fix(mothership): accurate stale-observation wording and lease scope notes

- A stale browser observation says it was not run this time, which stays
  true when an earlier delivery already ran it.
- The lifetimes module says what a terminal call leaves running, and that
  sign-out ends leased tools only.
- Tests check responses and recorded outcomes instead of mock calls.

Author: waleedlatif1

apps/sim/app/api/copilot/confirm/route.test.ts

@@ -332,10 +332,10 @@ describe('Copilot Confirm API Route', () => {
     )
 
     expect(response.status).toBe(409)
+    expect(await response.json()).toEqual({
+      error: 'The desktop app holds this tool call; only its own result settles it',
+    })
     expect(recordedOutcome()).toBe(CopilotConfirmOutcome.HeldByDesktop)
-    expect(completeAsyncToolCall).not.toHaveBeenCalled()
-    expect(completeClaimedAsyncToolCall).not.toHaveBeenCalled()
-    expect(publishToolConfirmation).not.toHaveBeenCalled()
   })
 
   it('does not publish when another terminal transition wins indeterminate claim reconciliation', async () => {

apps/sim/app/workspace/[workspaceId]/home/hooks/desktop-tool-lifetimes.ts

@@ -8,8 +8,10 @@ interface RunningTurnTools {
  * Turns with a desktop tool still running in this tab, keyed by the turn's stream id. A tool
  * outlives the chat view that started it (and any stream reader), so the turn, not the view, owns
  * its Stop. A turn is held only while one of its tools runs: each tool releases it as it settles.
- * Terminal commands take no lease: Stop ends them through the turn's resource activity, which
- * clears the agent's commands in each scope it touched.
+ * Terminal calls take no lease. A terminal call returns once its operation does (a `run` after
+ * its wait window), and the command it started keeps running in a terminal tab the user can see
+ * and control. Stop settles the agent's marks on that tab through the turn's resource activity;
+ * the process itself is the user's to end.
  */
 const runningTurns = new Map<string, RunningTurnTools>()
 
@@ -52,7 +54,10 @@ export function stopDesktopTools(streamId: string, reason: string): void {
   runningTurns.delete(streamId)
 }
 
-/** Cancels every running desktop tool in this tab, so none outlives the session that started it. */
+/**
+ * Cancels every leased desktop tool running in this tab (browser actions, local file reads and
+ * imports), so none outlives the session that started it.
+ */
 export function stopAllDesktopTools(reason: string): void {
   for (const turn of runningTurns.values()) turn.stop.abort(reason)
   runningTurns.clear()

apps/sim/lib/mothership/tools/client/browser-tool-execution.test.ts

@@ -1470,7 +1470,7 @@ describe('pre-dispatch drops still resolve the waiter', () => {
     expect(mockReportCompletion).toHaveBeenCalledWith(
       'stale-call-1',
       'error',
-      expect.stringContaining('never started'),
+      expect.stringContaining('not run this time'),
       expect.objectContaining({ staleEvent: true })
     )
     const [, , message] = mockReportCompletion.mock.calls[0] ?? []

apps/sim/lib/mothership/tools/client/browser-tool-execution.ts

@@ -109,8 +109,8 @@ const OUTCOME_UNKNOWN_MESSAGE =
   'The Sim window closed while this browser action was in flight. It may already have taken effect. Do not retry it automatically; take a fresh browser snapshot before deciding what to do.'
 const REPLAY_OUTCOME_UNKNOWN_MESSAGE =
   'This browser action was recorded before the Sim page reloaded, but its terminal result could not be recovered. It may already have taken effect. Do not retry it automatically; take a fresh browser snapshot before deciding what to do.'
-const STALE_OBSERVATION_NOT_STARTED_MESSAGE =
-  'Not run: this browser observation never started, because it reached the Sim desktop app too late to run safely. Nothing happened in the browser. Do not retry it in this turn; tell the user to keep this chat open in the Sim desktop app, or to ask again later.'
+const STALE_OBSERVATION_NOT_RUN_MESSAGE =
+  'Not run: this browser observation reached the Sim desktop app too late to run safely, so it was not run this time and has no result. An observation changes nothing in the browser. Do not retry it in this turn; tell the user to keep this chat open in the Sim desktop app, or to ask again later.'
 const STALE_STATEFUL_OUTCOME_UNKNOWN_MESSAGE =
   'This browser action was delivered too late to recover its exact result. It may already have taken effect. Do not retry it automatically; take a fresh browser snapshot before deciding what to do.'
 const REPLAY_GUARD_CAPACITY_MESSAGE =
@@ -595,7 +595,7 @@ export async function executeBrowserToolOnClient(
     logger.info('Skipping stale browser tool event', { toolCallId, toolName, age })
     const observationOnly = OBSERVATION_ONLY_BROWSER_TOOLS[toolName]
     const message = observationOnly
-      ? STALE_OBSERVATION_NOT_STARTED_MESSAGE
+      ? STALE_OBSERVATION_NOT_RUN_MESSAGE
       : STALE_STATEFUL_OUTCOME_UNKNOWN_MESSAGE
     retainAndReportTerminalCompletion(
       toolCallId,

apps/sim/lib/mothership/tools/client/completion.test.ts

@@ -121,7 +121,6 @@ describe('client tool completion reporting', () => {
     await expect(
       reportClientToolCompletionOnPageExit('tool-1', 'error', 'Browser failed')
     ).resolves.toBeUndefined()
-    expect(fetchMock).toHaveBeenCalledOnce()
   })
 
   it('rejects a non-success response', async () => {