fix(terminal): key agent-paste bracketing on live evidence, not shellForeground

Measured against a real pane: shellForeground is republished only at OSC 133
boundaries, so a shell without 133 integration leaves it latched true while an
agent owns the foreground. Vetoing on it silently reinstated the submit bug the
parent commit fixes - the parked draft was sent on paste with the gate in place.

Prefer process-confirmed agent identity when present, keep the restart-rehydrated
veto, and drop the shellForeground veto. Erring toward bracketing costs a literal
ESC[200~ in a non-2004 program; erring the other way sends the user's draft.

Refs STA-4294

src/renderer/src/components/terminal-pane/terminal-agent-paste-bracketing.test.ts

@@ -76,16 +76,15 @@ describe('shouldForceBracketedMultilinePasteForPane', () => {
     ).toBe(false)
   })
 
-  // Staleness: the agent row is retained on purpose and is not lifecycle-driven, so it
-  // can outlive the agent. Only evidence the pane is no longer the agent's may veto.
-  it('vetoes when OSC 133;D proved the foreground is back at the shell', () => {
+  it('brackets on a process-confirmed agent even with no status row', () => {
     expect(
       decide({
+        agentStatusByPaneKey: {},
         paneForegroundAgentByPaneKey: {
-          [AGENT_PANE_KEY]: foregroundEntry({ shellForeground: true })
+          [AGENT_PANE_KEY]: foregroundEntry({ agent: 'codex', shellForeground: false })
         }
       })
-    ).toBe(false)
+    ).toBe(true)
   })
 
   it('vetoes a row rehydrated from disk across an app restart', () => {
@@ -111,11 +110,14 @@ describe('shouldForceBracketedMultilinePasteForPane', () => {
     ).toBe(true)
   })
 
-  it('still brackets when foreground evidence exists but has not seen a shell', () => {
+  it('still brackets when shellForeground is latched true while an agent owns the pane', () => {
+    // Why: shellForeground is republished only at OSC 133 boundaries, so a shell with
+    // no 133 integration leaves it true while an agent runs. Measured live: vetoing on
+    // it reinstated the submit bug.
     expect(
       decide({
         paneForegroundAgentByPaneKey: {
-          [AGENT_PANE_KEY]: foregroundEntry({ agent: 'codex', shellForeground: false })
+          [AGENT_PANE_KEY]: foregroundEntry({ shellForeground: true })
         }
       })
     ).toBe(true)

src/renderer/src/components/terminal-pane/terminal-agent-paste-bracketing.ts

@@ -8,11 +8,6 @@ import type { PaneForegroundAgentEntry } from '../../store/slices/pane-foregroun
  * ConPTY never forwards it and that ConPTY can be on a remote host — so the client's
  * platform proves nothing about the PTY. Unbracketed, xterm rewrites the paste's
  * newlines to CR and the agent submits whatever draft was parked in its composer.
- *
- * An agent row is retained on purpose (an idle agent sits at `done`), so neither the
- * 30-minute freshness TTL nor `state !== 'done'` can gate this — both would strip
- * bracketing from a live but quiet agent. Only evidence that the pane is *no longer*
- * the agent's vetoes it.
  */
 export function shouldForceBracketedMultilinePasteForPane({
   isWindowsClient,
@@ -31,10 +26,9 @@ export function shouldForceBracketedMultilinePasteForPane({
     return true
   }
   const paneKey = makePaneKey(tabId, leafId)
-  // Why: OSC 133;D proved the foreground is back at the shell — process-grade evidence
-  // that outranks an agent row nothing has cleared yet.
-  if (paneForegroundAgentByPaneKey[paneKey]?.shellForeground === true) {
-    return false
+  // Why: a process-table confirmed agent is the strongest evidence there is.
+  if (isTuiAgent(paneForegroundAgentByPaneKey[paneKey]?.agent)) {
+    return true
   }
   const entry = agentStatusByPaneKey[paneKey]
   // Why: a row rehydrated from disk across a restart describes the previous process,
@@ -42,5 +36,12 @@ export function shouldForceBracketedMultilinePasteForPane({
   if (entry?.restoredUnconfirmed === true) {
     return false
   }
+  // Why NOT vetoed on shellForeground: that flag is republished only at OSC 133
+  // boundaries, so a shell without 133 integration leaves it latched true while an
+  // agent owns the foreground. Vetoing on it silently reinstates the submit bug —
+  // measured live. An idle agent also sits at `done` past the 30-minute freshness
+  // TTL, so neither state nor TTL can gate this either. Erring toward bracketing
+  // costs a literal ESC[200~ in a non-2004 program; erring the other way sends the
+  // user's parked draft.
   return isTuiAgent(entry?.agentType)
 }