feat: add terminal broadcast, context menu, and shell close

Complete multi-pane broadcast, standard terminal right-click actions, batch host selection with progress, and native ssh_close_shell so closing a pane tears down the remote shell.

Co-authored-by: Cursor <cursoragent@cursor.com>

test/utils/agentHarness.test.ts

@@ -1,6 +1,9 @@
 import { describe, expect, it } from 'vitest'
 import { analyzeAgentOutput } from '../../src/utils/agentOutput'
-import { assessAgentCommandRisk } from '../../src/utils/agentCommandRisk'
+import {
+  agentCommandNeedsConfirmation,
+  assessAgentCommandRisk,
+} from '../../src/utils/agentCommandRisk'
 
 describe('Agent command guardrail', () => {
   it.each([
@@ -17,6 +20,15 @@ describe('Agent command guardrail', () => {
   ])('classifies %s as %s', (command, expected) => {
     expect(assessAgentCommandRisk(command)).toBe(expected)
   })
+
+  it('uses strict, balanced and permissive confirmation thresholds', () => {
+    expect(agentCommandNeedsConfirmation('safe', 'strict')).toBe(false)
+    expect(agentCommandNeedsConfirmation('medium', 'strict')).toBe(true)
+    expect(agentCommandNeedsConfirmation('medium', 'balanced')).toBe(false)
+    expect(agentCommandNeedsConfirmation('high', 'balanced')).toBe(true)
+    expect(agentCommandNeedsConfirmation('high', 'permissive')).toBe(false)
+    expect(agentCommandNeedsConfirmation('critical', 'permissive')).toBe(false)
+  })
 })
 
 describe('Agent analyze_output tool', () => {