test(orchestration): cover agentCmdOverrides for a startupAgent launch

Anyone who worked around this bug by pointing agentCmdOverrides.cursor at their
own cursor-agent path must keep that override once the id resolves properly.
Verified it survives: `cursor-agent --beta '--force'`.

Also trims the conflict-guard comment to the repo's brevity rule; no logic
change.

src/main/runtime/orca-runtime.test.ts

@@ -12656,6 +12656,33 @@ describe('OrcaRuntimeService', () => {
     }
   )
 
+  // Why: a user who worked around this bug by pointing the override at their own
+  // cursor-agent path must keep that override once the id resolves properly.
+  it('honors an agentCmdOverrides entry for a startupAgent', async () => {
+    const spawn = vi.fn().mockResolvedValue({ id: 'pty-bg' })
+    const runtime = new OrcaRuntimeService({
+      ...store,
+      getSettings: () => ({
+        ...store.getSettings(),
+        disabledTuiAgents: [],
+        agentCmdOverrides: { cursor: 'cursor-agent --beta' },
+        agentDefaultArgs: { cursor: '--force' },
+        agentDefaultEnv: {}
+      })
+    })
+    runtime.setPtyController({
+      spawn,
+      write: () => true,
+      kill: () => true,
+      getForegroundProcess: async () => null
+    })
+
+    await runtime.createTerminal(`path:${TEST_WORKTREE_PATH}`, { startupAgent: 'cursor' })
+
+    const spawnCall = spawn.mock.calls[0]?.[0] as { command?: string } | undefined
+    expect(spawnCall?.command).toBe("cursor-agent --beta '--force'")
+  })
+
   // Why: folder workspaces have no repo, so command sniffing skipped them entirely
   // and spawned the bare string; an explicit agent must still resolve.
   it('resolves a startupAgent in a repo-less folder workspace', async () => {

src/main/runtime/orca-runtime.ts

@@ -23884,13 +23884,10 @@ export class OrcaRuntimeService {
       opts.claudeAgentTeamsSourceCommand
     const store = this.store
     if (opts.startupAgent) {
-      // Why: returning opts unresolved here would spawn a bare shell that can only
-      // time out waiting for an agent — the silent failure startupAgent exists to
-      // prevent. An explicit agent needs no command sniffing and no repo, so the
-      // only unresolvable cases are a contradictory caller or a dead runtime.
-      // `command` and `resumeProviderSession` count as contradictions too: the
-      // first would be silently overwritten, the second would pair resume identity
-      // with a freshly built launch.
+      // Why: falling through unresolved would spawn a bare shell that can only time
+      // out waiting for an agent. A caller-supplied launch contradicts the agent:
+      // `command` would be overwritten, `resumeProviderSession` would pair resume
+      // identity with a fresh launch.
       if (callerSuppliedLaunch || opts.command || opts.resumeProviderSession) {
         throw new Error(
           `startupAgent ${opts.startupAgent} cannot combine with a caller-supplied launch.`