[Bug]: Notification spam on IOS

[SturmOlling]

Before submitting

• I searched existing issues and did not find a duplicate.
• I included enough detail to reproduce or investigate the problem.

Area

apps/mobile

Steps to reproduce

I am getting ALL the notifications. 94 and counting from a single build command. I asked T3 to build using Astra light. it ran into usage limit and had asked a question. I shut down my computer. When I opened T3 code again my phone started chirping every other second ans stopped at 100 notifications.

Expected behavior

A single notification only

Actual behavior

I am getting ALL the notifications. 94 and counting from a single build command. I asked T3 to build using Astra light. it ran into usage limit and had asked a question. I shut down my computer. When I opened T3 code again my phone started chirping every other second ans stopped at 100 notifications.

Impact

Cosmetic issue

Version or commit

No response

Environment

IOS : 26.7 T3 0.0.40 (09e8de9)

Logs or stack traces

Screenshots, recordings, or supporting files

No response

Workaround

No response

[juliusmarminge]

Thanks for the report — this looks like a real iOS notification flood on desktop reconnect, not just one leftover “please answer this” alert.

What we think happened

You ran one Astra Light build, it hit a usage limit and asked a question, then the machine went away. Reopening T3 Code 0.0.40 (09e8de9c655a) republished agent-awareness state to the relay, and the phone treated that catch-up as new alerts (~94–100, about one every other second).

That matches the current code path:

• On server start, AgentAwarenessRelay republishes every awareness-active thread. Dedup lives only in memory, so a restart looks like a first publish. Ready/idle sessions project as completed, so catch-up is not limited to the thread that asked the question.
• Those publishes are not marked replay. Live Activity registration replay is already silent; environment startup catch-up is not.
• If Live Activities are off (or no card is armed), the relay alert path uses the current phase only. waiting_for_input has no 2-minute freshness window, so the usage-limit question can ring again after reconnect.
• Completed/failed alerts do have a 2-minute guard, but it keys off thread.updatedAt. Auto-settlement still stamps that to “now”, so old completions can look fresh. Astra Light child threads from one build can also show up as many separate “Done” rows.

Related, not a duplicate

• [Bug]: Auto-settlement marks old threads fresh and sends completion pushes on first launch #9057 (closed) was a first-launch after update auto-settle of an old backlog (~50+ iOS completion pushes). Same delivery hole, different trigger.
• fix(server): preserve automatic settlement timestamps #9254 is already in 0.0.40 — it fixed Settled-shelf settledAt timestamps, not this alert path.
• Open fix(server): stop stale settlement notifications #9081 would stop settlement from refreshing terminal awareness time. It does not mark startup catch-up as replay and does not add input-phase freshness, so it should not close this issue by itself.
• [Bug]: Subagent delegation incorrectly triggers completion notifications and unread status #5518 is false completion alerts during live subagent delegation, not a reconnect dump.

Next step

Keep this open for a server/relay fix (a mobile-only update will not be enough):

1. Treat environment startup catch-up as silent replay.
2. Land or rebase fix(server): stop stale settlement notifications #9081 so settlement cannot re-alert finished threads.
3. Do not alert on thread.settled republishes.
4. Apply previous-baseline / freshness rules to waiting_for_input on the notification-only path.

If you still have the banners, these details would help confirm the count (not required to keep this actionable):

• Were they all the same thread title, or many different ones?
• Roughly how long was the computer off?
• Were Live Activities enabled on the phone?

Temporary mute: disable agent notifications / T3 Connect publish on the linked device until reconnect settles.

[juliusmarminge]

Fixed by #13340 (merged to main: stop replaying historical terminal thread states as new Done alerts on server restart).