Skip to content

button/tap can silently no-op — exits 0 and --list still says running #136

Closed

Description

@R44VC0RP

ran into this while having an agent drive a sim headless (lock screen captures for live activity stuff).

at some point the helper's HID injection just died. serve-sim button lock, tap, gesture — all of them kept exiting 0 but nothing was landing on the device. --list was still reporting running: true the whole time. so from a script/agent's perspective everything looks healthy while literally no input works.

for a human this is a minor annoyance because you'd notice the screen not changing. for an agent it's brutal — it trusts the exit code, so it just keeps going and burns a ton of time debugging the wrong layer (we restarted springboard, tried notifyutil, eventually did a full sim reboot before figuring out it was the helper).

couldn't pin down the exact trigger. it survived normal use for hours and then died after a stretch of heavy lock/unlock cycling + a simctl shutdown/boot of the device. restarting the helper (--kill + fresh start) didn't always bring input back either, which was the confusing part.

what would fix the pain isn't even preventing the breakage, it's making it loud:

  • non-zero exit when the HID send doesn't actually land (or at least when the injector knows its connection is gone)
  • some health signal in --list beyond running true/false, like hid: ok/stale

happy to provide more details, we hit this on macOS 26 (tahoe) with xcode 26 sims, serve-sim 0.1.45.

Activity

  1. EvanBacon commented on Sep 27, 2026

    @EvanBacon
    Owner

    Triage (2026-09-27 maintenance): accepted as a real agent-pain bug — HID inject can die while CLI still exits 0 and --list stays running: true.

    Next steps (no Mac load this pass)

    1. Design fields for loud failure are already sketched in draft docs: observability design (error repro + perf timing) #162 (hid.health + input.inject, non-zero exit when hid.state=gone / no_client).
    2. Overlaps the fire-and-forget gap called out on Touch input has no visible effect on a headless iOS 27 simulator (Xcode 27) #153; not claiming the same root cause as headless digitizer no-op.
    3. Implementation blocked until docs: observability design (error repro + perf timing) #162 design review lands (or Evan green-lights a minimal --list hid.state slice without full telemetry).

    Leaving open. No cloud agent / Darwin repro this cycle.

  2. added a commit that references this issue on Nov 10, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions