From 39a8098b2ef5a0bd0b11c367e8ccdd9043ef4b6c Mon Sep 17 00:00:00 2001
From: Brennan Benson <79079362+brennanb2025@users.noreply.github.com>
Date: Tue, 4 Aug 2026 02:45:48 -0700
Subject: [PATCH] fix(mobile): stop a stale session-start report reverting a
 model pick
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Re-entering a chat tab re-delivers the same `agentStatus.model`, and the
reported-model effect re-applied it unconditionally — so picking a model, moving
to another tab, and coming back reverted the pill to the model the agent reported
at session start, which cannot have observed the `/model` sent after it. The
status stream reconnecting had the same effect.

A report is now only treated as evidence when the matched catalog id CHANGES for
that scope; a genuinely new report still supersedes a local pick. Mobile has no
screen read to confirm a switch against, so the repeat is all we can key off.
---
 ...mobile-native-chat-session-options.test.ts | 24 +++++++++++++++++++
 .../use-mobile-native-chat-session-options.ts | 16 +++++++++++++
 2 files changed, 40 insertions(+)

diff --git a/mobile/src/session/use-mobile-native-chat-session-options.test.ts b/mobile/src/session/use-mobile-native-chat-session-options.test.ts
index 8eff217445f..f8e811ca1e6 100644
--- a/mobile/src/session/use-mobile-native-chat-session-options.test.ts
+++ b/mobile/src/session/use-mobile-native-chat-session-options.test.ts
@@ -135,6 +135,30 @@ describe('useMobileNativeChatSessionOptions', () => {
     expect(effort!.kind).toMatchObject({ currentValue: 'low' })
   })
 
+  it('does not revive a stale session-start report over a newer local pick', async () => {
+    mount({ reportedModel: 'claude-sonnet-5' })
+    await act(async () => {
+      await api!.setOption('model', 'opus')
+    })
+    expect(api!.snapshot[0]!.kind).toMatchObject({ currentValue: 'opus' })
+    // Leaving the tab and returning re-delivers the SAME session-start report,
+    // which cannot have observed the `/model opus` sent after it.
+    update({ scopeKey: 'host\0worktree\0other' })
+    update({ scopeKey: 'host\0worktree\0tab' })
+    expect(api!.snapshot[0]).toMatchObject({ valueSource: 'dispatched' })
+    expect(api!.snapshot[0]!.kind).toMatchObject({ currentValue: 'opus' })
+  })
+
+  it('still lets a genuinely new report supersede a local pick', async () => {
+    mount({ reportedModel: 'claude-sonnet-5' })
+    await act(async () => {
+      await api!.setOption('model', 'opus')
+    })
+    update({ reportedModel: 'claude-haiku-4-5' })
+    expect(api!.snapshot[0]).toMatchObject({ valueSource: 'reported' })
+    expect(api!.snapshot[0]!.kind).toMatchObject({ currentValue: 'haiku' })
+  })
+
   it('keeps the live tab’s tracked model when other tabs overflow the record cap', async () => {
     mount()
     await act(async () => {
diff --git a/mobile/src/session/use-mobile-native-chat-session-options.ts b/mobile/src/session/use-mobile-native-chat-session-options.ts
index e91956565fe..7bfcd0467b0 100644
--- a/mobile/src/session/use-mobile-native-chat-session-options.ts
+++ b/mobile/src/session/use-mobile-native-chat-session-options.ts
@@ -39,11 +39,17 @@ type PendingOperation = { id: string; token: number }
 // scope cache. Bounded so long sessions across many tabs can't grow unbounded.
 const MOBILE_SESSION_OPTION_RECORD_CAP = 32
 const recordsByScope = new Map<string, NativeChatSessionOptionRecord>()
+// The catalog model id last taken from a hook report, per scope. Mobile cannot
+// read the agent's screen, so a repeat of the same report is not new evidence.
+const appliedReportByScope = new Map<string, string>()
 
 function getScopedRecord(scopeKey: string, agent: string): NativeChatSessionOptionRecord {
   const existing = recordsByScope.get(scopeKey)
   const record =
     existing && existing.agent === agent ? existing : createNativeChatSessionOptionRecord(agent)
+  if (record !== existing) {
+    appliedReportByScope.delete(scopeKey)
+  }
   // Why: delete-then-set on every read makes the touched scope most-recent, so
   // eviction only sheds the oldest UNTOUCHED tab. Insertion order alone would let
   // a long-lived active tab be the oldest key and lose its tracked model.
@@ -55,12 +61,14 @@ function getScopedRecord(scopeKey: string, agent: string): NativeChatSessionOpti
       break
     }
     recordsByScope.delete(oldest)
+    appliedReportByScope.delete(oldest)
   }
   return record
 }
 
 export function clearMobileSessionOptionRecordsForTests(): void {
   recordsByScope.clear()
+  appliedReportByScope.clear()
 }
 
 const EMPTY_SNAPSHOT: SessionOptionDescriptor[] = []
@@ -109,6 +117,14 @@ export function useMobileNativeChatSessionOptions(args: {
     if (!matched) {
       return
     }
+    // Why: the same report is re-delivered whenever the tab is re-entered or the
+    // status stream reconnects, and a session-start report cannot have observed a
+    // `/model` sent after it. Re-applying it would revert the user's pick. Only a
+    // report that CHANGES is evidence; the value itself still wins when it does.
+    if (appliedReportByScope.get(scopeKey) === matched) {
+      return
+    }
+    appliedReportByScope.set(scopeKey, matched)
     const record = getScopedRecord(scopeKey, agent)
     if (applyNativeChatReportedSessionOptions(record, { model: matched })) {
       bump()