ? { onUnconfirmedDelivery: args.onPromptDeliveryUnconfirmed }
+    : {}
+  // Why: no paste runs, so no paste seeds the chat's copy: a draft rode in on argv (Claude --prefill
+  // etc.), and a submitted prompt rode the launch line. Not with a launch file: the transcript then
+  // shows the pointer sentence, which would never prune this copy.
+  if (pasteDraftAfterLaunch === null && !args.promptInLaunchFile) {
+    if (promptDelivery === 'draft') {
+      seedNativeChatLaunchDraftForAgentTab({ tabId, agent, text: prompt })
+    } else if (promptDelivery === 'submit-after-ready') {
+      seedNativeChatLaunchPromptForAgentTab({ tabId, agent, text: prompt })
+    }
+  }
+  if (pasteDraftAfterLaunch === null && promptDelivery === 'draft') {
+    // A draft rode the launch command as an editable prefill; there is no turn to wait for.
+    args.onPromptDelivered?.()
+    return undefined
+  }
+  const content = pasteDraftAfterLaunch ?? prompt
+  const submitted = pasteDraftAfterLaunch === null || args.submitPastedPrompt
+  const timeoutNotice = createPasteReadinessTimeoutNotice({
+    worktreeId: args.worktreeId,
+    tabId,
+    agent,
+    submitted,
+    content
+  })
+  // Why the receipt for a carried prompt: callers post replies and resolve threads on this result.
+  const delivery =
+    pasteDraftAfterLaunch === null
+      ? waitForLaunchPromptReceipt({ tabId, agent, launchedAt: args.launchedAt, ...unconfirmed })
+      : deliverLaunchPromptToAgentTab({
+          tabId,
+          content,
+          agent,
+          submit: submitted,
+          forcePaste: true,
+          onTimeout: timeoutNotice.onTimeout,
+          ...unconfirmed
+        })
+  const result = delivery.then((delivered) =>
+    timeoutNotice.settle(delivered, args.onPromptDelivered)
+  )
+  if (promptDelivery === 'submit-after-ready') {
+    return result
+  }
+  void result.catch((error) => console.error('Prompt delivery failed after launch', error))
+  return undefined
+}
diff --git a/src/renderer/src/lib/launch-agent-startup-prompt-plan.ts b/src/renderer/src/lib/launch-agent-startup-prompt-plan.ts
index cd50ddff291..f82379f99a4 100644
--- a/src/renderer/src/lib/launch-agent-startup-prompt-plan.ts
+++ b/src/renderer/src/lib/launch-agent-startup-prompt-plan.ts
@@ -1,16 +1,12 @@
 import {
   buildAgentDraftLaunchPlan,
   buildAgentStartupPlan,
-  type AgentStartupPlan
+  planLaunchPrompt,
+  type AgentStartupPlan,
+  type AgentStartupPlanInputs
 } from '@/lib/tui-agent-startup'
-import { planStartupWithLaunchPrompt } from '../../../shared/startup-line-prompt-carry'
 import type { LaunchFile } from '../../../shared/launch-prompt-file'
 
-type StartupPlanBase = Omit<
-  Parameters<typeof buildAgentStartupPlan>[0],
-  'prompt' | 'allowEmptyPromptLaunch'
->
-
 export type LaunchAgentStartupPromptPlan = {
   startupPlan: AgentStartupPlan | null
   /** Written by the host before it types the launch line naming it. */
@@ -21,37 +17,41 @@ export type LaunchAgentStartupPromptPlan = {
 }
 
 /**
- * Decide how a new-tab launch delivers its prompt: agents whose CLI takes one get it on the launch
- * command, while agents that take text only after start launch clean and paste once ready.
+ * Decide how a new-tab launch delivers its prompt: on the agent's launch command or in a launch file
+ * where `carryLaunchPrompt` says it can ride, else pasted once the agent is ready.
  */
 export function planLaunchAgentStartupPrompt(args: {
-  base: StartupPlanBase
+  base: AgentStartupPlanInputs
   /** Already trimmed. */
   prompt: string
   promptDelivery: 'auto-submit' | 'draft' | 'submit-after-ready'
   isFollowupPath: boolean
-  /** A paired host of unknown version may neither stage a long line nor write a launch file. */
-  launchesOnPairedHost: boolean
+  /** See `launchHostIsPaired`. */
+  launchHostIsPaired: boolean
 }): LaunchAgentStartupPromptPlan {
   const { base, prompt, promptDelivery, isFollowupPath } = args
-  const hasPrompt = prompt.length > 0
-  const pasteAfterReady = (submit: boolean): LaunchAgentStartupPromptPlan => ({
-    startupPlan: buildAgentStartupPlan({ ...base, prompt: '', allowEmptyPromptLaunch: true }),
-    pasteDraftAfterLaunch: prompt,
+  const pasteAfterReady = (
+    startupPlan: AgentStartupPlan | null,
+    text: string,
+    submit: boolean
+  ): LaunchAgentStartupPromptPlan => ({
+    startupPlan,
+    pasteDraftAfterLaunch: text,
     submitPastedPrompt: submit
   })
+  const cleanPlan = (): AgentStartupPlan | null =>
+    buildAgentStartupPlan({ ...base, allowEmptyPromptLaunch: true })
 
-  if (hasPrompt && promptDelivery === 'draft') {
+  if (prompt.length > 0 && promptDelivery === 'draft') {
     const draftLaunchPlan = buildAgentDraftLaunchPlan({ ...base, draft: prompt })
     if (!draftLaunchPlan) {
-      return pasteAfterReady(false)
+      return pasteAfterReady(cleanPlan(), prompt, false)
     }
     return {
       startupPlan: {
         agent: draftLaunchPlan.agent,
         launchCommand: draftLaunchPlan.launchCommand,
         expectedProcess: draftLaunchPlan.expectedProcess,
-        followupPrompt: null,
         launchConfig: draftLaunchPlan.launchConfig,
         ...(draftLaunchPlan.sessionOptions
           ? { sessionOptions: draftLaunchPlan.sessionOptions }
@@ -65,29 +65,35 @@ export function planLaunchAgentStartupPrompt(args: {
       submitPastedPrompt: false
     }
   }
-  // Temporary, until paired hosts advertise staging: keep their paste after readiness.
-  if (
-    hasPrompt &&
-    (isFollowupPath || (args.launchesOnPairedHost && promptDelivery === 'submit-after-ready'))
-  ) {
-    return pasteAfterReady(promptDelivery === 'submit-after-ready')
+  if (prompt.length > 0 && isFollowupPath) {
+    return pasteAfterReady(cleanPlan(), prompt, promptDelivery === 'submit-after-ready')
   }
-  if (!hasPrompt || args.launchesOnPairedHost) {
-    const startupPlan = buildAgentStartupPlan({
-      ...base,
-      prompt,
-      allowEmptyPromptLaunch: !hasPrompt
-    })
-    // A paired host is sent a command, never a launch file, so a prompt that needs one is pasted.
-    return startupPlan?.launchFile
-      ? pasteAfterReady(true)
-      : { startupPlan, pasteDraftAfterLaunch: null, submitPastedPrompt: false }
+  // Why: a caller that waits for delivery gets a verdict from a paste; this client cannot observe a
+  // paired host's carried prompt reach its agent. Temporary, until paired hosts report that receipt.
+  if (prompt.length > 0 && args.launchHostIsPaired && promptDelivery === 'submit-after-ready') {
+    return pasteAfterReady(cleanPlan(), prompt, true)
   }
-  const carried = planStartupWithLaunchPrompt(base, prompt)
-  return {
-    startupPlan: carried.plan,
-    ...(carried.launchFile ? { launchFile: carried.launchFile } : {}),
-    pasteDraftAfterLaunch: null,
-    submitPastedPrompt: false
+  const planned = planLaunchPrompt({
+    ...base,
+    prompt,
+    launchHostIsPaired: args.launchHostIsPaired
+  })
+  if (!planned) {
+    return { startupPlan: null, pasteDraftAfterLaunch: null, submitPastedPrompt: false }
+  }
+  switch (planned.carry) {
+    case 'none':
+    case 'on-line':
+      return { startupPlan: planned.plan, pasteDraftAfterLaunch: null, submitPastedPrompt: false }
+    case 'launch-file':
+      return {
+        startupPlan: planned.plan,
+        launchFile: planned.launchFile,
+        pasteDraftAfterLaunch: null,
+        submitPastedPrompt: false
+      }
+    case 'paste-after-ready':
+      // Every non-draft new-tab prompt is submitted; a draft returned above.
+      return pasteAfterReady(planned.cleanPlan, planned.text, true)
   }
 }
diff --git a/src/renderer/src/lib/launch-file-host.test.ts b/src/renderer/src/lib/launch-file-host.test.ts
index c9fe3fa233a..299c4672db5 100644
--- a/src/renderer/src/lib/launch-file-host.test.ts
+++ b/src/renderer/src/lib/launch-file-host.test.ts
@@ -1,17 +1,17 @@
 import { afterEach, describe, expect, it, vi } from 'vitest'
-import { launchHostWritesLaunchFile } from './launch-file-host'
+import { launchHostIsPaired } from './launch-file-host'
 import { buildQuickComposerStartup } from '@/hooks/composer-state/quick-startup-plan'
 
-describe('whether a launch host writes its launch file', () => {
+describe('whether a launch lands on a paired host', () => {
   afterEach(() => {
     vi.unstubAllGlobals()
   })
 
-  it('is true for this machine and false for a paired host or a web client', () => {
-    expect(launchHostWritesLaunchFile({ activeRuntimeEnvironmentId: null })).toBe(true)
-    expect(launchHostWritesLaunchFile({ activeRuntimeEnvironmentId: 'env-1' })).toBe(false)
+  it('is false for this machine and true for another Orca or a web client', () => {
+    expect(launchHostIsPaired(null)).toBe(false)
+    expect(launchHostIsPaired('env-1')).toBe(true)
     vi.stubGlobal('window', { __ORCA_WEB_CLIENT__: true })
-    expect(launchHostWritesLaunchFile({ activeRuntimeEnvironmentId: null })).toBe(false)
+    expect(launchHostIsPaired(null)).toBe(true)
   })
 })
 
@@ -26,11 +26,11 @@ describe('the quick composer on a paired host', () => {
       platform: 'win32',
       shell: 'powershell',
       isRemote: false,
-      hostWritesLaunchFile: false,
+      launchHostIsPaired: true,
       telemetrySource: 'sidebar'
     })
     expect(startup.backendStartup).toBeUndefined()
     expect(startup.startupPlan?.launchFile).toBeUndefined()
-    expect(startup.startupPlan?.followupPrompt).toBe('fix the build\nthen run the tests')
+    expect(startup.startupPlan?.pastePromptAfterReady).toBe('fix the build\nthen run the tests')
   })
 })
diff --git a/src/renderer/src/lib/launch-file-host.ts b/src/renderer/src/lib/launch-file-host.ts
index 087e0e1402b..cb49325168c 100644
--- a/src/renderer/src/lib/launch-file-host.ts
+++ b/src/renderer/src/lib/launch-file-host.ts
@@ -1,14 +1,12 @@
-import type { GlobalSettings } from '../../../shared/global-settings-types'
-import { getActiveRuntimeTarget } from '@/runtime/runtime-client-target'
+import { isWebRuntimeSessionActive } from '@/runtime/web-runtime-session-environment'
 import { isWebClientLocation } from './web-client-location'
 
 /**
- * Whether the host a launch lands on writes the launch file its line names. A paired host (another
- * Orca this client drives) is sent a command and never the file, so a prompt that would need one is
- * pasted after the agent is ready instead. Temporary, until paired hosts are sent the prompt itself.
+ * Whether a launch lands on a paired Orca: another Orca this client drives, which may be an older
+ * build that neither stages a long line nor writes a launch file. `carryLaunchPrompt` then keeps
+ * the line to the typed budget and pastes the rest after the agent is ready. Temporary, until paired
+ * hosts advertise both.
  */
-export function launchHostWritesLaunchFile(
-  ownerSettings: Pick<GlobalSettings, 'activeRuntimeEnvironmentId'> | null | undefined
-): boolean {
-  return !isWebClientLocation() && getActiveRuntimeTarget(ownerSettings).kind === 'local'
+export function launchHostIsPaired(runtimeEnvironmentId: string | null | undefined): boolean {
+  return isWebClientLocation() || isWebRuntimeSessionActive(runtimeEnvironmentId)
 }
diff --git a/src/renderer/src/lib/launch-work-item-direct-agent-routing.ts b/src/renderer/src/lib/launch-work-item-direct-agent-routing.ts
index 639bce4be6f..8dee872c816 100644
--- a/src/renderer/src/lib/launch-work-item-direct-agent-routing.ts
+++ b/src/renderer/src/lib/launch-work-item-direct-agent-routing.ts
@@ -18,7 +18,7 @@ export function buildDirectWorkItemStartup(args: {
   launchConnectionId: string | null
   worktreePath: string
   repoProjectRuntime?: Parameters<typeof resolveSourceControlLaunchPlatform>[0]['projectRuntime']
-  launchesOnPairedHost: boolean
+  launchHostIsPaired: boolean
 }): ReturnType<typeof buildDirectWorkItemAgentStartupPlan> {
   const launchPlatform =
     args.launchPlatform ??
@@ -39,7 +39,7 @@ export function buildDirectWorkItemStartup(args: {
     ),
     // Why: SSH hosts run the plain `orca` shim, so the Linux-only `orca-ide` rename is not applied.
     isRemote: typeof args.launchConnectionId === 'string',
-    launchesOnPairedHost: args.launchesOnPairedHost
+    launchHostIsPaired: args.launchHostIsPaired
   })
 }
 
diff --git a/src/renderer/src/lib/launch-work-item-direct-agent.test.ts b/src/renderer/src/lib/launch-work-item-direct-agent.test.ts
index b5358ec6b3a..44b38b65066 100644
--- a/src/renderer/src/lib/launch-work-item-direct-agent.test.ts
+++ b/src/renderer/src/lib/launch-work-item-direct-agent.test.ts
@@ -26,7 +26,6 @@ describe('buildDirectWorkItemStartupOpts', () => {
       agent: 'codex',
       launchCommand: "codex 'review linked issue'",
       expectedProcess: 'codex',
-      followupPrompt: null,
       launchConfig: { agentArgs: '', agentEnv: {} },
       startupCommandDelivery: 'shell-ready'
     }
@@ -53,7 +52,6 @@ describe('buildDirectWorkItemStartupOpts', () => {
       agent: 'claude',
       launchCommand: "claude --prefill 'https://github.com/o/r/issues/12'",
       expectedProcess: 'claude',
-      followupPrompt: null,
       launchConfig: { agentArgs: '', agentEnv: {} }
     }
 
@@ -85,6 +83,7 @@ const settings = {
 describe('buildDirectWorkItemAgentStartupPlan', () => {
   it('omits native-chat preferences when the new workspace opens in terminal mode', () => {
     const result = buildDirectWorkItemAgentStartupPlan({
+      launchHostIsPaired: false,
       agent: 'codex',
       draftContent: 'Review issue 42',
       promptDelivery: 'draft',
@@ -99,6 +98,7 @@ describe('buildDirectWorkItemAgentStartupPlan', () => {
 
   it('applies native-chat preferences when the new workspace opens in chat', () => {
     const result = buildDirectWorkItemAgentStartupPlan({
+      launchHostIsPaired: false,
       agent: 'codex',
       draftContent: 'Review issue 42',
       promptDelivery: 'draft',
@@ -147,6 +147,7 @@ describe('buildDirectWorkItemAgentStartupPlan global arguments fallback', () =>
 
   it('resolves the global Agents arguments when the launch names none', () => {
     const result = buildDirectWorkItemAgentStartupPlan({
+      launchHostIsPaired: false,
       agent: 'codex',
       draftContent: 'Fix the broken checks',
       promptDelivery: 'draft',
@@ -160,6 +161,7 @@ describe('buildDirectWorkItemAgentStartupPlan global arguments fallback', () =>
 
   it('lets an explicit per-action value win over the global one', () => {
     const result = buildDirectWorkItemAgentStartupPlan({
+      launchHostIsPaired: false,
       agent: 'codex',
       agentArgs: '--model gpt-5',
       draftContent: 'Fix the broken checks',
@@ -173,3 +175,38 @@ describe('buildDirectWorkItemAgentStartupPlan global arguments fallback', () =>
     expect(result.startupPlan?.launchCommand).not.toContain('danger-full-access')
   })
 })
+
+describe('buildDirectWorkItemAgentStartupPlan submitted prompts', () => {
+  const submit = (agent: 'claude' | 'gemini', draftContent: string, launchHostIsPaired = false) =>
+    buildDirectWorkItemAgentStartupPlan({
+      launchHostIsPaired,
+      agent,
+      draftContent,
+      promptDelivery: 'submit-after-ready',
+      settings: { ...settings, openAgentTabsInChatByDefault: false },
+      launchPlatform: 'darwin',
+      nativeChatTranscriptIsLocalReadable: true
+    })
+
+  // The bug class the explicit carry outcome removes: a plan that exists but does not carry the
+  // prompt was reported as on the launch command, so no paste ran and the prompt was dropped.
+  it('leaves a file-sized prompt for the paste for an agent not measured reading the file', () => {
+    const result = submit('gemini', 'g'.repeat(20_000))
+    expect(result.promptOnLaunchCommand).toBe(false)
+    expect(result.startupPlan?.launchCommand).not.toContain('gggg')
+    expect(result.launchFile).toBeUndefined()
+  })
+
+  it('hands Claude the same prompt in a launch file, with the prompt to copy if refused', () => {
+    const prompt = 'c'.repeat(20_000)
+    const result = submit('claude', prompt)
+    expect(result.promptOnLaunchCommand).toBe(true)
+    expect(result.launchFile?.content).toBe(prompt)
+    expect(result.launchPrompt).toBe(prompt)
+  })
+
+  it('pastes on a paired host what its line cannot carry typed', () => {
+    expect(submit('claude', 'fix it', true).promptOnLaunchCommand).toBe(true)
+    expect(submit('claude', 'fix it\nthen run the tests', true).promptOnLaunchCommand).toBe(false)
+  })
+})
diff --git a/src/renderer/src/lib/launch-work-item-direct-agent.ts b/src/renderer/src/lib/launch-work-item-direct-agent.ts
index 2d3e50cee4b..cdd32ec8bed 100644
--- a/src/renderer/src/lib/launch-work-item-direct-agent.ts
+++ b/src/renderer/src/lib/launch-work-item-direct-agent.ts
@@ -3,6 +3,7 @@ import { track, tuiAgentToAgentKind } from '@/lib/telemetry'
 import {
   buildAgentDraftLaunchPlan,
   buildAgentStartupPlan,
+  planLaunchPrompt,
   type AgentStartupPlan
 } from '@/lib/tui-agent-startup'
 import type { AgentStartedTelemetry } from '@/lib/worktree-startup-payload'
@@ -18,8 +19,6 @@ import { translate } from '@/i18n/i18n'
 import { resolveInitialNativeChatSessionOptions } from '@/components/native-chat/native-chat-launch-session-options'
 import type { PersistedNativeChatSessionOptions } from '../../../shared/native-chat-session-options'
 import type { LaunchFile } from '../../../shared/launch-prompt-file'
-import { planStartupWithLaunchPrompt } from '../../../shared/startup-line-prompt-carry'
-import { agentPromptRidesLaunchCommand } from '../../../shared/tui-agent-startup'
 
 export function buildDirectWorkItemAgentStartupPlan(args: {
   agent: TuiAgent | null
@@ -42,11 +41,13 @@ export function buildDirectWorkItemAgentStartupPlan(args: {
   /** Why: SSH remotes deploy the CLI shim as plain `orca`, so the Linux-only
    * `orca-ide` rename must not be applied for remote launches. */
   isRemote?: boolean
-  /** A paired host of unknown version may neither stage a long line nor write a launch file. */
-  launchesOnPairedHost?: boolean
+  /** See `launchHostIsPaired`. */
+  launchHostIsPaired: boolean
 }): {
   startupPlan: AgentStartupPlan | null
   launchFile?: LaunchFile
+  /** The submitted prompt the launch command carries (on its line or in `launchFile`). */
+  launchPrompt?: string
   /** The launch command carries the content (a native draft or a submitted prompt); no paste runs. */
   promptOnLaunchCommand: boolean
   startupPlanFailed: boolean
@@ -76,33 +77,51 @@ export function buildDirectWorkItemAgentStartupPlan(args: {
     agentEnv: effectiveAgentEnv,
     sessionOptions
   }
-  // Temporary, until paired hosts advertise staging: they keep the paste after readiness.
-  if (
-    args.promptDelivery === 'submit-after-ready' &&
-    !args.launchesOnPairedHost &&
-    agentPromptRidesLaunchCommand(args.agent)
-  ) {
-    const carried = planStartupWithLaunchPrompt(planInputs, args.draftContent)
-    return {
-      startupPlan: carried.plan,
-      ...(carried.launchFile ? { launchFile: carried.launchFile } : {}),
-      promptOnLaunchCommand: carried.plan !== null,
-      startupPlanFailed: carried.plan === null
+  if (args.promptDelivery === 'submit-after-ready') {
+    const planned = planLaunchPrompt({
+      ...planInputs,
+      prompt: args.draftContent,
+      launchHostIsPaired: args.launchHostIsPaired
+    })
+    if (!planned) {
+      return { startupPlan: null, promptOnLaunchCommand: false, startupPlanFailed: true }
+    }
+    switch (planned.carry) {
+      case 'none':
+        return { startupPlan: planned.plan, promptOnLaunchCommand: true, startupPlanFailed: false }
+      case 'on-line':
+        return {
+          startupPlan: planned.plan,
+          launchPrompt: args.draftContent,
+          promptOnLaunchCommand: true,
+          startupPlanFailed: false
+        }
+      case 'launch-file':
+        return {
+          startupPlan: planned.plan,
+          launchFile: planned.launchFile,
+          launchPrompt: args.draftContent,
+          promptOnLaunchCommand: true,
+          startupPlanFailed: false
+        }
+      case 'paste-after-ready':
+        return {
+          startupPlan: planned.cleanPlan,
+          promptOnLaunchCommand: false,
+          startupPlanFailed: false
+        }
     }
   }
-  const draftLaunchPlan =
-    args.promptDelivery === 'submit-after-ready'
-      ? null
-      : buildAgentDraftLaunchPlan({
-          agent: args.agent,
-          draft: args.draftContent,
-          cmdOverrides: args.settings?.agentCmdOverrides ?? {},
-          platform: args.launchPlatform,
-          isRemote: args.isRemote,
-          agentArgs: effectiveAgentArgs,
-          agentEnv: effectiveAgentEnv,
-          sessionOptions
-        })
+  const draftLaunchPlan = buildAgentDraftLaunchPlan({
+    agent: args.agent,
+    draft: args.draftContent,
+    cmdOverrides: args.settings?.agentCmdOverrides ?? {},
+    platform: args.launchPlatform,
+    isRemote: args.isRemote,
+    agentArgs: effectiveAgentArgs,
+    agentEnv: effectiveAgentEnv,
+    sessionOptions
+  })
 
   if (draftLaunchPlan) {
     return {
@@ -110,7 +129,6 @@ export function buildDirectWorkItemAgentStartupPlan(args: {
         agent: draftLaunchPlan.agent,
         launchCommand: draftLaunchPlan.launchCommand,
         expectedProcess: draftLaunchPlan.expectedProcess,
-        followupPrompt: null,
         launchConfig: draftLaunchPlan.launchConfig,
         ...(draftLaunchPlan.sessionOptions
           ? { sessionOptions: draftLaunchPlan.sessionOptions }
@@ -153,7 +171,9 @@ export function buildDirectWorkItemStartupOpts(
   /** Unsent launch context, for the view-mode decision only. Set it for every
    *  draft launch — a natively-prefilled plan carries no `draftPrompt`. */
   launchDraftText?: string,
-  launchFile?: LaunchFile
+  /** What the launch command carries: its launch file, and the submitted prompt (see
+   *  `PtyPaneStartup.launchPrompt`). */
+  carried: { launchFile?: LaunchFile; launchPrompt?: string } = {}
 ): {
   startup?: {
     command: string
@@ -165,6 +185,7 @@ export function buildDirectWorkItemStartupOpts(
     sessionOptions?: AgentStartupPlan['sessionOptions']
     startupCommandDelivery?: StartupCommandDelivery
     launchFile?: LaunchFile
+    launchPrompt?: string
     telemetry?: AgentStartedTelemetry
   }
 } {
@@ -187,7 +208,8 @@ export function buildDirectWorkItemStartupOpts(
       ...(plan.startupCommandDelivery
         ? { startupCommandDelivery: plan.startupCommandDelivery }
         : {}),
-      ...(launchFile ? { launchFile } : {}),
+      ...(carried.launchFile ? { launchFile: carried.launchFile } : {}),
+      ...(carried.launchPrompt ? { launchPrompt: carried.launchPrompt } : {}),
       ...(telemetry ? { telemetry } : {})
     }
   }
diff --git a/src/renderer/src/lib/launch-work-item-direct-route-preparation.ts b/src/renderer/src/lib/launch-work-item-direct-route-preparation.ts
index 5a4a38a40ce..f19f8632015 100644
--- a/src/renderer/src/lib/launch-work-item-direct-route-preparation.ts
+++ b/src/renderer/src/lib/launch-work-item-direct-route-preparation.ts
@@ -4,7 +4,7 @@ import { getConnectionId } from '@/lib/connection-context'
 import { CLIENT_PLATFORM } from '@/lib/new-workspace'
 import { getLocalProjectExecutionRuntimeContext } from '@/lib/local-preflight-context'
 import { getRuntimeEnvironmentIdForWorktree } from '@/lib/worktree-runtime-owner'
-import { isWebRuntimeSessionActive } from '@/runtime/web-runtime-session'
+import { launchHostIsPaired } from '@/lib/launch-file-host'
 import type { LaunchFile } from '../../../shared/launch-prompt-file'
 import type {
   AgentSessionLaunchPlan,
@@ -21,6 +21,7 @@ export type DirectWorkItemAgentLaunchPreparation = {
   effectiveAgent: TuiAgent | null
   startupPlan: ReturnType<typeof buildDirectWorkItemStartup>['startupPlan']
   launchFile?: LaunchFile
+  launchPrompt?: string
   promptOnLaunchCommand: boolean
   startupPlanFailed: boolean
   structuredLaunch: boolean
@@ -74,7 +75,7 @@ export async function prepareDirectWorkItemAgentLaunch(args: {
         // Non-critical: activation still has the explicit startup below.
       })
   }
-  const { startupPlan, launchFile, promptOnLaunchCommand, startupPlanFailed } =
+  const { startupPlan, launchFile, launchPrompt, promptOnLaunchCommand, startupPlanFailed } =
     buildDirectWorkItemStartup({
       agent: effectiveAgent,
       agentArgs: args.agentArgs,
@@ -92,7 +93,7 @@ export async function prepareDirectWorkItemAgentLaunch(args: {
               CLIENT_PLATFORM
             ) ?? args.repoProjectRuntime)
           : undefined,
-      launchesOnPairedHost: isWebRuntimeSessionActive(
+      launchHostIsPaired: launchHostIsPaired(
         getRuntimeEnvironmentIdForWorktree(args.latestStore, args.worktreeId)
       )
     })
@@ -115,6 +116,7 @@ export async function prepareDirectWorkItemAgentLaunch(args: {
     effectiveAgent,
     startupPlan,
     ...(launchFile ? { launchFile } : {}),
+    ...(launchPrompt ? { launchPrompt } : {}),
     promptOnLaunchCommand,
     startupPlanFailed,
     structuredLaunch,
diff --git a/src/renderer/src/lib/launch-work-item-direct.test.ts b/src/renderer/src/lib/launch-work-item-direct.test.ts
index ee14844e612..8c4c2cf3448 100644
--- a/src/renderer/src/lib/launch-work-item-direct.test.ts
+++ b/src/renderer/src/lib/launch-work-item-direct.test.ts
@@ -2,7 +2,6 @@ import { beforeEach, describe, expect, it, vi } from 'vitest'
 import type { AppState } from '@/store'
 import type * as TuiAgentSelectionModule from '../../../shared/tui-agent-selection'
 import type * as TuiAgentStartupModule from '@/lib/tui-agent-startup'
-import type * as PromptCarryModule from '../../../shared/startup-line-prompt-carry'
 import type * as DirectAgentRoutingModule from '@/lib/launch-work-item-direct-agent-routing'
 
 const mocks = vi.hoisted(() => ({
@@ -104,17 +103,8 @@ vi.mock('@/lib/tui-agent-startup', async () => {
   return {
     ...actual,
     buildAgentDraftLaunchPlan: vi.fn(actual.buildAgentDraftLaunchPlan),
-    buildAgentStartupPlan: vi.fn(actual.buildAgentStartupPlan)
-  }
-})
-
-vi.mock('../../../shared/startup-line-prompt-carry', async () => {
-  const actual = await vi.importActual<typeof PromptCarryModule>(
-    '../../../shared/startup-line-prompt-carry'
-  )
-  return {
-    ...actual,
-    planStartupWithLaunchPrompt: vi.fn(actual.planStartupWithLaunchPrompt)
+    buildAgentStartupPlan: vi.fn(actual.buildAgentStartupPlan),
+    planLaunchPrompt: vi.fn(actual.planLaunchPrompt)
   }
 })
 
@@ -141,9 +131,12 @@ vi.mock('@/lib/launch-work-item-direct-agent-routing', async () => {
 import { launchWorkItemDirect } from './launch-work-item-direct'
 import { pasteDraftWhenAgentReady } from '@/lib/agent-paste-draft'
 import { beginDirectWorkItemStructuredLaunch } from '@/lib/launch-work-item-direct-agent-routing'
-import { buildAgentDraftLaunchPlan, buildAgentStartupPlan } from '@/lib/tui-agent-startup'
+import {
+  buildAgentDraftLaunchPlan,
+  buildAgentStartupPlan,
+  planLaunchPrompt
+} from '@/lib/tui-agent-startup'
 import { pickTuiAgent } from '../../../shared/tui-agent-selection'
-import { planStartupWithLaunchPrompt } from '../../../shared/startup-line-prompt-carry'
 
 const mockApi = {
   worktrees: {
@@ -600,7 +593,6 @@ describe('launchWorkItemDirect', () => {
       agent: 'cursor',
       launchCommand: 'cursor-agent',
       expectedProcess: 'cursor-agent',
-      followupPrompt: null,
       launchConfig: { agentArgs: '', agentEnv: {} }
     })
     mocks.store.createWorktree.mockResolvedValue({
@@ -813,12 +805,12 @@ describe('launchWorkItemDirect', () => {
       })
     ).resolves.toBe(true)
 
-    expect(planStartupWithLaunchPrompt).toHaveBeenCalledWith(
+    expect(planLaunchPrompt).toHaveBeenCalledWith(
       expect.objectContaining({
         agent: 'codex',
-        platform: 'linux'
-      }),
-      'Fix the failing checks.'
+        platform: 'linux',
+        prompt: 'Fix the failing checks.'
+      })
     )
   })
 })
diff --git a/src/renderer/src/lib/launch-work-item-direct.ts b/src/renderer/src/lib/launch-work-item-direct.ts
index ac4d58a7b58..fe1180d2a99 100644
--- a/src/renderer/src/lib/launch-work-item-direct.ts
+++ b/src/renderer/src/lib/launch-work-item-direct.ts
@@ -249,7 +249,7 @@ export async function launchWorkItemDirect(args: LaunchWorkItemDirectArgs): Prom
               startupPlan,
               launchSource,
               promptDelivery === 'draft' ? draftContent : undefined,
-              launchPreparation.launchFile
+              launchPreparation
             ))
       })
       return activationHolder.value !== false
diff --git a/src/renderer/src/lib/new-workspace.test.ts b/src/renderer/src/lib/new-workspace.test.ts
index 28fab8326ff..b6d130b6656 100644
--- a/src/renderer/src/lib/new-workspace.test.ts
+++ b/src/renderer/src/lib/new-workspace.test.ts
@@ -288,7 +288,7 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
         agent: 'aider',
         launchCommand: 'aider',
         expectedProcess: 'aider',
-        followupPrompt: 'fix the spinner',
+        pastePromptAfterReady: 'fix the spinner',
         launchConfig: { agentArgs: '', agentEnv: {} }
       }
     })
@@ -311,7 +311,7 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
         agent: 'aider',
         launchCommand: 'aider',
         expectedProcess: 'aider',
-        followupPrompt: 'fix the spinner',
+        pastePromptAfterReady: 'fix the spinner',
         launchConfig: { agentArgs: '', agentEnv: {} }
       }
     })
@@ -333,7 +333,7 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
         agent: 'aider',
         launchCommand: 'aider',
         expectedProcess: 'aider',
-        followupPrompt: 'fix the spinner',
+        pastePromptAfterReady: 'fix the spinner',
         launchConfig: { agentArgs: '', agentEnv: {} }
       }
     })
@@ -355,7 +355,7 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
         agent: 'claude',
         launchCommand: 'claude',
         expectedProcess: 'claude',
-        followupPrompt: 'fix the spinner',
+        pastePromptAfterReady: 'fix the spinner',
         launchConfig: { agentArgs: '', agentEnv: {} }
       }
     })
@@ -374,7 +374,7 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
         agent: 'aider',
         launchCommand: 'aider',
         expectedProcess: 'aider',
-        followupPrompt: 'fix the spinner',
+        pastePromptAfterReady: 'fix the spinner',
         launchConfig: { agentArgs: '', agentEnv: {} }
       }
     })
@@ -389,7 +389,6 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
         agent: 'claude',
         launchCommand: 'claude',
         expectedProcess: 'claude',
-        followupPrompt: null,
         launchConfig: { agentArgs: '', agentEnv: {} },
         draftPrompt: 'review this before sending'
       }
@@ -416,7 +415,7 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
           agent: 'aider',
           launchCommand: 'aider',
           expectedProcess: 'aider',
-          followupPrompt: 'fix the spinner',
+          pastePromptAfterReady: 'fix the spinner',
           launchConfig: { agentArgs: '', agentEnv: {} }
         }
       })
@@ -432,7 +431,6 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
         agent: 'claude',
         launchCommand: 'claude',
         expectedProcess: 'claude',
-        followupPrompt: null,
         launchConfig: { agentArgs: '', agentEnv: {} },
         draftPrompt: 'review this before sending'
       }
@@ -476,7 +474,6 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
         agent: 'codex',
         launchCommand: 'codex',
         expectedProcess: 'codex',
-        followupPrompt: null,
         launchConfig: { agentArgs: '', agentEnv: {} },
         draftPrompt: 'Linear context draft'
       }
@@ -506,7 +503,6 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
         agent: 'codex',
         launchCommand: 'codex',
         expectedProcess: 'codex',
-        followupPrompt: null,
         launchConfig: { agentArgs: '', agentEnv: {} },
         draftPrompt: 'https://github.com/stablyai/orca/pull/2051'
       }
@@ -563,7 +559,6 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
         agent: 'codex',
         launchCommand: 'codex',
         expectedProcess: 'codex',
-        followupPrompt: null,
         launchConfig: { agentArgs: '', agentEnv: {} },
         draftPrompt: 'linked draft'
       }
@@ -645,7 +640,6 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
         agent: 'codex',
         launchCommand: 'codex',
         expectedProcess: 'codex',
-        followupPrompt: null,
         launchConfig: { agentArgs: '', agentEnv: {} },
         draftPrompt: 'linked draft'
       }
@@ -684,7 +678,6 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
       agent: 'codex' as const,
       launchCommand: 'codex',
       expectedProcess: 'codex',
-      followupPrompt: null,
       launchConfig: { agentArgs: '', agentEnv: {} },
       draftPrompt: 'linked draft',
       launchToken: 'launch-token-1'
@@ -714,7 +707,6 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
       agent: 'codex' as const,
       launchCommand: 'codex',
       expectedProcess: 'codex',
-      followupPrompt: null,
       launchConfig: { agentArgs: '', agentEnv: {} },
       draftPrompt: 'linked draft',
       launchToken: 'launch-token-1'
@@ -775,7 +767,6 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
         agent: 'codex',
         launchCommand: 'codex',
         expectedProcess: 'codex',
-        followupPrompt: null,
         launchConfig: { agentArgs: '', agentEnv: {} },
         draftPrompt: 'old linked draft',
         launchToken: 'launch-token-old'
@@ -827,7 +818,7 @@ describe('ensureAgentStartupInTerminal prompt delivery', () => {
         agent: 'aider',
         launchCommand: 'aider',
         expectedProcess: 'aider',
-        followupPrompt: 'fix the spinner',
+        pastePromptAfterReady: 'fix the spinner',
         launchConfig: { agentArgs: '', agentEnv: {} }
       }
     })
diff --git a/src/renderer/src/lib/new-workspace.ts b/src/renderer/src/lib/new-workspace.ts
index 8a8d13f2652..e932c32811b 100644
--- a/src/renderer/src/lib/new-workspace.ts
+++ b/src/renderer/src/lib/new-workspace.ts
@@ -7,6 +7,7 @@ import {
 import { sendFollowupPromptWhenAgentReady } from '@/lib/agent-followup-delivery'
 import { showAgentLaunchPromptNotDeliveredNotice } from '@/lib/agent-launch-prompt-not-delivered-notice'
 import type { AgentStartupPlan } from '@/lib/tui-agent-startup'
+import type { ComposerAgentStartupPlan } from '@/lib/composer-agent-startup-plan'
 import type { LinkedWorkItemContext } from '@/lib/linked-work-item-context'
 import {
   beginAgentStartupDeliveryAttempt,
@@ -228,11 +229,11 @@ export function getWorkspaceSeedName(args: {
 export async function ensureAgentStartupInTerminal(args: {
   worktreeId: string
   primaryTabId?: string | null
-  startup: AgentStartupPlan
+  startup: ComposerAgentStartupPlan
 }): Promise<void> {
   const { worktreeId, primaryTabId, startup } = args
   const draftPrompt = startup.draftPrompt ?? null
-  if (startup.followupPrompt === null && draftPrompt === null) {
+  if (!startup.pastePromptAfterReady && draftPrompt === null) {
     return
   }
   const launchToken = ensureStartupLaunchToken(startup)
@@ -280,15 +281,14 @@ export async function ensureAgentStartupInTerminal(args: {
 async function deliverAgentStartupToTerminal(
   tabId: string,
   ptyId: string,
-  startup: AgentStartupPlan
+  startup: ComposerAgentStartupPlan
 ): Promise<void> {
   const draftPrompt = startup.draftPrompt ?? null
   const runtimeSettings = getSettingsForAgentTabRuntimeOwner(tabId)
-  // Why: followupPrompt is the legacy path for stdin-after-start agents
-  // (aider, goose, etc.) that need their initial prompt typed into the live
-  // session and submitted. Wait until the agent owns the PTY before writing.
-  if (startup.followupPrompt) {
-    const prompt = startup.followupPrompt
+  // Why: a stdin-after-start agent (aider, goose, etc.) needs its initial prompt typed into the
+  // live session and submitted. Wait until the agent owns the PTY before writing.
+  if (startup.pastePromptAfterReady) {
+    const prompt = startup.pastePromptAfterReady
     // Why: an argv agent's prompt left for paste (a WSL line too long to type, a paired host) waits
     // for the agent's composer and lands as one bracketed paste, not raw keys once the process shows.
     const delivered = agentPromptRidesLaunchCommand(startup.agent)
diff --git a/src/renderer/src/lib/pending-worktree-creation.ts b/src/renderer/src/lib/pending-worktree-creation.ts
index 1e703df742f..ef7536292e0 100644
--- a/src/renderer/src/lib/pending-worktree-creation.ts
+++ b/src/renderer/src/lib/pending-worktree-creation.ts
@@ -10,7 +10,7 @@ import type {
   WorkspaceLinkedItem,
   WorkspaceStatus
 } from '../../../shared/worktree/types'
-import type { AgentStartupPlan } from '@/lib/tui-agent-startup'
+import type { ComposerAgentStartupPlan } from '@/lib/composer-agent-startup-plan'
 import type { AgentStartedTelemetry } from '@/lib/worktree-startup-payload'
 import type { TaskSourceContext, WorkspaceRunContext } from '../../../shared/task-source-context'
 import type { AgentLaunchRoute } from '@/lib/agent-launch-routing'
@@ -103,7 +103,7 @@ export type WorktreeCreationRequest = {
   note: string
   /** Renderer-side launch plan used to seed the first terminal when the backend
    *  did not already spawn it. Null for blank-shell creates. */
-  startupPlan: AgentStartupPlan | null
+  startupPlan: ComposerAgentStartupPlan | null
   quickPrompt: string
   /** Launch context delivered only as an unsent TUI-input draft (argv prefill or
    *  startup paste); completion seeds the chat-composer copy from it. */
diff --git a/src/renderer/src/lib/source-control-agent-action-plan.ts b/src/renderer/src/lib/source-control-agent-action-plan.ts
index 223b3b879f0..b32d6cbc669 100644
--- a/src/renderer/src/lib/source-control-agent-action-plan.ts
+++ b/src/renderer/src/lib/source-control-agent-action-plan.ts
@@ -2,6 +2,7 @@ import {
   buildAgentDraftLaunchPlan,
   buildAgentStartupPlan,
   planAgentCliArgsSuffix,
+  planLaunchPrompt,
   type AgentStartupPlan
 } from '@/lib/tui-agent-startup'
 import { CLIENT_PLATFORM } from '@/lib/new-workspace'
@@ -14,6 +15,7 @@ import type { SessionOptionValue } from '../../../shared/native-chat-session-opt
 
 export type SourceControlLaunchPlanDelivery =
   | 'argv'
+  | 'argv-launch-file'
   | 'draft-native'
   | 'draft-paste'
   | 'paste-submit'
@@ -43,6 +45,8 @@ export function planSourceControlAgentActionLaunch(args: {
   /** Why: SSH remotes deploy the CLI shim as plain `orca`, so the Linux-only
    * `orca-ide` rename must not be applied for remote launches. */
   isRemote?: boolean
+  /** See `launchHostIsPaired`. */
+  launchHostIsPaired?: boolean
 }): SourceControlLaunchPlanResult {
   const agent = args.agent
   if (!agent) {
@@ -133,7 +137,6 @@ export function planSourceControlAgentActionLaunch(args: {
         agent: draftLaunchPlan.agent,
         launchCommand: draftLaunchPlan.launchCommand,
         expectedProcess: draftLaunchPlan.expectedProcess,
-        followupPrompt: null,
         launchConfig: draftLaunchPlan.launchConfig,
         ...(draftLaunchPlan.sessionOptions
           ? { sessionOptions: draftLaunchPlan.sessionOptions }
@@ -172,7 +175,7 @@ export function planSourceControlAgentActionLaunch(args: {
     })
     delivery = 'draft-paste'
   } else {
-    startupPlan = buildAgentStartupPlan({
+    const planned = planLaunchPrompt({
       agent,
       prompt: trimmedInput,
       cmdOverrides,
@@ -181,9 +184,26 @@ export function planSourceControlAgentActionLaunch(args: {
       isRemote,
       agentArgs: args.agentArgs,
       sessionOptions: args.sessionOptions,
-      allowEmptyPromptLaunch: false
+      launchHostIsPaired: args.launchHostIsPaired ?? false
     })
-    delivery = 'argv'
+    switch (planned?.carry) {
+      case undefined:
+        delivery = 'argv'
+        break
+      case 'none':
+      case 'on-line':
+        startupPlan = planned.plan
+        delivery = 'argv'
+        break
+      case 'launch-file':
+        startupPlan = planned.plan
+        delivery = 'argv-launch-file'
+        break
+      case 'paste-after-ready':
+        startupPlan = planned.cleanPlan
+        delivery = 'paste-submit'
+        break
+    }
   }
 
   if (!startupPlan) {
@@ -203,7 +223,7 @@ export function planSourceControlAgentActionLaunch(args: {
         ? 'The command input is prefilled as an editable draft by the agent launch command.'
         : delivery === 'draft-paste'
           ? 'The agent starts with no prompt, then Orca pastes the command input as an editable draft after the TUI is ready.'
-          : startupPlan.launchFile
+          : delivery === 'argv-launch-file'
             ? 'The command input goes in a private file the host writes; the launch command tells the agent to read it, submitted as the first turn.'
             : 'The command input is included in the launch command and submitted as the first turn.'
 
diff --git a/src/renderer/src/lib/tui-agent-startup.test.ts b/src/renderer/src/lib/tui-agent-startup.test.ts
index 7ce665544b3..095ac81920c 100644
--- a/src/renderer/src/lib/tui-agent-startup.test.ts
+++ b/src/renderer/src/lib/tui-agent-startup.test.ts
@@ -1,9 +1,6 @@
 import { describe, expect, it } from 'vitest'
-import {
-  buildAgentDraftLaunchPlan,
-  buildAgentStartupPlan,
-  isShellProcess
-} from './tui-agent-startup'
+import { planLaunchForTest } from '../../../shared/launch-prompt-plan.test-fixture'
+import { buildAgentDraftLaunchPlan, isShellProcess } from './tui-agent-startup'
 import { resolveTuiAgentLaunchArgs } from '../../../shared/tui-agent-launch-defaults'
 
 const emptyLaunchConfig = (agentCommand: string) => ({
@@ -15,7 +12,7 @@ const emptyLaunchConfig = (agentCommand: string) => ({
 describe('buildAgentStartupPlan', () => {
   it('passes Claude prompts as a positional interactive argument', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'claude',
         prompt: 'Fix the bug',
         cmdOverrides: {},
@@ -25,14 +22,15 @@ describe('buildAgentStartupPlan', () => {
       agent: 'claude',
       launchCommand: "claude 'Fix the bug'",
       expectedProcess: 'claude',
-      followupPrompt: null,
+      carry: 'on-line',
+      pasteAfterReady: null,
       launchConfig: emptyLaunchConfig('claude')
     })
   })
 
   it('uses Gemini interactive prompt mode instead of dropping the prompt', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'gemini',
         prompt: 'Investigate this regression',
         cmdOverrides: {},
@@ -42,14 +40,15 @@ describe('buildAgentStartupPlan', () => {
       agent: 'gemini',
       launchCommand: "gemini --prompt-interactive 'Investigate this regression'",
       expectedProcess: 'gemini',
-      followupPrompt: null,
+      carry: 'on-line',
+      pasteAfterReady: null,
       launchConfig: emptyLaunchConfig('gemini')
     })
   })
 
   it('uses Antigravity interactive prompt mode with the agy binary', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'antigravity',
         prompt: 'Investigate this regression',
         cmdOverrides: {},
@@ -59,14 +58,15 @@ describe('buildAgentStartupPlan', () => {
       agent: 'antigravity',
       launchCommand: "agy --prompt-interactive 'Investigate this regression'",
       expectedProcess: 'agy',
-      followupPrompt: null,
+      carry: 'on-line',
+      pasteAfterReady: null,
       launchConfig: emptyLaunchConfig('agy')
     })
   })
 
   it('launches aider first and injects the draft prompt after startup', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'aider',
         prompt: 'Refactor the parser',
         cmdOverrides: {},
@@ -76,14 +76,15 @@ describe('buildAgentStartupPlan', () => {
       agent: 'aider',
       launchCommand: 'aider',
       expectedProcess: 'aider',
-      followupPrompt: 'Refactor the parser',
+      carry: 'paste-after-ready',
+      pasteAfterReady: 'Refactor the parser',
       launchConfig: emptyLaunchConfig('aider')
     })
   })
 
   it('launches Autohand Code first and injects the draft prompt after startup', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'autohand',
         prompt: 'Add tests for the parser',
         cmdOverrides: {},
@@ -93,14 +94,15 @@ describe('buildAgentStartupPlan', () => {
       agent: 'autohand',
       launchCommand: 'autohand',
       expectedProcess: 'autohand',
-      followupPrompt: 'Add tests for the parser',
+      carry: 'paste-after-ready',
+      pasteAfterReady: 'Add tests for the parser',
       launchConfig: emptyLaunchConfig('autohand')
     })
   })
 
   it('launches Ante first and injects the draft prompt after startup', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'ante',
         prompt: 'Summarize the failing tests',
         cmdOverrides: {},
@@ -110,14 +112,15 @@ describe('buildAgentStartupPlan', () => {
       agent: 'ante',
       launchCommand: 'ante',
       expectedProcess: 'ante',
-      followupPrompt: 'Summarize the failing tests',
+      carry: 'paste-after-ready',
+      pasteAfterReady: 'Summarize the failing tests',
       launchConfig: emptyLaunchConfig('ante')
     })
   })
 
   it('passes the prompt to Trae as a positional argv behind a `--` separator', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'trae',
         prompt: 'Summarize the failing tests',
         cmdOverrides: {},
@@ -127,7 +130,8 @@ describe('buildAgentStartupPlan', () => {
       agent: 'trae',
       launchCommand: "traecli -- 'Summarize the failing tests'",
       expectedProcess: 'traecli',
-      followupPrompt: null,
+      carry: 'on-line',
+      pasteAfterReady: null,
       launchConfig: emptyLaunchConfig('traecli')
     })
   })
@@ -135,7 +139,7 @@ describe('buildAgentStartupPlan', () => {
   // Why: without the separator these dispatch to Trae's `help`/`config` subcommands instead.
   it('keeps subcommand-shaped Trae prompts as the positional prompt', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'trae',
         prompt: 'help me name this config',
         cmdOverrides: {},
@@ -146,7 +150,7 @@ describe('buildAgentStartupPlan', () => {
 
   it('delivers the Muse prompt after its composer is ready', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'muse',
         prompt: 'Summarize the failing tests',
         cmdOverrides: {},
@@ -156,7 +160,8 @@ describe('buildAgentStartupPlan', () => {
       agent: 'muse',
       launchCommand: 'muse --trust-workspace',
       expectedProcess: 'muse',
-      followupPrompt: 'Summarize the failing tests',
+      carry: 'paste-after-ready',
+      pasteAfterReady: 'Summarize the failing tests',
       launchConfig: {
         ...emptyLaunchConfig('muse'),
         agentCommand: 'muse --trust-workspace'
@@ -166,7 +171,7 @@ describe('buildAgentStartupPlan', () => {
 
   it('passes the prompt to Prime Agent as a positional argv behind a `--` separator', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'prime-agent',
         prompt: 'Summarize the failing tests',
         cmdOverrides: {},
@@ -176,7 +181,8 @@ describe('buildAgentStartupPlan', () => {
       agent: 'prime-agent',
       launchCommand: "prime-agent -- 'Summarize the failing tests'",
       expectedProcess: 'prime-agent',
-      followupPrompt: null,
+      carry: 'on-line',
+      pasteAfterReady: null,
       launchConfig: emptyLaunchConfig('prime-agent')
     })
   })
@@ -184,7 +190,7 @@ describe('buildAgentStartupPlan', () => {
   // Why: without the separator these dispatch to Prime Agent's `help`/`agents` subcommands instead.
   it('keeps subcommand-shaped Prime Agent prompts as the positional prompt', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'prime-agent',
         prompt: 'help me name this config',
         cmdOverrides: {},
@@ -195,7 +201,7 @@ describe('buildAgentStartupPlan', () => {
 
   it('uses cursor-agent as the actual launch binary', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'cursor',
         prompt: 'Review this file',
         cmdOverrides: {},
@@ -205,14 +211,15 @@ describe('buildAgentStartupPlan', () => {
       agent: 'cursor',
       launchCommand: "cursor-agent 'Review this file'",
       expectedProcess: 'cursor-agent',
-      followupPrompt: null,
+      carry: 'on-line',
+      pasteAfterReady: null,
       launchConfig: emptyLaunchConfig('cursor-agent')
     })
   })
 
   it('applies command overrides without changing the prompt syntax contract', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'droid',
         prompt: 'Ship the fix',
         cmdOverrides: { droid: '/opt/factory/bin/droid' },
@@ -222,14 +229,15 @@ describe('buildAgentStartupPlan', () => {
       agent: 'droid',
       launchCommand: "/opt/factory/bin/droid 'Ship the fix'",
       expectedProcess: 'droid',
-      followupPrompt: null,
+      carry: 'on-line',
+      pasteAfterReady: null,
       launchConfig: emptyLaunchConfig('/opt/factory/bin/droid')
     })
   })
 
   it('passes Copilot prompts with the -i flag for an interactive session', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'copilot',
         prompt: 'Fix the bug',
         cmdOverrides: {},
@@ -239,14 +247,15 @@ describe('buildAgentStartupPlan', () => {
       agent: 'copilot',
       launchCommand: "copilot -i 'Fix the bug'",
       expectedProcess: 'copilot',
-      followupPrompt: null,
+      carry: 'on-line',
+      pasteAfterReady: null,
       launchConfig: emptyLaunchConfig('copilot')
     })
   })
 
   it('launches Grok with the prompt as a positional argv', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'grok',
         prompt: 'Trace the failing test',
         cmdOverrides: {},
@@ -256,14 +265,15 @@ describe('buildAgentStartupPlan', () => {
       agent: 'grok',
       launchCommand: "grok -- 'Trace the failing test'",
       expectedProcess: 'grok',
-      followupPrompt: null,
+      carry: 'on-line',
+      pasteAfterReady: null,
       launchConfig: emptyLaunchConfig('grok')
     })
   })
 
   it('launches Devin first and injects the prompt after startup', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'devin',
         prompt: 'Trace the failing test',
         cmdOverrides: {},
@@ -274,7 +284,8 @@ describe('buildAgentStartupPlan', () => {
       agent: 'devin',
       launchCommand: "devin '--permission-mode' 'bypass' '--respect-workspace-trust' 'false'",
       expectedProcess: 'devin',
-      followupPrompt: 'Trace the failing test',
+      carry: 'paste-after-ready',
+      pasteAfterReady: 'Trace the failing test',
       launchConfig: {
         agentCommand: "devin '--permission-mode' 'bypass' '--respect-workspace-trust' 'false'",
         agentArgs: '--permission-mode bypass --respect-workspace-trust false',
@@ -285,7 +296,7 @@ describe('buildAgentStartupPlan', () => {
 
   it('launches Command Code by its unambiguous binary with a positional prompt', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'command-code',
         prompt: 'Fix the issue',
         cmdOverrides: {},
@@ -295,14 +306,15 @@ describe('buildAgentStartupPlan', () => {
       agent: 'command-code',
       launchCommand: "command-code --trust 'Fix the issue'",
       expectedProcess: 'command-code',
-      followupPrompt: null,
+      carry: 'on-line',
+      pasteAfterReady: null,
       launchConfig: emptyLaunchConfig('command-code --trust')
     })
   })
 
   it('returns null when there is no prompt to inject', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'codex',
         prompt: '   ',
         cmdOverrides: {},
diff --git a/src/renderer/src/lib/tui-agent-startup.ts b/src/renderer/src/lib/tui-agent-startup.ts
index a61e9003d60..7b51cdc27f2 100644
--- a/src/renderer/src/lib/tui-agent-startup.ts
+++ b/src/renderer/src/lib/tui-agent-startup.ts
@@ -3,6 +3,7 @@ export {
   buildAgentDraftLaunchPlan,
   buildAgentStartupPlan,
   planAgentCliArgsSuffix,
+  planLaunchPrompt,
   isShellProcess,
   quoteStartupArg,
   resolveStartupShell
@@ -10,5 +11,7 @@ export {
 export type {
   AgentCliArgsPlan,
   AgentDraftLaunchPlan,
-  AgentStartupPlan
+  AgentLaunchPromptArgs,
+  AgentStartupPlan,
+  AgentStartupPlanInputs
 } from '../../../shared/tui-agent-startup'
diff --git a/src/renderer/src/lib/worktree-creation-background-mount.test.ts b/src/renderer/src/lib/worktree-creation-background-mount.test.ts
index b44e87687be..955ed1779c8 100644
--- a/src/renderer/src/lib/worktree-creation-background-mount.test.ts
+++ b/src/renderer/src/lib/worktree-creation-background-mount.test.ts
@@ -40,7 +40,6 @@ function makeAgentRequest(): WorktreeCreationRequest {
       agent: 'codex',
       launchCommand: 'codex',
       expectedProcess: 'codex',
-      followupPrompt: null,
       launchConfig: { agentArgs: '', agentEnv: {} }
     },
     quickPrompt: '',
diff --git a/src/renderer/src/lib/worktree-creation-flow-ready-toast.test.ts b/src/renderer/src/lib/worktree-creation-flow-ready-toast.test.ts
index 68af47babfd..d0de17e0681 100644
--- a/src/renderer/src/lib/worktree-creation-flow-ready-toast.test.ts
+++ b/src/renderer/src/lib/worktree-creation-flow-ready-toast.test.ts
@@ -206,7 +206,6 @@ describe('a creation that finishes after the user moved on (#9944)', () => {
           agent: 'claude',
           launchCommand: 'claude',
           expectedProcess: 'claude',
-          followupPrompt: null,
           launchConfig: { agentArgs: '', agentEnv: {} }
         }
       })
diff --git a/src/renderer/src/lib/worktree-creation-flow-startup.ts b/src/renderer/src/lib/worktree-creation-flow-startup.ts
index 5201719192a..98698278abc 100644
--- a/src/renderer/src/lib/worktree-creation-flow-startup.ts
+++ b/src/renderer/src/lib/worktree-creation-flow-startup.ts
@@ -28,6 +28,7 @@ export function buildWorktreeCreationStartupOpt(
     ...(request.launchDraftPrompt ? { launchDraftText: request.launchDraftPrompt } : {}),
     ...(plan.startupCommandDelivery ? { startupCommandDelivery: plan.startupCommandDelivery } : {}),
     ...(plan.launchFile ? { launchFile: plan.launchFile } : {}),
+    ...(plan.launchPrompt ? { launchPrompt: plan.launchPrompt } : {}),
     // Why: command-code shows its prompt in the tab status before the first
     // hook fires, so the prompt is threaded through here.
     ...(request.agent === 'command-code' && request.quickPrompt.trim().length > 0
diff --git a/src/renderer/src/lib/worktree-creation-flow-stranded-surface.test.ts b/src/renderer/src/lib/worktree-creation-flow-stranded-surface.test.ts
index 7a678434a4e..97db288744f 100644
--- a/src/renderer/src/lib/worktree-creation-flow-stranded-surface.test.ts
+++ b/src/renderer/src/lib/worktree-creation-flow-stranded-surface.test.ts
@@ -232,7 +232,7 @@ describe('a throw after createWorktree succeeds no longer strands the creation s
         launchCommand: 'codex',
         expectedProcess: 'codex',
         draftPrompt: 'draft context',
-        followupPrompt: 'follow-up context',
+        pastePromptAfterReady: 'follow-up context',
         launchConfig: { agentArgs: '', agentEnv: {} }
       }
     })
diff --git a/src/renderer/src/lib/worktree-creation-flow.test.ts b/src/renderer/src/lib/worktree-creation-flow.test.ts
index 92dea7e3991..cd1cd6c2666 100644
--- a/src/renderer/src/lib/worktree-creation-flow.test.ts
+++ b/src/renderer/src/lib/worktree-creation-flow.test.ts
@@ -626,7 +626,6 @@ describe('staged background worktree creation', () => {
           agent: 'claude',
           launchCommand: 'claude --prefill x',
           expectedProcess: 'claude',
-          followupPrompt: null,
           launchConfig: { agentArgs: '', agentEnv: {} }
         },
         launchDraftPrompt: 'https://github.com/o/r/issues/12'
@@ -664,7 +663,6 @@ describe('staged background worktree creation', () => {
           agent: 'claude',
           launchCommand: 'claude --prefill x',
           expectedProcess: 'claude',
-          followupPrompt: null,
           launchConfig: { agentArgs: '', agentEnv: {} }
         },
         launchDraftPrompt: 'https://github.com/o/r/issues/12'
@@ -697,7 +695,6 @@ describe('staged background worktree creation', () => {
           agent,
           launchCommand: `${agent} --prefill x`,
           expectedProcess: agent,
-          followupPrompt: null,
           launchConfig: { agentArgs: '', agentEnv: {} }
         },
         launchDraftPrompt: draft
@@ -732,7 +729,6 @@ describe('staged background worktree creation', () => {
           agent: 'claude',
           launchCommand: "claude --prefill 'https://github.com/o/r/issues/12'",
           expectedProcess: 'claude',
-          followupPrompt: null,
           launchConfig: { agentArgs: '', agentEnv: {} }
         },
         launchDraftPrompt: 'https://github.com/o/r/issues/12'
@@ -761,7 +757,6 @@ describe('staged background worktree creation', () => {
           agent: 'claude',
           launchCommand: 'claude',
           expectedProcess: 'claude',
-          followupPrompt: null,
           launchConfig: { agentArgs: '', agentEnv: {} }
         }
       })
@@ -789,7 +784,6 @@ describe('staged background worktree creation', () => {
           agent: 'codex',
           launchCommand: 'codex',
           expectedProcess: 'codex',
-          followupPrompt: null,
           launchConfig: { agent: 'codex', command: 'codex' },
           draftPrompt: 'ship it'
         } as never
diff --git a/src/renderer/src/lib/worktree-startup-payload.ts b/src/renderer/src/lib/worktree-startup-payload.ts
index ab3b3b44d5b..46244464315 100644
--- a/src/renderer/src/lib/worktree-startup-payload.ts
+++ b/src/renderer/src/lib/worktree-startup-payload.ts
@@ -33,6 +33,8 @@ export type WorktreeStartupPayload = {
   startupCommandDelivery?: StartupCommandDelivery
   /** Written by the host before it types `command`, which names it by placeholder. */
   launchFile?: LaunchFile
+  /** The submitted prompt `command` carries; handed back to copy if the host refuses the spawn. */
+  launchPrompt?: string
   initialAgentStatus?: { agent: TuiAgent; prompt: string }
   sessionOptions?: Record<string, SessionOptionValue>
   telemetry?: AgentStartedTelemetry
diff --git a/src/renderer/src/store/terminals/terminal-actions.ts b/src/renderer/src/store/terminals/terminal-actions.ts
index c0448be5ee3..0b1306ec71f 100644
--- a/src/renderer/src/store/terminals/terminal-actions.ts
+++ b/src/renderer/src/store/terminals/terminal-actions.ts
@@ -207,6 +207,7 @@ export type TerminalActions = {
       delivery?: 'terminal-paste'
       startupCommandDelivery?: StartupCommandDelivery
       launchFile?: LaunchFile
+      launchPrompt?: string
       env?: Record<string, string>
       envToDelete?: string[]
       launchConfig?: SleepingAgentLaunchConfig
@@ -234,6 +235,7 @@ export type TerminalActions = {
     delivery?: 'terminal-paste'
     startupCommandDelivery?: StartupCommandDelivery
     launchFile?: LaunchFile
+    launchPrompt?: string
     env?: Record<string, string>
     envToDelete?: string[]
     launchConfig?: SleepingAgentLaunchConfig
diff --git a/src/renderer/src/store/terminals/terminal-state.ts b/src/renderer/src/store/terminals/terminal-state.ts
index a91dfbdbd26..f1b9f0f38bb 100644
--- a/src/renderer/src/store/terminals/terminal-state.ts
+++ b/src/renderer/src/store/terminals/terminal-state.ts
@@ -67,6 +67,7 @@ export type TerminalState = {
       delivery?: 'terminal-paste'
       startupCommandDelivery?: StartupCommandDelivery
       launchFile?: LaunchFile
+      launchPrompt?: string
       env?: Record<string, string>
       envToDelete?: string[]
       launchConfig?: SleepingAgentLaunchConfig
diff --git a/src/shared/cmd-startup-arg-quoting.win32.test.ts b/src/shared/cmd-startup-arg-quoting.win32.test.ts
index 1e7dd8424b4..132dbed105f 100644
--- a/src/shared/cmd-startup-arg-quoting.win32.test.ts
+++ b/src/shared/cmd-startup-arg-quoting.win32.test.ts
@@ -2,6 +2,7 @@ import { existsSync, mkdirSync, mkdtempSync, readFileSync, writeFileSync } from
 import { tmpdir } from 'node:os'
 import { join } from 'node:path'
 import { afterAll, beforeAll, describe, expect, it } from 'vitest'
+import { planLaunchForTest } from './launch-prompt-plan.test-fixture'
 import { runProcess } from './child-process/run-process'
 import {
   WINDOWS_ARGUMENT_CORPUS,
@@ -9,7 +10,6 @@ import {
 } from './child-process/__fixtures__/windows-argument-corpus'
 import { removeTreeSync } from './windows-transient-lock-removal'
 import { quoteStartupArg } from './tui-agent-startup-shell'
-import { buildAgentStartupPlan } from './tui-agent-startup'
 import { removeLaunchFile, writeLaunchFile } from './launch-file-writing'
 
 /**
@@ -64,7 +64,7 @@ describeOnWindows('cmd launch-line prompt quoting', () => {
   it('never lets a multi-line prompt reach cmd as commands: the line names a launch file', async () => {
     const marker = join(dir, 'pwned.txt')
     const prompt = `Fix the build\r\n& echo PWNED> "${marker}"\nthen run the tests`
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude',
       prompt,
       cmdOverrides: { claude: shim },
@@ -97,7 +97,7 @@ describeOnWindows('cmd launch-line prompt quoting', () => {
   it('names a launch file under a home with an apostrophe, accents, CJK and `%`', async () => {
     const home = join(dir, "O'Brien José 张伟 100%")
     mkdirSync(home)
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude',
       prompt: 'Fix the build\nthen run the tests',
       cmdOverrides: { claude: shim },
diff --git a/src/shared/freebuff-launch.test.ts b/src/shared/freebuff-launch.test.ts
index b58dc4bd091..2714f36fdbe 100644
--- a/src/shared/freebuff-launch.test.ts
+++ b/src/shared/freebuff-launch.test.ts
@@ -1,5 +1,6 @@
 import { describe, expect, it } from 'vitest'
-import { buildAgentDraftLaunchPlan, buildAgentStartupPlan } from './tui-agent-startup'
+import { planLaunchForTest } from './launch-prompt-plan.test-fixture'
+import { buildAgentDraftLaunchPlan } from './tui-agent-startup'
 import { isExpectedAgentProcess, recognizeAgentProcess } from './agent-process-recognition'
 import { pickTuiAgent } from './tui-agent-selection'
 
@@ -8,7 +9,7 @@ describe('Freebuff terminal launches', () => {
     'delivers task text through the interactive terminal on %s',
     (platform) => {
       const prompt = '--help\nFix "quoted" paths and $HOME handling'
-      const plan = buildAgentStartupPlan({
+      const plan = planLaunchForTest({
         agent: 'freebuff',
         prompt,
         cmdOverrides: {},
@@ -17,7 +18,7 @@ describe('Freebuff terminal launches', () => {
       expect(plan).toMatchObject({
         launchCommand: 'freebuff',
         expectedProcess: 'freebuff',
-        followupPrompt: prompt
+        pasteAfterReady: prompt
       })
       expect(
         buildAgentDraftLaunchPlan({
@@ -32,14 +33,14 @@ describe('Freebuff terminal launches', () => {
 
   it('uses the same launch contract on SSH hosts', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'freebuff',
         prompt: 'Inspect this folder',
         cmdOverrides: {},
         platform: 'linux',
         isRemote: true
       })
-    ).toMatchObject({ launchCommand: 'freebuff', followupPrompt: 'Inspect this folder' })
+    ).toMatchObject({ launchCommand: 'freebuff', pasteAfterReady: 'Inspect this folder' })
   })
 
   it.each(['/usr/local/bin/freebuff', 'C:\\tools\\freebuff.exe'])(
diff --git a/src/shared/freebuff-startup.test.ts b/src/shared/freebuff-startup.test.ts
index 65806790af4..34616b61eb2 100644
--- a/src/shared/freebuff-startup.test.ts
+++ b/src/shared/freebuff-startup.test.ts
@@ -23,8 +23,7 @@ describe('Freebuff startup', () => {
 
     expect(freebuff).toMatchObject({
       launchCommand: 'freebuff --debug',
-      expectedProcess: 'freebuff',
-      followupPrompt: null
+      expectedProcess: 'freebuff'
     })
     expect(codebuff).toMatchObject({ launchCommand: 'codebuff', expectedProcess: 'codebuff' })
   })
diff --git a/src/shared/launch-file-directory-grant.test.ts b/src/shared/launch-file-directory-grant.test.ts
index bf50b3e03dc..5b5b1114342 100644
--- a/src/shared/launch-file-directory-grant.test.ts
+++ b/src/shared/launch-file-directory-grant.test.ts
@@ -2,18 +2,18 @@ import { mkdirSync, mkdtempSync, realpathSync, rmSync, symlinkSync } from 'node:
 import { tmpdir } from 'node:os'
 import { join } from 'node:path'
 import { afterEach, beforeEach, describe, expect, it } from 'vitest'
+import { planLaunchForTest } from './launch-prompt-plan.test-fixture'
 import { writeLaunchFile } from './launch-file-writing'
 import {
   buildLaunchFilePointer,
   carryInLaunchFile,
   launchFileDirectoryPlaceholder
 } from './launch-prompt-file'
-import { buildAgentStartupPlan } from './tui-agent-startup'
 import type { TuiAgent } from './tui-agent'
 
 function planWithLaunchFile(agent: TuiAgent, shell: 'posix' | 'cmd' | 'powershell' = 'posix') {
   const { prompt, launchFile } = carryInLaunchFile('the brief', true)
-  const plan = buildAgentStartupPlan({
+  const plan = planLaunchForTest({
     agent,
     prompt,
     cmdOverrides: {},
@@ -53,7 +53,7 @@ describe('the launch-file directory grant on the agent command line', () => {
     const { plan, pointer } = planWithLaunchFile('codex')
     expect(plan?.launchCommand).toBe(`codex '${pointer}'`)
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'claude',
         prompt: 'fix it',
         cmdOverrides: {},
diff --git a/src/shared/launch-file-writing.test.ts b/src/shared/launch-file-writing.test.ts
index 5f8620e36cd..cbdf971d4c2 100644
--- a/src/shared/launch-file-writing.test.ts
+++ b/src/shared/launch-file-writing.test.ts
@@ -20,12 +20,7 @@ import {
   sweepStaleLaunchFiles,
   writeLaunchFile
 } from './launch-file-writing'
-import {
-  MAX_INLINE_LAUNCH_PROMPT_CHARS,
-  buildLaunchFilePointer,
-  parseLaunchFile,
-  planLaunchPrompt
-} from './launch-prompt-file'
+import { buildLaunchFilePointer, carryInLaunchFile, parseLaunchFile } from './launch-prompt-file'
 import { quoteStartupArg } from './tui-agent-startup-shell'
 
 let baseDirectory: string
@@ -38,35 +33,30 @@ afterEach(() => {
   rmSync(baseDirectory, { recursive: true, force: true })
 })
 
-describe('planLaunchPrompt', () => {
-  it('keeps a prompt at the ceiling inline', () => {
-    const prompt = 'a'.repeat(MAX_INLINE_LAUNCH_PROMPT_CHARS)
-    expect(planLaunchPrompt(prompt)).toEqual({ prompt })
-  })
-
-  it('moves a 20,000-character prompt into a launch file', () => {
+describe('carryInLaunchFile', () => {
+  it('moves a prompt into a launch file and names it by a pointer', () => {
     const prompt = 'b'.repeat(20_000)
-    const planned = planLaunchPrompt(prompt)
-    expect(planned.launchFile).toMatchObject({ content: prompt, sensitive: false })
-    expect(planned.prompt).toBe(buildLaunchFilePointer(planned.launchFile!.placeholder))
+    const carried = carryInLaunchFile(prompt, false)
+    expect(carried.launchFile).toMatchObject({ content: prompt, sensitive: false })
+    expect(carried.prompt).toBe(buildLaunchFilePointer(carried.launchFile.placeholder))
   })
 
-  it('moves a short sensitive prompt into a launch file', () => {
-    const planned = planLaunchPrompt('token dcap_secret', { sensitive: true })
-    expect(planned.prompt).not.toContain('dcap_secret')
-    expect(planned.launchFile).toMatchObject({ content: 'token dcap_secret', sensitive: true })
+  it('keeps a sensitive prompt off the pointer', () => {
+    const carried = carryInLaunchFile('token dcap_secret', true)
+    expect(carried.prompt).not.toContain('dcap_secret')
+    expect(carried.launchFile).toMatchObject({ content: 'token dcap_secret', sensitive: true })
   })
 
   it('mints a fresh placeholder each time', () => {
-    const first = planLaunchPrompt('x', { sensitive: true }).launchFile!.placeholder
-    const second = planLaunchPrompt('x', { sensitive: true }).launchFile!.placeholder
+    const first = carryInLaunchFile('x', true).launchFile.placeholder
+    const second = carryInLaunchFile('x', true).launchFile.placeholder
     expect(first).not.toBe(second)
   })
 })
 
 describe('parseLaunchFile', () => {
-  it('accepts what planLaunchPrompt mints and rejects anything else', () => {
-    const { launchFile } = planLaunchPrompt('x', { sensitive: true })
+  it('accepts what carryInLaunchFile mints and rejects anything else', () => {
+    const { launchFile } = carryInLaunchFile('x', true)
     expect(parseLaunchFile(launchFile)).toEqual(launchFile)
     expect(parseLaunchFile({ ...launchFile, placeholder: 'HOME' })).toBeUndefined()
     expect(parseLaunchFile({ ...launchFile, sensitive: 'yes' })).toBeUndefined()
@@ -81,7 +71,7 @@ describe('parseLaunchFile', () => {
 
 describe('writeLaunchFile', () => {
   it('writes a 0600 file in a private 0700 directory and puts its path in the line and env', () => {
-    const { prompt, launchFile } = planLaunchPrompt('c'.repeat(20_000))
+    const { prompt, launchFile } = carryInLaunchFile('c'.repeat(20_000), false)
     const written = writeLaunchFile({
       launchFile: launchFile!,
       command: `claude '${prompt}'`,
@@ -102,7 +92,7 @@ describe('writeLaunchFile', () => {
   it("refuses an apostrophe in the path when the line's quoting is unknown, leaving nothing behind", () => {
     const quoted = join(baseDirectory, "it's")
     mkdirSync(quoted)
-    const { launchFile } = planLaunchPrompt('x', { sensitive: true })
+    const { launchFile } = carryInLaunchFile('x', true)
     expect(() => writeLaunchFile({ launchFile: launchFile!, baseDirectory: quoted })).toThrow(
       LaunchFileUnavailableError
     )
@@ -110,7 +100,7 @@ describe('writeLaunchFile', () => {
   })
 
   it('refuses when the temp directory cannot be written', () => {
-    const { launchFile } = planLaunchPrompt('x', { sensitive: true })
+    const { launchFile } = carryInLaunchFile('x', true)
     expect(() =>
       writeLaunchFile({ launchFile: launchFile!, baseDirectory: join(baseDirectory, 'missing') })
     ).toThrow(
@@ -119,7 +109,7 @@ describe('writeLaunchFile', () => {
   })
 
   it('removes the whole directory', () => {
-    const { launchFile } = planLaunchPrompt('x', { sensitive: true })
+    const { launchFile } = carryInLaunchFile('x', true)
     const written = writeLaunchFile({ launchFile: launchFile!, baseDirectory })
     removeLaunchFile(written)
     expect(existsSync(written.directory)).toBe(false)
@@ -178,7 +168,7 @@ describePosix('a launch line naming a file under an unusual home directory', ()
     async (user) => {
       const home = join(baseDirectory, user)
       mkdirSync(home)
-      const { prompt, launchFile } = planLaunchPrompt('brief', { sensitive: true })
+      const { prompt, launchFile } = carryInLaunchFile('brief', true)
       const written = writeLaunchFile({
         launchFile: { ...launchFile!, quoting: 'posix' },
         command: `printf '%s' ${quoteStartupArg(prompt, 'posix')}`,
diff --git a/src/shared/launch-file-writing.ts b/src/shared/launch-file-writing.ts
index b857163344a..b3d55e20b98 100644
--- a/src/shared/launch-file-writing.ts
+++ b/src/shared/launch-file-writing.ts
@@ -20,7 +20,7 @@ import {
   type LaunchFile
 } from './launch-prompt-file'
 import { quoteStartupArg, type AgentStartupShell } from './tui-agent-startup-shell'
-import { typedStartupLineFits } from './typed-startup-line'
+import { startupLineNeedsStaging } from './startup-command-staging'
 import type { WslLaunchDirectory } from './wsl-launch-directory'
 
 const LAUNCH_FILE_DIR_PREFIX = 'orca-launch-file-'
@@ -69,8 +69,8 @@ export function launchFilePathInQuotedRun(
 }
 
 export class LaunchFileUnavailableError extends Error {
-  constructor(reason: string) {
-    super(describeLaunchFileUnavailable(reason))
+  constructor(reason: string, carrier: 'file' | 'staged-line' = 'file') {
+    super(describeLaunchFileUnavailable(reason, carrier))
   }
 }
 
@@ -161,6 +161,19 @@ export function writeLaunchFile(args: {
   }
 }
 
+/** Whether a WSL spawn needs its distro's launch directory: for a launch file, or to stage the
+ *  long or multi-line agent line Orca built. Only then is the distro probed. */
+export function spawnNeedsWslLaunchDirectory(args: {
+  launchFile?: LaunchFile
+  command?: string
+  orcaBuiltLine: boolean
+}): boolean {
+  return (
+    args.launchFile !== undefined ||
+    (args.orcaBuiltLine && args.command !== undefined && startupLineNeedsStaging(args.command))
+  )
+}
+
 /**
  * The write site's launch file for a spawn. A WSL session writes it into the distro, and refuses
  * when the distro directory is unknown and a file or a staged line needs it: a Windows path means
@@ -175,11 +188,15 @@ export function writeSpawnLaunchFile(args: {
   wslDirectory: WslLaunchDirectory | undefined
 }): WrittenLaunchFile | undefined {
   const wslDirectory = args.wslDistro ? args.wslDirectory : undefined
-  const needsDirectory =
-    args.launchFile !== undefined ||
-    (args.orcaBuiltLine && args.command !== undefined && !typedStartupLineFits(args.command))
-  if (args.wslDistro && needsDirectory && wslDirectory?.distro !== args.wslDistro) {
-    throw new LaunchFileUnavailableError("the WSL distro's home directory could not be reached")
+  if (
+    args.wslDistro &&
+    spawnNeedsWslLaunchDirectory(args) &&
+    wslDirectory?.distro !== args.wslDistro
+  ) {
+    throw new LaunchFileUnavailableError(
+      "the WSL distro's home directory could not be reached",
+      args.launchFile ? 'file' : 'staged-line'
+    )
   }
   return args.launchFile
     ? writeLaunchFile({
diff --git a/src/shared/launch-prompt-carrier.test.ts b/src/shared/launch-prompt-carrier.test.ts
index 3f73193b164..81b69019970 100644
--- a/src/shared/launch-prompt-carrier.test.ts
+++ b/src/shared/launch-prompt-carrier.test.ts
@@ -1,43 +1,51 @@
 import { describe, expect, it } from 'vitest'
-import { buildAgentStartupPlan } from './tui-agent-startup'
+import { planLaunchForTest } from './launch-prompt-plan.test-fixture'
 
 const base = { agent: 'claude' as const, cmdOverrides: {} }
 
 describe('the one launch-prompt decision every launch path builds through', () => {
   // Why: the composer, phone quick commands and background sessions call the builder directly.
-  it('points a single-line prompt past cmd’s line cap at a launch file', () => {
+  it('points a single-line Windows prompt past the typed budget at a launch file', () => {
     const prompt = 'y'.repeat(9_000)
-    const plan = buildAgentStartupPlan({ ...base, prompt, platform: 'win32', shell: 'cmd' })
+    const plan = planLaunchForTest({ ...base, prompt, platform: 'win32', shell: 'cmd' })
     expect(plan?.launchFile?.content).toBe(prompt)
     expect(plan?.launchCommand.length).toBeLessThan(8_191)
   })
 
   it('points a prompt past the argv ceiling at a launch file on a POSIX host too', () => {
     const prompt = 'z'.repeat(20_000)
-    expect(buildAgentStartupPlan({ ...base, prompt, platform: 'linux' })?.launchFile?.content).toBe(
+    expect(planLaunchForTest({ ...base, prompt, platform: 'linux' })?.launchFile?.content).toBe(
       prompt
     )
   })
 
-  it('keeps a sensitive prompt off the line however short it is', () => {
-    const plan = buildAgentStartupPlan({
+  it('keeps a paired host to the typed budget, pasting what it may neither stage nor read', () => {
+    const short = planLaunchForTest({
       ...base,
-      prompt: 'token dcap_secret',
+      prompt: 'fix it',
       platform: 'linux',
-      sensitive: true
+      launchHostIsPaired: true
     })
-    expect(plan?.launchFile).toMatchObject({ content: 'token dcap_secret', sensitive: true })
-    expect(plan?.launchCommand).not.toContain('dcap_secret')
+    expect(short?.carry).toBe('on-line')
+    const long = planLaunchForTest({
+      ...base,
+      prompt: 'fix it\nthen run the tests',
+      platform: 'linux',
+      launchHostIsPaired: true
+    })
+    expect(long?.carry).toBe('paste-after-ready')
+    expect(long?.pasteAfterReady).toBe('fix it\nthen run the tests')
+    expect(long?.launchCommand).not.toContain('run the tests')
   })
 
   // Why: the host writes a WSL session's staged line and launch file into the distro.
   it('plans a WSL launch like any Linux launch: long lines stay typed, huge ones get a file', () => {
-    const typed = buildAgentStartupPlan({ ...base, platform: 'linux', prompt: 'one\ntwo' })
+    const typed = planLaunchForTest({ ...base, platform: 'linux', prompt: 'one\ntwo' })
     expect(typed?.launchCommand).toContain('two')
-    expect(typed?.followupPrompt).toBeNull()
+    expect(typed?.pasteAfterReady).toBeNull()
     const huge = 'c'.repeat(20_000)
-    const pointed = buildAgentStartupPlan({ ...base, platform: 'linux', prompt: huge })
+    const pointed = planLaunchForTest({ ...base, platform: 'linux', prompt: huge })
     expect(pointed?.launchFile?.content).toBe(huge)
-    expect(pointed?.followupPrompt).toBeNull()
+    expect(pointed?.pasteAfterReady).toBeNull()
   })
 })
diff --git a/src/shared/launch-prompt-carry.test.ts b/src/shared/launch-prompt-carry.test.ts
new file mode 100644
index 00000000000..f058bfc7bb9
--- /dev/null
+++ b/src/shared/launch-prompt-carry.test.ts
@@ -0,0 +1,130 @@
+import { describe, expect, it } from 'vitest'
+import { MAX_INLINE_LAUNCH_PROMPT_CHARS, carryInLaunchFile } from './launch-prompt-file'
+import { planLaunchPrompt, type AgentLaunchPromptArgs } from './tui-agent-startup'
+import type { TuiAgent } from './tui-agent'
+import { RUNTIME_CAPABILITIES } from './protocol-version'
+import { AGENT_LAUNCH_PROMPT_CARRY_RUNTIME_CAPABILITY } from './agent-launch-runtime-capability'
+import { TYPED_STARTUP_LINE_BUDGET_BYTES, typedStartupLineFits } from './typed-startup-line'
+
+function plan(
+  agent: TuiAgent,
+  prompt: string,
+  extra: Partial<Omit<AgentLaunchPromptArgs, 'agent' | 'prompt'>> = {}
+) {
+  return planLaunchPrompt({ agent, prompt, cmdOverrides: {}, platform: 'darwin', ...extra })
+}
+
+describe('where a launch prompt rides', () => {
+  it('builds a plain launch for an empty prompt', () => {
+    expect(plan('claude', '  ')).toMatchObject({ carry: 'none' })
+  })
+
+  it('carries a short prompt on the line', () => {
+    const planned = plan('claude', 'explain this repo')
+    expect(planned?.carry).toBe('on-line')
+    expect(planned?.carry === 'on-line' && planned.plan.launchCommand).toContain(
+      'explain this repo'
+    )
+  })
+
+  it.each([
+    ['a 600-byte', 'x'.repeat(600)],
+    ['a multi-line', 'first line\nsecond line'],
+    ['a key-bearing', 'see\tthis \x1b[31mred']
+  ])('carries %s prompt on the line of a POSIX host, which stages it', (_label, prompt) => {
+    const planned = plan('codex', prompt)
+    expect(planned?.carry).toBe('on-line')
+    expect(planned?.carry === 'on-line' && planned.plan.launchCommand).toContain(prompt)
+  })
+
+  it.each<TuiAgent>(['claude', 'codex'])(
+    'points %s at a launch file past the argv ceiling, with the full text in the file',
+    (agent) => {
+      const prompt = `${'y'.repeat(MAX_INLINE_LAUNCH_PROMPT_CHARS)}z`
+      const planned = plan(agent, prompt)
+      if (planned?.carry !== 'launch-file') {
+        throw new Error(`expected a launch file, got ${planned?.carry}`)
+      }
+      expect(planned.launchFile).toMatchObject({ content: prompt, sensitive: false })
+      expect(planned.plan.launchCommand).toContain(planned.launchFile.placeholder)
+      expect(planned.plan.launchCommand).not.toContain('yyyy')
+    }
+  )
+
+  // The bug class this outcome removes: a plan that exists but does not carry the prompt.
+  it('leaves a file-sized prompt for the paste for an agent not measured reading the file', () => {
+    const prompt = 'g'.repeat(MAX_INLINE_LAUNCH_PROMPT_CHARS + 1)
+    const planned = plan('gemini', prompt)
+    if (planned?.carry !== 'paste-after-ready') {
+      throw new Error(`expected the paste, got ${planned?.carry}`)
+    }
+    expect(planned.text).toBe(prompt)
+    expect(planned.cleanPlan.launchCommand).not.toContain('gggg')
+  })
+
+  it('leaves a stdin-after-start agent’s prompt for the paste', () => {
+    const planned = plan('aider', 'fix it')
+    expect(planned).toMatchObject({ carry: 'paste-after-ready', text: 'fix it' })
+  })
+
+  it('names a launch file its caller wrote, quoted for the line', () => {
+    const { prompt, launchFile } = carryInLaunchFile('worker brief', true)
+    const planned = plan('claude', prompt, { launchFile })
+    expect(planned).toMatchObject({
+      carry: 'launch-file',
+      launchFile: { ...launchFile, quoting: 'posix' }
+    })
+  })
+})
+
+describe('a host that types the line raw', () => {
+  it.each<['cmd' | 'powershell']>([['cmd'], ['powershell']])(
+    'keeps a %s line to the typed budget and points Claude at a file past it',
+    (shell) => {
+      const short = plan('claude', 'fix it', { platform: 'win32', shell })
+      expect(short?.carry).toBe('on-line')
+      const long = plan('claude', 'y'.repeat(600), { platform: 'win32', shell })
+      expect(long?.carry).toBe('launch-file')
+    }
+  )
+
+  it('points Codex at a file for a Windows-damaged prompt, and pastes it for Gemini', () => {
+    const prompt = 'fix the build\nthen run the tests'
+    expect(plan('codex', prompt, { platform: 'win32', shell: 'cmd' })?.carry).toBe('launch-file')
+    expect(plan('gemini', prompt, { platform: 'win32', shell: 'cmd' })).toMatchObject({
+      carry: 'paste-after-ready',
+      text: prompt
+    })
+  })
+
+  it('points a PowerShell prompt with a double quote at a file', () => {
+    expect(plan('claude', 'say "hi"', { platform: 'win32', shell: 'powershell' })?.carry).toBe(
+      'launch-file'
+    )
+  })
+
+  it('pastes on a paired host what its line cannot carry typed, even for Claude', () => {
+    const extra = { platform: 'linux' as const, launchHostIsPaired: true }
+    expect(plan('claude', 'fix it', extra)?.carry).toBe('on-line')
+    expect(plan('claude', 'first line\nsecond line', extra)?.carry).toBe('paste-after-ready')
+    expect(plan('claude', 'z'.repeat(20_000), extra)?.carry).toBe('paste-after-ready')
+  })
+})
+
+describe('whether a line can be typed as it is', () => {
+  it('holds a line to half of macOS MAX_CANON', () => {
+    expect(typedStartupLineFits('x'.repeat(TYPED_STARTUP_LINE_BUDGET_BYTES))).toBe(true)
+    expect(typedStartupLineFits('x'.repeat(TYPED_STARTUP_LINE_BUDGET_BYTES + 1))).toBe(false)
+  })
+
+  it.each(['\n', '\r', '\t', '\x1b', '\x03', '\x7f'])('never types a line with %j', (byte) => {
+    expect(typedStartupLineFits(`a${byte}b`)).toBe(false)
+  })
+})
+
+describe('the capability clients gate a prompted launch on', () => {
+  it('is advertised by every host that delivers a prompt its typed line cannot carry as typed', () => {
+    // An older host types any prompt into the line, so its absence is the gate.
+    expect(RUNTIME_CAPABILITIES).toContain(AGENT_LAUNCH_PROMPT_CARRY_RUNTIME_CAPABILITY)
+  })
+})
diff --git a/src/shared/launch-prompt-carry.ts b/src/shared/launch-prompt-carry.ts
new file mode 100644
index 00000000000..c9e708f45f8
--- /dev/null
+++ b/src/shared/launch-prompt-carry.ts
@@ -0,0 +1,167 @@
+/** Where a launch prompt rides: decided once here, for every launch path, and switched on by each. */
+import {
+  MAX_INLINE_LAUNCH_PROMPT_CHARS,
+  carryInLaunchFile,
+  launchFileDirectoryPlaceholder,
+  type LaunchFile
+} from './launch-prompt-file'
+import { TUI_AGENT_CONFIG } from './tui-agent-config'
+import {
+  isPosixStartupShell,
+  quoteStartupArg,
+  resolveStartupShell,
+  type AgentStartupShell
+} from './tui-agent-startup-shell'
+import { hasControlByte, typedStartupLineFits } from './typed-startup-line'
+import type { TuiAgent } from './tui-agent'
+
+/**
+ * Whether a Windows shell would damage `prompt` as one quoted argument. No quoting keeps a control
+ * byte literal: cmd types the line and a line break submits it early. PowerShell runs a short line
+ * from its argv, so its damage is the hand-off to the agent (measured): legacy argument passing (5.1
+ * always, 7.x into a `.cmd` shim) splits at an inner `"` and turns a trailing `\` into `"`, and a
+ * shim's cmd.exe expands `%NAME%`; a lone `%` stays literal.
+ */
+export function windowsShellDamagesPrompt(prompt: string, shell: AgentStartupShell): boolean {
+  if (isPosixStartupShell(shell)) {
+    return false
+  }
+  return (
+    hasControlByte(prompt) ||
+    (shell === 'powershell' &&
+      (prompt.includes('"') || /%[^%]+%/.test(prompt) || prompt.endsWith('\\')))
+  )
+}
+
+/** Why a prefill draft could not be launched, in the user's words, when the Windows shell is why. */
+export function windowsDraftRefusal(draft: string, shell: AgentStartupShell): string | null {
+  return windowsShellDamagesPrompt(draft.trim(), shell)
+    ? "The host's Windows shell would break this draft on the agent's command line (it has a line " +
+        'break or other control character, or on PowerShell a double quote, a %NAME% pair or a ' +
+        'trailing backslash), so the agent was not started. Start it without the draft and paste ' +
+        'the draft once it opens.'
+    : null
+}
+
+/**
+ * Where a launch prompt went. Only the carried outcomes hold the plan to launch, so a caller must
+ * switch on `carry` before it can launch anything, and cannot mistake a paste for a delivery.
+ */
+export type LaunchPromptPlan<P> =
+  | { carry: 'none'; plan: P }
+  | { carry: 'on-line'; plan: P }
+  | { carry: 'launch-file'; plan: P; launchFile: LaunchFile }
+  | { carry: 'paste-after-ready'; cleanPlan: P; text: string }
+
+export type LaunchPromptCarry = LaunchPromptPlan<unknown>['carry']
+
+/** Why a create that cannot paste after the agent is ready refuses a prompt that needs it. */
+export function launchPromptNeedsPasteRefusal(
+  agent: TuiAgent,
+  created: 'terminal' | 'session'
+): string {
+  return TUI_AGENT_CONFIG[agent].promptInjectionMode === 'stdin-after-start'
+    ? `${agent} takes its prompt only after it starts, so no ${created} was created. Start the ` +
+        'agent and paste the prompt once it opens.'
+    : `${agent} cannot take this prompt on its command line here (too long, or broken by the ` +
+        `host's Windows shell), so no ${created} was created. Start the agent without it and ` +
+        'paste the prompt once it opens.'
+}
+
+export type CarriedPlanArgs = {
+  agent: TuiAgent
+  prompt: string
+  platform: NodeJS.Platform
+  shell?: AgentStartupShell
+  /** The file `prompt` already points at, when the caller wrote its own. */
+  launchFile?: LaunchFile
+  /** Another Orca this client drives, possibly an older one: it is sent a command it may neither
+   *  stage nor accompany with a launch file. Temporary, until paired hosts advertise both. */
+  launchHostIsPaired?: boolean
+}
+
+export function agentReadsLaunchFile(agent: TuiAgent): boolean {
+  return TUI_AGENT_CONFIG[agent].readsLaunchFile === true
+}
+
+/**
+ * The one carry rule. The prompt rides the agent's line, which a host that stages (POSIX, SSH, WSL)
+ * stages when it is long or multi-line. It rides a launch file when it is past the argv ceiling,
+ * when a Windows shell would damage it, or when the host types the line raw (a Windows host, a
+ * paired Orca) and the line is past the typed budget. A launch file goes only to an agent measured
+ * reading one, on a host that writes it; otherwise the agent starts clean and the prompt is pasted
+ * once it is ready.
+ */
+export function carryLaunchPrompt<A extends CarriedPlanArgs, P extends { launchCommand: string }>(
+  args: A,
+  buildLine: (args: A) => P | null
+): LaunchPromptPlan<P> | null {
+  const text = args.prompt.trim()
+  const shell = resolveStartupShell(args.platform, args.shell)
+  const clean = (): P | null => buildLine({ ...args, prompt: '', launchFile: undefined })
+  if (!text) {
+    const plan = clean()
+    return plan && { carry: 'none', plan }
+  }
+  if (args.launchFile) {
+    const plan = buildLine(args)
+    return plan && { carry: 'launch-file', plan, launchFile: withQuoting(args.launchFile, shell) }
+  }
+  const pasteAfterReady = (): LaunchPromptPlan<P> | null => {
+    const cleanPlan = clean()
+    return cleanPlan && { carry: 'paste-after-ready', cleanPlan, text }
+  }
+  const mode = TUI_AGENT_CONFIG[args.agent].promptInjectionMode
+  if (mode === 'stdin-after-start') {
+    return pasteAfterReady()
+  }
+  const viaLaunchFile = (): LaunchPromptPlan<P> | null => {
+    // Why: an agent not measured reading the file would stop on an approval or refuse the path.
+    if (args.launchHostIsPaired === true || !agentReadsLaunchFile(args.agent)) {
+      return pasteAfterReady()
+    }
+    const pointer = carryInLaunchFile(text, false)
+    const plan = buildLine({ ...args, prompt: pointer.prompt, launchFile: pointer.launchFile })
+    return (
+      plan && { carry: 'launch-file', plan, launchFile: withQuoting(pointer.launchFile, shell) }
+    )
+  }
+  const readsEnv = mode === 'hermes-query'
+  if (
+    text.length > MAX_INLINE_LAUNCH_PROMPT_CHARS ||
+    (!readsEnv && windowsShellDamagesPrompt(text, shell))
+  ) {
+    return viaLaunchFile()
+  }
+  const plan = buildLine(args)
+  if (!plan) {
+    // Hermes reads its prompt from the env and refuses one past that budget, counted in bytes; a
+    // one-character query building proves the budget, not the command, refused it.
+    return readsEnv && buildLine({ ...args, prompt: '.' }) ? viaLaunchFile() : null
+  }
+  // Why the typed budget (#24257's): nothing stages a line the host types raw, and a long typed
+  // line is truncated or submitted early. Hermes's line never holds the text.
+  const hostTypesLineRaw = args.launchHostIsPaired === true || args.platform === 'win32'
+  if (!readsEnv && hostTypesLineRaw && !typedStartupLineFits(plan.launchCommand)) {
+    return viaLaunchFile()
+  }
+  return { carry: 'on-line', plan }
+}
+
+/** The host writes the path inside this line's quoting, so the file carries which one it is. */
+function withQuoting(launchFile: LaunchFile, shell: AgentStartupShell): LaunchFile {
+  return { ...launchFile, quoting: shell }
+}
+
+/** The host puts the launch file's private directory where the placeholder is, like its path. */
+export function launchFileDirectoryGrant(
+  agent: TuiAgent,
+  launchFile: LaunchFile | undefined,
+  shell: AgentStartupShell
+): string {
+  const flag = TUI_AGENT_CONFIG[agent].launchFileDirectoryFlag
+  if (!launchFile || !flag) {
+    return ''
+  }
+  return ` ${quoteStartupArg(`${flag}=${launchFileDirectoryPlaceholder(launchFile.placeholder)}`, shell)}`
+}
diff --git a/src/shared/launch-prompt-file.ts b/src/shared/launch-prompt-file.ts
index 4c563404fc7..0c99fc1251e 100644
--- a/src/shared/launch-prompt-file.ts
+++ b/src/shared/launch-prompt-file.ts
@@ -49,20 +49,6 @@ export function isLaunchFilePointer(prompt: string): boolean {
   )
 }
 
-/**
- * The prompt a launch line carries, and the file behind it when the prompt is too
- * long or sensitive. The pointer names a placeholder the execution host replaces.
- */
-export function planLaunchPrompt(
-  prompt: string,
-  options: { sensitive?: boolean } = {}
-): { prompt: string; launchFile?: LaunchFile } {
-  const sensitive = options.sensitive === true
-  return sensitive || prompt.length > MAX_INLINE_LAUNCH_PROMPT_CHARS
-    ? carryInLaunchFile(prompt, sensitive)
-    : { prompt }
-}
-
 export function carryInLaunchFile(
   content: string,
   sensitive: boolean
@@ -103,9 +89,17 @@ export function parseLaunchFile(value: unknown): LaunchFile | undefined {
 
 const LAUNCH_FILE_UNAVAILABLE_CODE = 'launch_file_unavailable'
 
-/** The refusal a host sends when it could not write a launch file; it reaches the user as is. */
-export function describeLaunchFileUnavailable(reason: string): string {
-  return `Orca could not write the file that carries the agent's prompt (${reason}), so the agent was not started. [${LAUNCH_FILE_UNAVAILABLE_CODE}]`
+/** The refusal a host sends when it could not write what carries the prompt (a launch file, or the
+ *  staged script holding a long line); it reaches the user as is. */
+export function describeLaunchFileUnavailable(
+  reason: string,
+  carrier: 'file' | 'staged-line' = 'file'
+): string {
+  const what =
+    carrier === 'file'
+      ? "the file that carries the agent's prompt"
+      : "the script that carries the agent's launch line and prompt"
+  return `Orca could not write ${what} (${reason}), so the agent was not started. [${LAUNCH_FILE_UNAVAILABLE_CODE}]`
 }
 
 /** Also matches the `launch_file_unavailable: <reason>` form earlier builds sent. */
diff --git a/src/shared/launch-prompt-plan.test-fixture.ts b/src/shared/launch-prompt-plan.test-fixture.ts
new file mode 100644
index 00000000000..782eba00086
--- /dev/null
+++ b/src/shared/launch-prompt-plan.test-fixture.ts
@@ -0,0 +1,44 @@
+import type { LaunchFile } from './launch-prompt-file'
+import type { LaunchPromptCarry } from './launch-prompt-carry'
+import {
+  planLaunchPrompt,
+  type AgentLaunchPromptArgs,
+  type AgentStartupPlan
+} from './tui-agent-startup'
+
+/** A prompted plan flattened for tests that assert on the line it builds. */
+export type PlannedLaunchForTest = AgentStartupPlan & {
+  carry: LaunchPromptCarry
+  launchFile?: LaunchFile
+  /** The prompt left for the paste after ready; null when the launch carried it. */
+  pasteAfterReady: string | null
+}
+
+/** `planLaunchPrompt`, flattened; null where it answers null, or for an empty prompt unless
+ *  `allowEmptyPromptLaunch`, as the builder did before the outcome was explicit. */
+export function planLaunchForTest(
+  args: AgentLaunchPromptArgs & { allowEmptyPromptLaunch?: boolean }
+): PlannedLaunchForTest | null {
+  const { allowEmptyPromptLaunch, ...launchArgs } = args
+  if (allowEmptyPromptLaunch !== true && !args.prompt.trim()) {
+    return null
+  }
+  const planned = planLaunchPrompt(launchArgs)
+  if (!planned) {
+    return null
+  }
+  switch (planned.carry) {
+    case 'none':
+    case 'on-line':
+      return { ...planned.plan, carry: planned.carry, pasteAfterReady: null }
+    case 'launch-file':
+      return {
+        ...planned.plan,
+        carry: planned.carry,
+        launchFile: planned.launchFile,
+        pasteAfterReady: null
+      }
+    case 'paste-after-ready':
+      return { ...planned.cleanPlan, carry: planned.carry, pasteAfterReady: planned.text }
+  }
+}
diff --git a/src/shared/local-build-compatibility-contract.json b/src/shared/local-build-compatibility-contract.json
index 0e90a402914..bfebe8c975c 100644
--- a/src/shared/local-build-compatibility-contract.json
+++ b/src/shared/local-build-compatibility-contract.json
@@ -3,9 +3,9 @@
   "appId": "com.stablyai.orca",
   "stateSchemaVersion": 1,
   "readableStateSchemaVersions": [1],
-  "daemonProtocolVersion": 40,
+  "daemonProtocolVersion": 41,
   "attachableDaemonProtocolVersions": [
     1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26,
-    27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40
+    27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41
   ]
 }
diff --git a/src/shared/local-build-compatibility-contract.ts b/src/shared/local-build-compatibility-contract.ts
index 75a414b2a5f..afda2d8f227 100644
--- a/src/shared/local-build-compatibility-contract.ts
+++ b/src/shared/local-build-compatibility-contract.ts
@@ -3,9 +3,9 @@ export const LOCAL_BUILD_COMPATIBILITY_CONTRACT = {
   appId: 'com.stablyai.orca',
   stateSchemaVersion: 1,
   readableStateSchemaVersions: [1],
-  daemonProtocolVersion: 40,
+  daemonProtocolVersion: 41,
   attachableDaemonProtocolVersions: [
     1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26,
-    27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40
+    27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41
   ]
 } as const
diff --git a/src/shared/omp-fresh-launch-shell.test.ts b/src/shared/omp-fresh-launch-shell.test.ts
index ebdc7581a9a..5c41cdde06f 100644
--- a/src/shared/omp-fresh-launch-shell.test.ts
+++ b/src/shared/omp-fresh-launch-shell.test.ts
@@ -3,8 +3,8 @@ import { mkdtemp, writeFile, readFile, rm } from 'node:fs/promises'
 import { tmpdir } from 'node:os'
 import { delimiter, join } from 'node:path'
 import { afterEach, expect, it } from 'vitest'
+import { planLaunchForTest } from './launch-prompt-plan.test-fixture'
 import { runProcess } from './child-process/run-process'
-import { buildAgentStartupPlan } from './tui-agent-startup'
 
 const roots: string[] = []
 afterEach(async () => {
@@ -27,7 +27,7 @@ it.each(shells)(
     const config = join(root, 'fresh settings.yml')
     const capture = join(root, 'argv')
     await writeFile(config, 'autoResume: false\n')
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'omp',
       prompt: 'task with spaces',
       cmdOverrides: {},
diff --git a/src/shared/omp-fresh-launch.test.ts b/src/shared/omp-fresh-launch.test.ts
index eb8c94e2e5c..f7c6bf68d22 100644
--- a/src/shared/omp-fresh-launch.test.ts
+++ b/src/shared/omp-fresh-launch.test.ts
@@ -1,6 +1,6 @@
 import { describe, expect, it } from 'vitest'
+import { planLaunchForTest } from './launch-prompt-plan.test-fixture'
 import { withFreshOmpLaunch } from './omp-fresh-launch'
-import { buildAgentStartupPlan } from './tui-agent-startup'
 
 describe('OMP fresh launch intent', () => {
   it.each(['omp', 'omp launch', 'omp --model provider/model', 'omp --config user.yml'])(
@@ -45,7 +45,7 @@ describe('OMP fresh launch intent', () => {
     expect(withFreshOmpLaunch('omp', 'cmd')).toContain('omp --config "%ORCA_OMP_FRESH_CONFIG%"')
   })
   it('keeps fresh intent out of saved resume command and environment', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'omp',
       prompt: 'new task',
       cmdOverrides: {},
@@ -56,7 +56,7 @@ describe('OMP fresh launch intent', () => {
     expect(plan?.env).toBeUndefined()
   })
   it('requires host-owned configuration for fresh SSH launches', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'omp',
       prompt: 'new task',
       cmdOverrides: {},
diff --git a/src/shared/qoder-agent.test.ts b/src/shared/qoder-agent.test.ts
index 8db665e7209..2ad44aab180 100644
--- a/src/shared/qoder-agent.test.ts
+++ b/src/shared/qoder-agent.test.ts
@@ -1,6 +1,7 @@
 import { readFileSync } from 'node:fs'
 import { join } from 'node:path'
 import { describe, expect, it } from 'vitest'
+import { planLaunchForTest } from './launch-prompt-plan.test-fixture'
 import {
   recognizeAgentProcessFromCommandLine,
   isExpectedAgentProcess
@@ -11,7 +12,6 @@ import {
   normalizeTerminalTitle,
   detectAgentStatusFromTitle
 } from './agent-detection'
-import { buildAgentStartupPlan } from './tui-agent-startup'
 import { getAgentResumeArgv } from './agent-session-resume'
 import { createHookListenerState } from './agent-hook-listener/listener-state'
 import { normalizeAndAccept } from './agent-hook-listener-test-harness'
@@ -121,14 +121,14 @@ it('replays captured unauthenticated startup, prompt, idle and resume hooks', ()
 it.each(['darwin', 'linux', 'win32'] as const)(
   'launches a Qoder prompt safely on %s',
   (platform) => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'qoder',
       prompt: "fix Bob's branch; echo $HOME",
       cmdOverrides: {},
       platform
     })
     expect(plan?.expectedProcess).toBe('qodercli')
-    expect(plan?.followupPrompt).toBeNull()
+    expect(plan?.pasteAfterReady).toBeNull()
     expect(plan?.launchCommand).toBe(
       platform === 'win32'
         ? "qodercli --prompt-interactive 'fix Bob''s branch; echo $HOME'"
diff --git a/src/shared/startup-command-staging.node-pty.test.ts b/src/shared/startup-command-staging.node-pty.test.ts
index 2619a92c805..f0e202b4393 100644
--- a/src/shared/startup-command-staging.node-pty.test.ts
+++ b/src/shared/startup-command-staging.node-pty.test.ts
@@ -23,7 +23,7 @@ import * as pty from 'node-pty'
 import { afterAll, describe, expect, it } from 'vitest'
 import { resolveFishBinary } from './fish-binary-requirement'
 import { writeLaunchFile, type WrittenLaunchFile } from './launch-file-writing'
-import { buildLaunchFilePointer, planLaunchPrompt } from './launch-prompt-file'
+import { buildLaunchFilePointer, carryInLaunchFile } from './launch-prompt-file'
 import { stageStartupCommand } from './startup-command-staging'
 import { buildStartupCommandSubmission } from './startup-command-submission'
 import { quoteStartupArg } from './tui-agent-startup-shell'
@@ -181,7 +181,7 @@ describeShells('a launch file named on a real command line', () => {
   for (const shell of SHELLS) {
     it(`hands ${shell.name}'s agent a pointer to the full prompt`, async () => {
       const prompt = `${HOSTILE}\n${'z'.repeat(20_000)}\n`
-      const planned = planLaunchPrompt(prompt)
+      const planned = carryInLaunchFile(prompt, false)
       let written: WrittenLaunchFile | undefined
       const run = await launchInRealShell(shell, planned.prompt, (command) => {
         written = writeLaunchFile({
diff --git a/src/shared/startup-command-staging.ts b/src/shared/startup-command-staging.ts
index 84f66c3d89b..2f3129e9a4e 100644
--- a/src/shared/startup-command-staging.ts
+++ b/src/shared/startup-command-staging.ts
@@ -68,11 +68,16 @@ export function shouldStageStartupCommand(args: {
     return args.orcaBuiltLine === true
   }
   return (
-    !typedStartupLineFits(stripSubmitTerminator(args.command)) &&
+    startupLineNeedsStaging(args.command) &&
     (SOURCING_SHELLS.has(shellName) || args.orcaBuiltLine === true)
   )
 }
 
+/** Whether a line is too long or multi-line to type as it is, in a shell that can source a script. */
+export function startupLineNeedsStaging(command: string): boolean {
+  return !typedStartupLineFits(stripSubmitTerminator(command))
+}
+
 function stagedScriptLine(shellName: string | null, quotedPath: string): string {
   if (shellName === 'fish') {
     // Why eval: fish runs a sourced file without job control; eval runs it as if typed.
diff --git a/src/shared/startup-line-prompt-carry.test.ts b/src/shared/startup-line-prompt-carry.test.ts
deleted file mode 100644
index cbcc56b871e..00000000000
--- a/src/shared/startup-line-prompt-carry.test.ts
+++ /dev/null
@@ -1,175 +0,0 @@
-import { describe, expect, it } from 'vitest'
-import { MAX_INLINE_LAUNCH_PROMPT_CHARS } from './launch-prompt-file'
-import { planStartupWithLaunchPrompt } from './startup-line-prompt-carry'
-import type { TuiAgent } from './tui-agent'
-import { RUNTIME_CAPABILITIES } from './protocol-version'
-import { AGENT_LAUNCH_PROMPT_CARRY_RUNTIME_CAPABILITY } from './agent-launch-runtime-capability'
-import {
-  TYPED_STARTUP_LINE_BUDGET_BYTES,
-  WINDOWS_TYPED_STARTUP_LINE_MAX_CHARS,
-  typedStartupLineFits,
-  windowsTypedStartupLineFits
-} from './typed-startup-line'
-
-function plan(
-  agent: TuiAgent,
-  prompt: string,
-  extra: {
-    platform?: NodeJS.Platform
-    shell?: 'cmd' | 'powershell'
-    sensitive?: boolean
-  } = {}
-) {
-  return planStartupWithLaunchPrompt(
-    {
-      agent,
-      cmdOverrides: {},
-      platform: extra.platform ?? 'darwin',
-      ...(extra.shell ? { shell: extra.shell } : {})
-    },
-    prompt,
-    extra.sensitive ? { sensitive: true } : {}
-  )
-}
-
-describe('a launch prompt on the command line', () => {
-  it('carries a short prompt inline', () => {
-    const { plan: startup, launchFile } = plan('claude', 'explain this repo')
-    expect(launchFile).toBeUndefined()
-    expect(startup?.launchCommand).toContain('explain this repo')
-    expect(startup?.followupPrompt).toBeNull()
-  })
-
-  it.each([
-    ['a 600-byte', 'x'.repeat(600)],
-    ['a multi-line', 'first line\nsecond line'],
-    ['a key-bearing', 'see\tthis \x1b[31mred']
-  ])('carries %s prompt whole on a POSIX host, which stages the typed line', (_label, prompt) => {
-    const { plan: startup, launchFile } = plan('codex', prompt)
-    expect(launchFile).toBeUndefined()
-    expect(startup?.launchCommand).toContain(prompt)
-    expect(startup?.followupPrompt).toBeNull()
-  })
-
-  it('points at a launch file past the argv ceiling, with the full text in the file', () => {
-    const prompt = `${'y'.repeat(MAX_INLINE_LAUNCH_PROMPT_CHARS)}z`
-    const { plan: startup, launchFile } = plan('claude', prompt)
-    expect(launchFile?.content).toBe(prompt)
-    expect(launchFile?.sensitive).toBe(false)
-    expect(startup?.launchCommand).toContain(launchFile?.placeholder)
-    expect(startup?.launchCommand).not.toContain('yyyy')
-  })
-
-  it('points a sensitive prompt at a launch file whatever its size', () => {
-    const { plan: startup, launchFile } = plan('claude', 'token dcap_abc', { sensitive: true })
-    expect(launchFile).toMatchObject({ content: 'token dcap_abc', sensitive: true })
-    expect(startup?.launchCommand).not.toContain('dcap_abc')
-  })
-
-  it.each([
-    ['cmd', 'first line\nsecond line'],
-    ['powershell', 'first line\r\nsecond line']
-  ] as const)(
-    'points a multi-line prompt at a launch file on a Windows %s host, which cannot stage',
-    (shell, prompt) => {
-      const { plan: startup, launchFile } = plan('codex', prompt, { platform: 'win32', shell })
-      expect(launchFile?.content).toBe(prompt)
-      expect(startup?.launchCommand).not.toContain('first line')
-      expect(typedStartupLineFits(startup?.launchCommand ?? '\n')).toBe(true)
-    }
-  )
-
-  it('keeps a long single-line prompt inline on Windows up to cmd’s line cap', () => {
-    const inline = plan('claude', 'x'.repeat(600), { platform: 'win32', shell: 'cmd' })
-    expect(inline.launchFile).toBeUndefined()
-    expect(inline.plan?.launchCommand).toContain('x'.repeat(600))
-
-    const pastCap = plan('claude', 'x'.repeat(WINDOWS_TYPED_STARTUP_LINE_MAX_CHARS), {
-      platform: 'win32',
-      shell: 'cmd'
-    })
-    expect(pastCap.launchFile?.content).toBe('x'.repeat(WINDOWS_TYPED_STARTUP_LINE_MAX_CHARS))
-  })
-
-  it('carries a multi-line Hermes prompt in its spawn env, whose typed line never holds it', () => {
-    const multiLine = 'line one\nline two'
-    const { plan: startup, launchFile } = plan('hermes', multiLine, { platform: 'win32' })
-    expect(launchFile).toBeUndefined()
-    expect(startup?.launchCommand).not.toContain('line one')
-    expect(Object.values(startup?.env ?? {})).toContain(multiLine)
-  })
-
-  it('leaves a Hermes prompt past its env budget for the paste after ready', () => {
-    const { plan: startup, launchFile } = plan('hermes', 'x'.repeat(30_000))
-    expect(launchFile).toBeUndefined()
-    expect(startup?.launchCommand).toBe('hermes --tui')
-    expect(startup?.followupPrompt).toBe('x'.repeat(30_000))
-  })
-
-  it('leaves CJK Hermes text under 16,384 chars but past its env bytes for the paste', () => {
-    const cjk = '修'.repeat(9_000)
-    const { plan: startup, launchFile } = plan('hermes', cjk)
-    expect(launchFile).toBeUndefined()
-    expect(startup?.followupPrompt).toBe(cjk)
-  })
-
-  it('puts a prompt that needs a launch file there only for an agent measured reading it', () => {
-    const long = 'y'.repeat(MAX_INLINE_LAUNCH_PROMPT_CHARS + 1)
-    expect(plan('claude', long).launchFile?.content).toBe(long)
-    expect(plan('codex', long).launchFile?.content).toBe(long)
-    for (const agent of ['gemini', 'cursor', 'antigravity', 'grok'] as const) {
-      const { plan: startup, launchFile } = plan(agent, long)
-      expect(launchFile).toBeUndefined()
-      expect(startup?.followupPrompt).toBe(long)
-      expect(startup?.launchCommand).not.toContain(long)
-    }
-  })
-
-  it('pastes a Windows-damaged prompt for an agent not measured reading its launch file', () => {
-    const { plan: startup, launchFile } = plan('gemini', 'line one\nline two', {
-      platform: 'win32',
-      shell: 'powershell'
-    })
-    expect(launchFile).toBeUndefined()
-    expect(startup?.followupPrompt).toBe('line one\nline two')
-  })
-
-  it('leaves a stdin-after-start agent’s prompt for its caller to paste', () => {
-    const { plan: startup, launchFile } = plan('aider', 'fix it')
-    expect(launchFile).toBeUndefined()
-    expect(startup?.followupPrompt).toBe('fix it')
-  })
-
-  it('launches clean for an empty prompt', () => {
-    const { plan: startup, launchFile } = plan('claude', '   ')
-    expect(launchFile).toBeUndefined()
-    expect(startup?.launchCommand).toBe('claude')
-  })
-})
-
-describe('whether a line can be typed as it is', () => {
-  it('holds a POSIX line to half of macOS MAX_CANON', () => {
-    expect(typedStartupLineFits('x'.repeat(TYPED_STARTUP_LINE_BUDGET_BYTES))).toBe(true)
-    expect(typedStartupLineFits('x'.repeat(TYPED_STARTUP_LINE_BUDGET_BYTES + 1))).toBe(false)
-    expect(typedStartupLineFits('é'.repeat(TYPED_STARTUP_LINE_BUDGET_BYTES / 2 + 1))).toBe(false)
-  })
-
-  it('holds a Windows line only to cmd’s cap', () => {
-    expect(windowsTypedStartupLineFits('x'.repeat(WINDOWS_TYPED_STARTUP_LINE_MAX_CHARS))).toBe(true)
-    expect(windowsTypedStartupLineFits('x'.repeat(WINDOWS_TYPED_STARTUP_LINE_MAX_CHARS + 1))).toBe(
-      false
-    )
-  })
-
-  it.each(['\n', '\r', '\t', '\x1b', '\x03', '\x7f'])('never types a line with %j', (byte) => {
-    expect(typedStartupLineFits(`a${byte}b`)).toBe(false)
-    expect(windowsTypedStartupLineFits(`a${byte}b`)).toBe(false)
-  })
-})
-
-describe('the capability clients gate a prompted launch on', () => {
-  it('is advertised by every host that delivers a prompt its typed line cannot carry as typed', () => {
-    // An older host types any prompt into the line, so its absence is the gate.
-    expect(RUNTIME_CAPABILITIES).toContain(AGENT_LAUNCH_PROMPT_CARRY_RUNTIME_CAPABILITY)
-  })
-})
diff --git a/src/shared/startup-line-prompt-carry.ts b/src/shared/startup-line-prompt-carry.ts
deleted file mode 100644
index cdf89e5978c..00000000000
--- a/src/shared/startup-line-prompt-carry.ts
+++ /dev/null
@@ -1,32 +0,0 @@
-/**
- * The launch-prompt plan for callers that route sensitive prompts: the decision itself (line, launch
- * file, or paste) is the builder's (`carryLaunchPrompt`), so every launch path shares it.
- */
-
-import type { LaunchFile } from './launch-prompt-file'
-import { buildAgentStartupPlan, type AgentStartupPlan } from './tui-agent-startup'
-
-type StartupPlanInputs = Omit<
-  Parameters<typeof buildAgentStartupPlan>[0],
-  'prompt' | 'allowEmptyPromptLaunch'
->
-
-export type LaunchPromptStartupPlan = {
-  plan: AgentStartupPlan | null
-  /** Written by the execution host before it types the line naming it. */
-  launchFile?: LaunchFile
-}
-
-export function planStartupWithLaunchPrompt(
-  inputs: StartupPlanInputs,
-  prompt: string,
-  options: { sensitive?: boolean } = {}
-): LaunchPromptStartupPlan {
-  const plan = buildAgentStartupPlan({
-    ...inputs,
-    prompt,
-    allowEmptyPromptLaunch: true,
-    ...(options.sensitive ? { sensitive: true } : {})
-  })
-  return { plan, ...(plan?.launchFile ? { launchFile: plan.launchFile } : {}) }
-}
diff --git a/src/shared/startup-plan-launch-file.ts b/src/shared/startup-plan-launch-file.ts
deleted file mode 100644
index 6cff8596509..00000000000
--- a/src/shared/startup-plan-launch-file.ts
+++ /dev/null
@@ -1,126 +0,0 @@
-/** How a startup plan carries a launch file: which prompts move into one, and what the line adds. */
-import {
-  MAX_INLINE_LAUNCH_PROMPT_CHARS,
-  carryInLaunchFile,
-  launchFileDirectoryPlaceholder,
-  type LaunchFile
-} from './launch-prompt-file'
-import { TUI_AGENT_CONFIG } from './tui-agent-config'
-import {
-  isPosixStartupShell,
-  quoteStartupArg,
-  resolveStartupShell,
-  type AgentStartupShell
-} from './tui-agent-startup-shell'
-import { hasControlByte, windowsTypedStartupLineFits } from './typed-startup-line'
-import type { TuiAgent } from './tui-agent'
-
-/**
- * Whether a Windows shell would damage `prompt` as one quoted argument; the closed rule is this plus
- * cmd's 8,191-character line cap (`carryLaunchPrompt`). No quoting keeps a control byte literal: cmd
- * types the line and a line break submits it early. PowerShell runs a short line from its argv, so
- * its damage is the hand-off to the agent (measured): legacy argument passing (5.1 always, 7.x into a
- * `.cmd` shim) splits at an inner `"` and turns a trailing `\` into `"`, and a shim's cmd.exe expands
- * `%NAME%`; a lone `%` stays literal.
- */
-export function windowsShellDamagesPrompt(prompt: string, shell: AgentStartupShell): boolean {
-  if (isPosixStartupShell(shell)) {
-    return false
-  }
-  return (
-    hasControlByte(prompt) ||
-    (shell === 'powershell' &&
-      (prompt.includes('"') || /%[^%]+%/.test(prompt) || prompt.endsWith('\\')))
-  )
-}
-
-/** Why a prefill draft could not be launched, in the user's words, when the Windows shell is why. */
-export function windowsDraftRefusal(draft: string, shell: AgentStartupShell): string | null {
-  return windowsShellDamagesPrompt(draft.trim(), shell)
-    ? "The host's Windows shell would break this draft on the agent's command line (it has a line " +
-        'break or other control character, or on PowerShell a double quote, a %NAME% pair or a ' +
-        'trailing backslash), so the agent was not started. Start it without the draft and paste ' +
-        'the draft once it opens.'
-    : null
-}
-
-type CarriedPlanArgs = {
-  agent: TuiAgent
-  prompt: string
-  platform: NodeJS.Platform
-  shell?: AgentStartupShell
-  allowEmptyPromptLaunch?: boolean
-  launchFile?: LaunchFile
-  hostWritesLaunchFile?: boolean
-  sensitive?: boolean
-}
-
-export function agentReadsLaunchFile(agent: TuiAgent): boolean {
-  return TUI_AGENT_CONFIG[agent].readsLaunchFile === true
-}
-
-/**
- * Where a launch prompt rides, decided once for every launch path: on the line, which a POSIX host
- * stages when it is long or multi-line; in a launch file named by a pointer when the prompt is past
- * the argv ceiling, sensitive, damaged by a Windows shell, or past cmd's line cap; or, where no
- * launch file can be written (a paired host) or the agent is not known to read one, left in
- * `followupPrompt` for the paste after ready.
- */
-export function carryLaunchPrompt<
-  A extends CarriedPlanArgs,
-  P extends { launchCommand: string; followupPrompt: string | null }
->(args: A, buildOnLine: (args: A) => P | null): P | null {
-  const text = args.prompt.trim()
-  const mode = TUI_AGENT_CONFIG[args.agent].promptInjectionMode
-  if (!text || args.launchFile || mode === 'stdin-after-start') {
-    return buildOnLine(args)
-  }
-  const readsEnv = mode === 'hermes-query'
-  const pasteAfterReady = (): P | null => {
-    const clean = buildOnLine({ ...args, prompt: '', allowEmptyPromptLaunch: true })
-    return clean && { ...clean, followupPrompt: text }
-  }
-  const viaLaunchFile = (): P | null => {
-    // Why: an agent not measured reading the file would stop on an approval or refuse the path.
-    if (args.hostWritesLaunchFile === false || !agentReadsLaunchFile(args.agent)) {
-      return pasteAfterReady()
-    }
-    const pointer = carryInLaunchFile(text, args.sensitive === true)
-    return buildOnLine({ ...args, prompt: pointer.prompt, launchFile: pointer.launchFile })
-  }
-  const shell = resolveStartupShell(args.platform, args.shell)
-  if (
-    args.sensitive === true ||
-    text.length > MAX_INLINE_LAUNCH_PROMPT_CHARS ||
-    (!readsEnv && windowsShellDamagesPrompt(text, shell))
-  ) {
-    return viaLaunchFile()
-  }
-  const onLine = buildOnLine(args)
-  if (!onLine || readsEnv) {
-    // Hermes reads its prompt from the env and refuses one past that budget, counted in bytes; a
-    // one-character query building proves the budget, not the command, refused it.
-    return !onLine && readsEnv && buildOnLine({ ...args, prompt: '.' }) ? viaLaunchFile() : onLine
-  }
-  return args.platform === 'win32' && !windowsTypedStartupLineFits(onLine.launchCommand)
-    ? viaLaunchFile()
-    : onLine
-}
-
-/** The host writes the path inside this line's quoting, so the file carries which one it is. */
-export function launchFileProps(launchFile: LaunchFile | undefined, shell: AgentStartupShell) {
-  return launchFile ? { launchFile: { ...launchFile, quoting: shell } } : {}
-}
-
-/** The host puts the launch file's private directory where the placeholder is, like its path. */
-export function launchFileDirectoryGrant(
-  agent: TuiAgent,
-  launchFile: LaunchFile | undefined,
-  shell: AgentStartupShell
-): string {
-  const flag = TUI_AGENT_CONFIG[agent].launchFileDirectoryFlag
-  if (!launchFile || !flag) {
-    return ''
-  }
-  return ` ${quoteStartupArg(`${flag}=${launchFileDirectoryPlaceholder(launchFile.placeholder)}`, shell)}`
-}
diff --git a/src/shared/tui-agent-prompt-transport.test.ts b/src/shared/tui-agent-prompt-transport.test.ts
index f117493dc71..a10ee091a4a 100644
--- a/src/shared/tui-agent-prompt-transport.test.ts
+++ b/src/shared/tui-agent-prompt-transport.test.ts
@@ -1,8 +1,8 @@
 /**
  * The two answers to "how does this agent take its prompt" must stay one answer.
  *
- * `buildAgentStartupPlan` decides by building a command and leaving `followupPrompt` null when the
- * text went into it. `agentPromptRidesLaunchCommand` has to give the same verdict BEFORE a command
+ * `planLaunchPrompt` decides by building a command and answering `paste-after-ready` when the text
+ * did not go into it. `agentPromptRidesLaunchCommand` has to give the same verdict BEFORE a command
  * exists, because `agent.launch` picks a delivery while it is still choosing what to create. Two
  * readings of one table is exactly the shape that drifts, so this pins them together across every
  * agent: add an agent, or change its injection mode, and the disagreement fails here rather than
@@ -10,8 +10,9 @@
  */
 
 import { describe, expect, it } from 'vitest'
+import { planLaunchForTest } from './launch-prompt-plan.test-fixture'
 import { TUI_AGENT_CONFIG } from './tui-agent-config'
-import { agentPromptRidesLaunchCommand, buildAgentStartupPlan } from './tui-agent-startup'
+import { agentPromptRidesLaunchCommand } from './tui-agent-startup'
 import type { TuiAgent } from './tui-agent'
 
 // oxlint-disable-next-line typescript/consistent-type-assertions -- SAFETY: the config is declared as a total record over TuiAgent, so its own keys are that union.
@@ -20,7 +21,7 @@ const ALL_AGENTS = Object.keys(TUI_AGENT_CONFIG) as TuiAgent[]
 const PROMPT = 'summarize the diff'
 
 function planFor(agent: TuiAgent) {
-  return buildAgentStartupPlan({
+  return planLaunchForTest({
     agent,
     prompt: PROMPT,
     cmdOverrides: {},
@@ -39,9 +40,9 @@ describe('the prompt-transport predicate against the plan it predicts', () => {
   it.each(ALL_AGENTS)('agrees with the built plan for %s', (agent) => {
     const plan = planFor(agent)
     expect(plan).not.toBeNull()
-    // `followupPrompt` is the plan saying "the command does NOT carry this"; the predicate must
+    // `paste-after-ready` is the plan saying "the command does NOT carry this"; the predicate must
     // say the same thing, and it is read before any plan is built.
-    expect(plan!.followupPrompt === null).toBe(agentPromptRidesLaunchCommand(agent))
+    expect(plan!.pasteAfterReady === null).toBe(agentPromptRidesLaunchCommand(agent))
   })
 
   it('puts the prompt in the launch command exactly when it says it does', () => {
@@ -50,7 +51,7 @@ describe('the prompt-transport predicate against the plan it predicts', () => {
       if (!agentPromptRidesLaunchCommand(agent)) {
         // The command must not smuggle the text in some other way.
         expect(plan!.launchCommand).not.toContain(PROMPT)
-        expect(plan!.followupPrompt).toBe(PROMPT)
+        expect(plan!.pasteAfterReady).toBe(PROMPT)
         continue
       }
       // Hermes hands long text through an env var, so the command names the variable, not the text.
diff --git a/src/shared/tui-agent-resume-startup.ts b/src/shared/tui-agent-resume-startup.ts
index f4924b83c47..659e6dadd49 100644
--- a/src/shared/tui-agent-resume-startup.ts
+++ b/src/shared/tui-agent-resume-startup.ts
@@ -62,7 +62,6 @@ export function buildAgentResumeStartupPlan(args: {
     agent: args.agent,
     launchCommand,
     expectedProcess: TUI_AGENT_CONFIG[args.agent].expectedProcess,
-    followupPrompt: null,
     launchConfig,
     ...(args.agent === 'codex' ? { startupCommandDelivery: 'shell-ready' as const } : {}),
     ...(Object.keys(applied).length > 0 ? { sessionOptions: { ...applied } } : {}),
diff --git a/src/shared/tui-agent-startup-hermes.test.ts b/src/shared/tui-agent-startup-hermes.test.ts
index 4cb34a4ab7f..401f7e8b658 100644
--- a/src/shared/tui-agent-startup-hermes.test.ts
+++ b/src/shared/tui-agent-startup-hermes.test.ts
@@ -1,5 +1,5 @@
 import { describe, expect, it } from 'vitest'
-import { buildAgentStartupPlan } from './tui-agent-startup'
+import { planLaunchForTest } from './launch-prompt-plan.test-fixture'
 import {
   unwrapPosixShellScript,
   unwrapPowerShellScript
@@ -11,7 +11,7 @@ import {
 
 describe('hermes startup plans', () => {
   it('moves Hermes command override flags after the chat subcommand', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: 'run privately',
       cmdOverrides: { hermes: 'hermes --tui --provider anthropic' },
@@ -36,7 +36,7 @@ describe('hermes startup plans', () => {
       expected: "& 'C:\\Tools\\hermes.exe' 'chat'"
     }
   ])('preserves Windows paths in Hermes command overrides on $shell', (testCase) => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: 'run it',
       cmdOverrides: { hermes: testCase.override },
@@ -48,7 +48,7 @@ describe('hermes startup plans', () => {
   })
 
   it('removes a configured duplicate chat subcommand', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: 'run it',
       cmdOverrides: { hermes: 'hermes --provider copilot chat --tui' },
@@ -62,7 +62,7 @@ describe('hermes startup plans', () => {
   })
 
   it('preserves an option value named chat', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: 'run it',
       cmdOverrides: { hermes: 'hermes --profile chat --tui' },
@@ -73,7 +73,7 @@ describe('hermes startup plans', () => {
   })
 
   it('keeps Orca ownership of the Hermes startup query and TUI mode', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: 'automation prompt',
       cmdOverrides: { hermes: 'hermes --query override --cli' },
@@ -90,7 +90,7 @@ describe('hermes startup plans', () => {
   })
 
   it('preserves wrapper tokens before the Hermes executable', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: 'run it',
       cmdOverrides: { hermes: 'uv run hermes --tui' },
@@ -101,7 +101,7 @@ describe('hermes startup plans', () => {
   })
 
   it('selects the final Hermes executable token in a wrapper', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: 'run it',
       cmdOverrides: { hermes: 'sudo -u hermes hermes --tui' },
@@ -114,7 +114,7 @@ describe('hermes startup plans', () => {
   })
 
   it('selects the wrapped executable when the wrapper and command both name Hermes', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: 'run it',
       cmdOverrides: { hermes: 'sudo -u hermes hermes chat --tui' },
@@ -127,7 +127,7 @@ describe('hermes startup plans', () => {
   })
 
   it('does not mistake a Hermes option value for a wrapped executable', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: 'run it',
       cmdOverrides: { hermes: 'hermes chat --resume hermes --tui' },
@@ -140,7 +140,7 @@ describe('hermes startup plans', () => {
   })
 
   it('preserves POSIX environment-assignment command prefixes', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: 'run it',
       cmdOverrides: { hermes: 'HERMES_HOME=/tmp/test uv run hermes --tui' },
@@ -154,7 +154,7 @@ describe('hermes startup plans', () => {
 
   it('rejects a Hermes command override with no identifiable executable', () => {
     expect(
-      buildAgentStartupPlan({
+      planLaunchForTest({
         agent: 'hermes',
         prompt: 'run it',
         cmdOverrides: { hermes: 'custom-agent --tui' },
@@ -164,7 +164,7 @@ describe('hermes startup plans', () => {
   })
 
   it('leaves a Hermes query past the safe Windows environment limit for the paste', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: 'x'.repeat(24_000),
       cmdOverrides: {},
@@ -172,13 +172,13 @@ describe('hermes startup plans', () => {
     })
 
     expect(plan?.launchFile).toBeUndefined()
-    expect(plan?.followupPrompt).toBe('x'.repeat(24_000))
+    expect(plan?.pasteAfterReady).toBe('x'.repeat(24_000))
   })
 
   it.each(['quote "this"', 'print %PATH%', 'toggle !feature!', 'inspect C:\\repo\\'])(
     'keeps a complex cmd Hermes query out of command text: %s',
     (prompt) => {
-      const plan = buildAgentStartupPlan({
+      const plan = planLaunchForTest({
         agent: 'hermes',
         prompt,
         cmdOverrides: {},
@@ -192,7 +192,7 @@ describe('hermes startup plans', () => {
   )
 
   it('uses the Windows remote default shell for SSH Hermes queries', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: 'run remotely',
       cmdOverrides: {},
@@ -206,7 +206,7 @@ describe('hermes startup plans', () => {
   })
 
   it('measures POSIX Hermes query limits in UTF-8 bytes, pasting past them', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: '界'.repeat(9_000),
       cmdOverrides: {},
@@ -214,11 +214,11 @@ describe('hermes startup plans', () => {
     })
 
     expect(plan?.launchFile).toBeUndefined()
-    expect(plan?.followupPrompt).toBe('界'.repeat(9_000))
+    expect(plan?.pasteAfterReady).toBe('界'.repeat(9_000))
   })
 
   it('keeps empty Hermes launches on the interactive TUI command', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: '',
       cmdOverrides: {},
@@ -227,6 +227,6 @@ describe('hermes startup plans', () => {
     })
 
     expect(plan?.launchCommand).toBe('hermes --tui')
-    expect(plan?.followupPrompt).toBeNull()
+    expect(plan?.pasteAfterReady).toBeNull()
   })
 })
diff --git a/src/shared/tui-agent-startup.test.ts b/src/shared/tui-agent-startup.test.ts
index 27b487ba598..5ba4245faf1 100644
--- a/src/shared/tui-agent-startup.test.ts
+++ b/src/shared/tui-agent-startup.test.ts
@@ -1,8 +1,8 @@
 import { describe, expect, it } from 'vitest'
+import { planLaunchForTest } from './launch-prompt-plan.test-fixture'
 import {
   buildAgentDraftLaunchPlan,
   buildAgentResumeStartupPlan,
-  buildAgentStartupPlan,
   buildShellCommandFromArgv,
   planAgentCliArgsSuffix
 } from './tui-agent-startup'
@@ -48,7 +48,7 @@ describe('tui agent startup plans', () => {
   // Structured native chat stopped reading the configured arguments; a terminal launch must
   // still spell every token of them, in order, exactly as the user wrote them.
   it('passes the whole configured argument string to a terminal launch', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude',
       prompt: '',
       agentArgs: resolveTuiAgentLaunchArgs('claude', {
@@ -64,7 +64,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('uses POSIX quoting when the target shell is Linux', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude',
       prompt: "fix Bob's branch",
       cmdOverrides: {},
@@ -75,7 +75,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('uses PowerShell quoting by default when the target shell is Windows', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude',
       prompt: "fix Bob's \u2018quoted\u2019 branch",
       cmdOverrides: {},
@@ -93,7 +93,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('uses cmd escaping when requested explicitly', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude',
       prompt: 'fix "quoted" & %PATH%',
       cmdOverrides: {},
@@ -108,7 +108,7 @@ describe('tui agent startup plans', () => {
     'never types a line break into %s: a multi-line prompt rides a launch file',
     (shell) => {
       const prompt = 'fix the build\r\n& echo PWNED\nthen run tests'
-      const plan = buildAgentStartupPlan({
+      const plan = planLaunchForTest({
         agent: 'claude',
         prompt,
         cmdOverrides: {},
@@ -124,7 +124,7 @@ describe('tui agent startup plans', () => {
   )
 
   it('keeps a multi-line prompt inline for a POSIX shell on a POSIX host, which stages it', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude',
       prompt: 'line one\nline two',
       cmdOverrides: {},
@@ -138,7 +138,7 @@ describe('tui agent startup plans', () => {
 
   // Why: a Windows host stages nothing, so Git Bash would get the raw line break typed.
   it('points Git Bash on a Windows host at a launch file for a multi-line prompt', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude',
       prompt: 'line one\nline two',
       cmdOverrides: {},
@@ -163,7 +163,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('terminates Grok options before a flag-shaped POSIX prompt', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'grok',
       prompt: '--version',
       cmdOverrides: {},
@@ -174,7 +174,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('terminates Grok options before a flag-shaped PowerShell prompt', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'grok',
       prompt: '-h',
       cmdOverrides: {},
@@ -185,7 +185,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('terminates Grok options before a subcommand-shaped cmd prompt', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'grok',
       prompt: 'help',
       cmdOverrides: {},
@@ -197,7 +197,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('places the Grok prompt separator after configured agent arguments', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'grok',
       prompt: '--version',
       cmdOverrides: {},
@@ -209,7 +209,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('does not add the Grok prompt separator to other argv agents', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'codex',
       prompt: '--version',
       cmdOverrides: {},
@@ -225,7 +225,7 @@ describe('tui agent startup plans', () => {
     { platform: 'win32' as const, shell: 'cmd' as const }
   ])('delivers multiline Hermes queries through a child-only expansion on $shell', (testCase) => {
     const prompt = 'first line\nsecond "quoted" line with %PATH%'
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt,
       cmdOverrides: {},
@@ -235,7 +235,7 @@ describe('tui agent startup plans', () => {
     })
 
     expect(plan?.launchCommand).not.toContain(prompt)
-    expect(plan?.followupPrompt).toBeNull()
+    expect(plan?.pasteAfterReady).toBeNull()
     expect(plan?.launchConfig.agentCommand).toBe('hermes --tui')
     expect(plan?.env?.ORCA_HERMES_STARTUP_QUERY).toBe(prompt)
     const script =
@@ -259,7 +259,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('uses a sh invocation that POSIX-host PowerShell can parse', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'hermes',
       prompt: 'run it',
       cmdOverrides: {},
@@ -276,7 +276,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('does not launch Codex with the Orca profile when agent status hooks are enabled', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'codex',
       prompt: 'fix it',
       cmdOverrides: {},
@@ -288,7 +288,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('keeps plain empty Codex startup on the fast delivery path', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'codex',
       prompt: '',
       cmdOverrides: {},
@@ -300,13 +300,14 @@ describe('tui agent startup plans', () => {
       agent: 'codex',
       launchCommand: 'codex',
       expectedProcess: 'codex',
-      followupPrompt: null,
+      carry: 'none',
+      pasteAfterReady: null,
       launchConfig: { agentCommand: 'codex', agentArgs: '', agentEnv: {} }
     })
   })
 
   it('launches Claude without Orca settings injection', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude',
       prompt: 'fix it',
       cmdOverrides: {},
@@ -318,7 +319,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('uses the Linux Orca CLI command for Claude Agent Teams launches', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude-agent-teams',
       prompt: '',
       cmdOverrides: {},
@@ -334,7 +335,7 @@ describe('tui agent startup plans', () => {
     // `orca-ide` GNOME-screen-reader workaround), so a remote launch must not
     // emit `orca-ide claude-teams` — that name is not on the remote PATH and
     // `claude-teams` is rejected by the relay's CLI switch (issue #6500).
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude-agent-teams',
       prompt: '',
       cmdOverrides: {},
@@ -349,7 +350,7 @@ describe('tui agent startup plans', () => {
   it('keeps the Windows orca.cmd shim for Claude Agent Teams on SSH remotes', () => {
     // Why: the Windows remote shim is also `orca.cmd`, matching the local
     // win32 override, so remoteness must not alter the Windows command.
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude-agent-teams',
       prompt: '',
       cmdOverrides: {},
@@ -365,7 +366,7 @@ describe('tui agent startup plans', () => {
     // Why: the `orca-ide` rename is still required for a local Linux desktop
     // install (avoids shadowing the GNOME Orca screen reader), so an explicit
     // isRemote:false must preserve it.
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude-agent-teams',
       prompt: '',
       cmdOverrides: {},
@@ -378,7 +379,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('launches OpenClaude as a distinct argv agent', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'openclaude',
       prompt: 'fix it',
       cmdOverrides: {},
@@ -389,13 +390,14 @@ describe('tui agent startup plans', () => {
       agent: 'openclaude',
       launchCommand: "openclaude 'fix it'",
       expectedProcess: 'openclaude',
-      followupPrompt: null,
+      carry: 'on-line',
+      pasteAfterReady: null,
       launchConfig: { agentCommand: 'openclaude', agentArgs: '', agentEnv: {} }
     })
   })
 
   it('launches Mistral Vibe through the installed vibe executable', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'mistral-vibe',
       prompt: 'fix it',
       cmdOverrides: {},
@@ -406,13 +408,14 @@ describe('tui agent startup plans', () => {
       agent: 'mistral-vibe',
       launchCommand: 'vibe',
       expectedProcess: 'vibe',
-      followupPrompt: 'fix it',
+      carry: 'paste-after-ready',
+      pasteAfterReady: 'fix it',
       launchConfig: { agentCommand: 'vibe', agentArgs: '', agentEnv: {} }
     })
   })
 
   it('launches Qwen Code through the installed qwen executable', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'qwen-code',
       prompt: 'fix it',
       cmdOverrides: {},
@@ -423,7 +426,8 @@ describe('tui agent startup plans', () => {
       agent: 'qwen-code',
       launchCommand: 'qwen',
       expectedProcess: 'qwen',
-      followupPrompt: 'fix it',
+      carry: 'paste-after-ready',
+      pasteAfterReady: 'fix it',
       launchConfig: { agentCommand: 'qwen', agentArgs: '', agentEnv: {} }
     })
   })
@@ -440,7 +444,7 @@ describe('tui agent startup plans', () => {
   ] as const)(
     'launches Muse in %s mode on %s/%s before delivering its prompt',
     (_, platform, shell, defaults, command) => {
-      const plan = buildAgentStartupPlan({
+      const plan = planLaunchForTest({
         agent: 'muse',
         prompt: 'fix it',
         cmdOverrides: {},
@@ -453,7 +457,7 @@ describe('tui agent startup plans', () => {
         agent: 'muse',
         launchCommand: command,
         expectedProcess: 'muse',
-        followupPrompt: 'fix it'
+        pasteAfterReady: 'fix it'
       })
     }
   )
@@ -461,19 +465,19 @@ describe('tui agent startup plans', () => {
   it.each(['exec', 'resume', '--help'])(
     'delivers the reserved Muse prompt %s as text',
     (prompt) => {
-      const plan = buildAgentStartupPlan({
+      const plan = planLaunchForTest({
         agent: 'muse',
         prompt,
         cmdOverrides: {},
         platform: 'linux'
       })
       expect(plan?.launchCommand).toBe('muse --trust-workspace')
-      expect(plan?.followupPrompt).toBe(prompt)
+      expect(plan?.pasteAfterReady).toBe(prompt)
     }
   )
 
   it('leaves Claude command overrides untouched', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude',
       prompt: 'fix it',
       cmdOverrides: { claude: 'claude --dangerously-skip-permissions' },
@@ -484,7 +488,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('leaves Codex command overrides untouched', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'codex',
       prompt: 'fix it',
       cmdOverrides: { codex: 'codex --profile work' },
@@ -587,7 +591,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('appends shell-quoted CLI arguments before prompt delivery flags', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude',
       prompt: 'fix it',
       cmdOverrides: {},
@@ -601,7 +605,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('uses PowerShell quoting for CLI arguments on Windows', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude',
       prompt: 'fix it',
       cmdOverrides: {},
@@ -613,7 +617,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('carries agent launch environment defaults into startup plans', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'goose',
       prompt: '',
       cmdOverrides: {},
@@ -632,7 +636,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('captures empty args and env as explicit launch config values', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'claude',
       prompt: '',
       cmdOverrides: {},
@@ -649,7 +653,7 @@ describe('tui agent startup plans', () => {
     const agentDefaultArgs = normalizeTuiAgentArgsRecord({
       opencode: '--dangerously-skip-permissions'
     })
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'opencode',
       prompt: 'fix it',
       cmdOverrides: {},
@@ -709,7 +713,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('appends Kiro trust defaults to the chat subcommand that accepts them', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'kiro',
       prompt: 'fix it',
       cmdOverrides: {},
@@ -721,7 +725,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('launches Continue through the documented cn binary', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'continue',
       prompt: 'fix it',
       cmdOverrides: {},
@@ -795,7 +799,7 @@ describe('tui agent startup plans', () => {
   })
 
   it('launches Devin with stdin-after-start prompt delivery', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'devin',
       prompt: 'fix the tests',
       cmdOverrides: {},
@@ -806,7 +810,8 @@ describe('tui agent startup plans', () => {
       agent: 'devin',
       launchCommand: "devin '--permission-mode' 'bypass' '--respect-workspace-trust' 'false'",
       expectedProcess: 'devin',
-      followupPrompt: 'fix the tests',
+      carry: 'paste-after-ready',
+      pasteAfterReady: 'fix the tests',
       launchConfig: {
         agentCommand: "devin '--permission-mode' 'bypass' '--respect-workspace-trust' 'false'",
         agentArgs: '--permission-mode bypass --respect-workspace-trust false',
diff --git a/src/shared/tui-agent-startup.ts b/src/shared/tui-agent-startup.ts
index 99009b63ffa..ee016edf1f5 100644
--- a/src/shared/tui-agent-startup.ts
+++ b/src/shared/tui-agent-startup.ts
@@ -13,9 +13,9 @@ import type { LaunchFile } from './launch-prompt-file'
 import {
   carryLaunchPrompt,
   launchFileDirectoryGrant,
-  launchFileProps,
-  windowsShellDamagesPrompt
-} from './startup-plan-launch-file'
+  windowsShellDamagesPrompt,
+  type LaunchPromptPlan
+} from './launch-prompt-carry'
 import { TUI_AGENT_CONFIG } from './tui-agent-config'
 import type { StartupCommandDelivery } from './codex-startup-delivery'
 import { buildSleepingAgentLaunchConfig } from './sleeping-agent-launch-config'
@@ -31,7 +31,6 @@ export type AgentStartupPlan = {
   agent: TuiAgent
   launchCommand: string
   expectedProcess: string
-  followupPrompt: string | null
   launchConfig: SleepingAgentLaunchConfig
   launchToken?: string
   draftPrompt?: string | null
@@ -40,21 +39,17 @@ export type AgentStartupPlan = {
   /** Values actually emitted into this launch command, kept as base model ids
    * so the native-chat surface can render only launch-backed state. */
   sessionOptions?: Record<string, SessionOptionValue>
-  /** Holds the prompt the command points at; the host writes it before typing the command. */
-  launchFile?: LaunchFile
 }
 
 function appliedSessionOptionProps(values: Record<string, SessionOptionValue>) {
   return Object.keys(values).length > 0 ? { sessionOptions: { ...values } } : {}
 }
 
-export type AgentStartupPlanArgs = {
+export type AgentStartupPlanInputs = {
   agent: TuiAgent
-  prompt: string
   cmdOverrides: Partial<Record<TuiAgent, string>>
   platform: NodeJS.Platform
   shell?: AgentStartupShell
-  allowEmptyPromptLaunch?: boolean
   agentArgs?: string | null
   agentEnv?: Record<string, string> | null
   sessionOptions?: Record<string, SessionOptionValue>
@@ -62,25 +57,38 @@ export type AgentStartupPlanArgs = {
   /** Why: SSH remotes deploy the CLI shim as plain `orca`, so the Linux-only
    * `orca-ide` rename must be skipped for remote launches. */
   isRemote?: boolean
-  /** The file `prompt` points at, when the caller already moved the prompt into one. */
+}
+
+/** An agent launch with no prompt; a launch that offers one is planned by `planLaunchPrompt`. */
+export function buildAgentStartupPlan(
+  args: AgentStartupPlanInputs & { prompt?: ''; allowEmptyPromptLaunch?: boolean }
+): AgentStartupPlan | null {
+  return args.allowEmptyPromptLaunch === true
+    ? buildPlanWithPromptOnLine({ ...args, prompt: '' })
+    : null
+}
+
+export type AgentLaunchPromptArgs = AgentStartupPlanInputs & {
+  prompt: string
+  /** The file `prompt` already points at, when the caller wrote its own. */
   launchFile?: LaunchFile
-  /** False for a paired host, which is sent a command and never a launch file: a prompt that would
-   *  need one launches clean and is left in `followupPrompt` for the paste after ready. */
-  hostWritesLaunchFile?: boolean
-  /** The caller minted a secret in the prompt: it rides a launch file, never argv or history. */
-  sensitive?: boolean
+  /** Another Orca this client drives; see `CarriedPlanArgs`. */
+  launchHostIsPaired?: boolean
 }
 
 /**
- * The one place that decides whether a launch prompt rides the line, a launch file, or the paste
- * after ready (`carryLaunchPrompt`); every launch path builds through it.
+ * A launch that offers a prompt: where it rides (`carryLaunchPrompt`) and the plan to launch. Every
+ * launch path plans its prompt here, and must switch on the outcome to reach the plan.
  */
-export function buildAgentStartupPlan(args: AgentStartupPlanArgs): AgentStartupPlan | null {
+export function planLaunchPrompt(
+  args: AgentLaunchPromptArgs
+): LaunchPromptPlan<AgentStartupPlan> | null {
   return carryLaunchPrompt(args, buildPlanWithPromptOnLine)
 }
 
-function buildPlanWithPromptOnLine(args: AgentStartupPlanArgs): AgentStartupPlan | null {
-  const { agent, prompt, cmdOverrides, platform, allowEmptyPromptLaunch = false } = args
+/** Builds the line with `prompt` on it; the carry rule decides whether it should. */
+function buildPlanWithPromptOnLine(args: AgentLaunchPromptArgs): AgentStartupPlan | null {
+  const { agent, prompt, cmdOverrides, platform } = args
   const shell = resolveStartupShell(platform, args.shell)
   const trimmedPrompt = prompt.trim()
   const config = TUI_AGENT_CONFIG[agent]
@@ -108,24 +116,18 @@ function buildPlanWithPromptOnLine(args: AgentStartupPlanArgs): AgentStartupPlan
   })
 
   if (!trimmedPrompt) {
-    if (!allowEmptyPromptLaunch) {
-      return null
-    }
     return {
       agent,
       launchCommand,
       expectedProcess: config.expectedProcess,
-      followupPrompt: null,
       launchConfig,
       ...appliedSessionOptionProps(baseCommand.appliedSessionOptions),
       ...(args.agentEnv ? { env: { ...args.agentEnv } } : {})
     }
   }
 
-  const launchFile = args.launchFile
   const quotedPrompt = quoteStartupArg(trimmedPrompt, shell)
-  const fileProps = launchFileProps(launchFile, shell)
-  const grant = launchFileDirectoryGrant(agent, launchFile, shell)
+  const grant = launchFileDirectoryGrant(agent, args.launchFile, shell)
 
   if (config.promptInjectionMode === 'argv') {
     const promptSeparator = config.argvPromptSeparator ? ` ${config.argvPromptSeparator}` : ''
@@ -140,12 +142,10 @@ function buildPlanWithPromptOnLine(args: AgentStartupPlanArgs): AgentStartupPlan
             )
           : `${launchCommand}${grant}${promptSeparator} ${quotedPrompt}`,
       expectedProcess: config.expectedProcess,
-      followupPrompt: null,
       launchConfig,
       ...appliedSessionOptionProps(baseCommand.appliedSessionOptions),
       ...(agent === 'codex' ? { startupCommandDelivery: 'shell-ready' as const } : {}),
-      ...(args.agentEnv ? { env: { ...args.agentEnv } } : {}),
-      ...fileProps
+      ...(args.agentEnv ? { env: { ...args.agentEnv } } : {})
     }
   }
 
@@ -154,11 +154,9 @@ function buildPlanWithPromptOnLine(args: AgentStartupPlanArgs): AgentStartupPlan
       agent,
       launchCommand: `${launchCommand}${grant} --prompt ${quotedPrompt}`,
       expectedProcess: config.expectedProcess,
-      followupPrompt: null,
       launchConfig,
       ...appliedSessionOptionProps(baseCommand.appliedSessionOptions),
-      ...(args.agentEnv ? { env: { ...args.agentEnv } } : {}),
-      ...fileProps
+      ...(args.agentEnv ? { env: { ...args.agentEnv } } : {})
     }
   }
 
@@ -181,12 +179,9 @@ function buildPlanWithPromptOnLine(args: AgentStartupPlanArgs): AgentStartupPlan
       // only bounds and quotes the native invocation before starting the TUI.
       launchCommand: queryPlan.command,
       expectedProcess: config.expectedProcess,
-      followupPrompt: null,
       launchConfig,
       ...appliedSessionOptionProps(baseCommand.appliedSessionOptions),
-      ...(queryPlan.env ? { env: queryPlan.env } : {}),
-      // Hermes reads its prompt from the env, where a line break is harmless.
-      ...launchFileProps(args.launchFile, shell)
+      ...(queryPlan.env ? { env: queryPlan.env } : {})
     }
   }
 
@@ -195,11 +190,9 @@ function buildPlanWithPromptOnLine(args: AgentStartupPlanArgs): AgentStartupPlan
       agent,
       launchCommand: `${launchCommand}${grant} --prompt-interactive ${quotedPrompt}`,
       expectedProcess: config.expectedProcess,
-      followupPrompt: null,
       launchConfig,
       ...appliedSessionOptionProps(baseCommand.appliedSessionOptions),
-      ...(args.agentEnv ? { env: { ...args.agentEnv } } : {}),
-      ...fileProps
+      ...(args.agentEnv ? { env: { ...args.agentEnv } } : {})
     }
   }
 
@@ -208,29 +201,20 @@ function buildPlanWithPromptOnLine(args: AgentStartupPlanArgs): AgentStartupPlan
       agent,
       launchCommand: `${launchCommand}${grant} -i ${quotedPrompt}`,
       expectedProcess: config.expectedProcess,
-      followupPrompt: null,
       launchConfig,
       ...appliedSessionOptionProps(baseCommand.appliedSessionOptions),
-      ...(args.agentEnv ? { env: { ...args.agentEnv } } : {}),
-      ...fileProps
+      ...(args.agentEnv ? { env: { ...args.agentEnv } } : {})
     }
   }
 
-  return {
-    agent,
-    launchCommand,
-    expectedProcess: config.expectedProcess,
-    followupPrompt: trimmedPrompt,
-    launchConfig,
-    ...appliedSessionOptionProps(baseCommand.appliedSessionOptions),
-    ...(args.agentEnv ? { env: { ...args.agentEnv } } : {})
-  }
+  // `stdin-after-start`: no line carries its prompt; `carryLaunchPrompt` pastes it instead.
+  return null
 }
 
 /**
  * Whether this agent's prompt rides the launch command rather than the live PTY.
  *
- * The same question `buildAgentStartupPlan` answers by returning `followupPrompt: null`, asked
+ * The same question `planLaunchPrompt` answers with any outcome but `paste-after-ready`, asked
  * before a command exists — a caller deciding how to deliver a prompt has to know which half it is
  * getting while it is still choosing what to create. Derived from the one injection table rather
  * than restating it, and pinned against the builder for every agent by
diff --git a/src/shared/typed-startup-line.ts b/src/shared/typed-startup-line.ts
index 409b6bc03a3..4829179e303 100644
--- a/src/shared/typed-startup-line.ts
+++ b/src/shared/typed-startup-line.ts
@@ -2,9 +2,6 @@
  *  and 1 KiB did not. */
 export const TYPED_STARTUP_LINE_BUDGET_BYTES = 512
 
-/** cmd.exe's documented line cap, the smallest of the Windows shells Orca types into. */
-export const WINDOWS_TYPED_STARTUP_LINE_MAX_CHARS = 8191
-
 const encoder = new TextEncoder()
 
 /** Any C0 byte or DEL: no quoter escapes them, so a line editor reads them as keys. */
@@ -23,8 +20,3 @@ export function hasControlByte(line: string): boolean {
 export function typedStartupLineFits(line: string): boolean {
   return !hasControlByte(line) && encoder.encode(line).byteLength <= TYPED_STARTUP_LINE_BUDGET_BYTES
 }
-
-/** The same question for a Windows host, which cannot stage: only its own line cap applies. */
-export function windowsTypedStartupLineFits(line: string): boolean {
-  return !hasControlByte(line) && line.length <= WINDOWS_TYPED_STARTUP_LINE_MAX_CHARS
-}
diff --git a/src/shared/windows-shell-prompt-damage.test.ts b/src/shared/windows-shell-prompt-damage.test.ts
index 2ae35845503..579f2639981 100644
--- a/src/shared/windows-shell-prompt-damage.test.ts
+++ b/src/shared/windows-shell-prompt-damage.test.ts
@@ -1,11 +1,12 @@
 import { describe, expect, it } from 'vitest'
+import { planLaunchForTest } from './launch-prompt-plan.test-fixture'
 import { buildLaunchFilePointer, isLaunchFilePointer } from './launch-prompt-file'
-import { buildAgentDraftLaunchPlan, buildAgentStartupPlan } from './tui-agent-startup'
+import { buildAgentDraftLaunchPlan } from './tui-agent-startup'
 import type { AgentStartupShell } from './tui-agent-startup-shell'
-import { windowsDraftRefusal } from './startup-plan-launch-file'
+import { windowsDraftRefusal } from './launch-prompt-carry'
 
 function plan(prompt: string, shell: AgentStartupShell) {
-  return buildAgentStartupPlan({
+  return planLaunchForTest({
     agent: 'claude',
     prompt,
     cmdOverrides: {},
@@ -76,17 +77,17 @@ describe('a prompt a Windows shell would damage on the launch line', () => {
   })
 
   it('launches clean and leaves the prompt for the paste when the host writes no launch file', () => {
-    const startup = buildAgentStartupPlan({
+    const startup = planLaunchForTest({
       agent: 'claude',
       prompt: 'fix the build\nthen run the tests',
       cmdOverrides: {},
       platform: 'win32',
       shell: 'powershell',
-      hostWritesLaunchFile: false
+      launchHostIsPaired: true
     })
     expect(startup?.launchFile).toBeUndefined()
     expect(startup?.launchCommand).toBe('claude')
-    expect(startup?.followupPrompt).toBe('fix the build\nthen run the tests')
+    expect(startup?.pasteAfterReady).toBe('fix the build\nthen run the tests')
   })
 
   it('tells its own pointer apart from a prompt that merely mentions one', () => {
diff --git a/src/shared/wsl-launch-directory.test.ts b/src/shared/wsl-launch-directory.test.ts
index c14dee379a2..4b4c20687db 100644
--- a/src/shared/wsl-launch-directory.test.ts
+++ b/src/shared/wsl-launch-directory.test.ts
@@ -3,7 +3,7 @@ import { tmpdir } from 'node:os'
 import { join } from 'node:path'
 import { afterEach, beforeEach, describe, expect, it } from 'vitest'
 import { LaunchFileUnavailableError, writeSpawnLaunchFile } from './launch-file-writing'
-import { buildLaunchFilePointer, planLaunchPrompt, type LaunchFile } from './launch-prompt-file'
+import { buildLaunchFilePointer, carryInLaunchFile, type LaunchFile } from './launch-prompt-file'
 import { stageStartupCommand } from './startup-command-staging'
 import { parseWslLaunchDirectory, type WslLaunchDirectory } from './wsl-launch-directory'
 
@@ -21,7 +21,7 @@ afterEach(() => {
 })
 
 function launchFile(content = 'fix the build'): LaunchFile {
-  const planned = planLaunchPrompt(content, { sensitive: true })
+  const planned = carryInLaunchFile(content, true)
   return { ...planned.launchFile!, quoting: 'posix' }
 }
 
diff --git a/src/shared/zcode-orchestration-contract.test.ts b/src/shared/zcode-orchestration-contract.test.ts
index ba25cbceaef..039dd522f28 100644
--- a/src/shared/zcode-orchestration-contract.test.ts
+++ b/src/shared/zcode-orchestration-contract.test.ts
@@ -1,6 +1,7 @@
 import { describe, expect, it } from 'vitest'
+import { planLaunchForTest } from './launch-prompt-plan.test-fixture'
 import { getAgentSessionOptionCatalog } from './agent-session-option-catalog'
-import { buildAgentStartupPlan, agentPromptRidesLaunchCommand } from './tui-agent-startup'
+import { agentPromptRidesLaunchCommand } from './tui-agent-startup'
 import { TUI_AGENT_CONFIG } from './tui-agent-config'
 import { YOLO_TUI_AGENT_ARGS } from './tui-agent-permissions'
 import { recognizeAgentProcess } from './agent-process-recognition'
@@ -32,7 +33,7 @@ describe('ZCode orchestration and send-to-agent contract', () => {
   })
 
   it('keeps the prompt out of the launch command and hands it back as a follow-up', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'zcode',
       prompt: 'refactor the parser',
       cmdOverrides: {},
@@ -40,11 +41,11 @@ describe('ZCode orchestration and send-to-agent contract', () => {
     })
     expect(plan?.launchCommand).toBe('zcode')
     expect(plan?.launchCommand).not.toContain('refactor the parser')
-    expect(plan?.followupPrompt).toBe('refactor the parser')
+    expect(plan?.pasteAfterReady).toBe('refactor the parser')
   })
 
   it('carries a mode chosen through agent args onto the launch command', () => {
-    const plan = buildAgentStartupPlan({
+    const plan = planLaunchForTest({
       agent: 'zcode',
       prompt: '',
       cmdOverrides: {},