PR 30390 real development verification

Baseline: github/main 5bdc2afc23
PR head: ebf626ee89b7b220bfe1195fb3363d1d23819b0a
Linux VM; repository Playwright 1.60.0; /usr/bin/google-chrome.
One background dev server at a time, with isolated development homes.
Project added through Add project > Local folder, pointing at a real git repository.
Real Codex gpt-6.1-sol turns. No mocks, fixtures, or database record injection.

Setup commands:
  git worktree add ../main github/main
  (cd ../main && timeout 600 vp i)
  setsid nohup vp run dev > /tmp/main-dev.log 2>&1 &
  Stop main by captured process-group PID before starting PR.
  git checkout users/spoukyii/runtime-subagent-title-emoji
  timeout 600 vp i
  T3CODE_HOME=/home/user/t3code/.t3 setsid nohup vp run dev > /tmp/dev.log 2>&1 &
Ports from dev-runner: main web 8693/server 16733; PR web 5733/server 13773.

Provider setup:
Installed Codex 0.164.0 emits explicit agent-path titles and empty prompts.
Its real native subagent completed CHILD_OK; parent completed PARENT_OK.
To reach the prompt fallback, selected supported Codex 0.156.1 in Settings > Providers.
Added gpt-6.1-sol as a custom model there. Children inherited the parent model.
One baseline turn had a proxy 503. Retrying completed successfully.
Explicit child model selection on 0.156.1 was rejected; inherited gpt-6.1-sol worked.

Exact changed behavior:
Child prompt: 76 lowercase a characters + emoji U+1F600 + the CHILD_OK instruction.
The real native spawn_agent call produced this prompt and child.
Browser CDP debugger observed the existing function's local coordinator at return.
No function replacement, direct projection invocation, or synthetic records.
Stack: projectedSubagentsToRuntime -> ThreadRelationshipsPanel -> React.

Before on main, actual running client:
{
  "title": "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa\ud83d...",
  "prompt": "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa😀 Reply exactly CHILD_OK. Do not use tools or create subagents.",
  "explicitTitle": null,
  "model": "gpt-6.1-sol",
  "status": "completed",
  "result": "CHILD_OK",
  "childThreadId": "thread:provider:codex:native-thread:01a1a3c2-f298-7723-aace-0a4ed5163883",
  "wellFormed": false,
  "length": 80,
  "tailCodeUnits": [
    "61",
    "61",
    "61",
    "61",
    "d83d",
    "2e",
    "2e",
    "2e"
  ],
  "callers": [
    "projectedSubagentsToRuntime",
    "ThreadRelationshipsPanel",
    "react_stack_bottom_frame"
  ]
}

After on PR, same app-created record copied while both servers were stopped:
{
  "title": "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa...",
  "prompt": "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa😀 Reply exactly CHILD_OK. Do not use tools or create subagents.",
  "explicitTitle": null,
  "model": "gpt-6.1-sol",
  "status": "completed",
  "result": "CHILD_OK",
  "childThreadId": "thread:provider:codex:native-thread:01a1a3c2-f298-7723-aace-0a4ed5163883",
  "wellFormed": true,
  "length": 79,
  "tailCodeUnits": [
    "61",
    "61",
    "61",
    "61",
    "61",
    "2e",
    "2e",
    "2e"
  ],
  "callers": [
    "projectedSubagentsToRuntime",
    "ThreadRelationshipsPanel",
    "react_stack_bottom_frame"
  ]
}

Fresh provider turn on PR, three actual children:
[
  {
    "title": "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa...",
    "prompt": "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa😀 Reply exactly CHILD_OK. Do not use tools or create subagents.",
    "explicitTitle": null,
    "model": "gpt-6.1-sol",
    "status": "completed",
    "result": "CHILD_OK",
    "childThreadId": "thread:provider:codex:native-thread:01a1a3ca-a875-71a3-b7b8-ef08e8607c3c",
    "wellFormed": true,
    "length": 79,
    "tailCodeUnits": [
      "61",
      "61",
      "61",
      "61",
      "61",
      "2e",
      "2e",
      "2e"
    ],
    "callers": [
      "projectedSubagentsToRuntime",
      "ThreadRelationshipsPanel",
      "react_stack_bottom_frame"
    ]
  },
  {
    "title": "Short 😀 Reply exactly SHORT_OK. Do not use tools or create subagents.",
    "prompt": "Short 😀 Reply exactly SHORT_OK. Do not use tools or create subagents.",
    "explicitTitle": null,
    "model": "gpt-6.1-sol",
    "status": "completed",
    "result": "SHORT_OK",
    "childThreadId": "thread:provider:codex:native-thread:01a1a3ca-d319-79a2-bc28-70991ab17c0a",
    "wellFormed": true,
    "length": 70,
    "tailCodeUnits": [
      "62",
      "61",
      "67",
      "65",
      "6e",
      "74",
      "73",
      "2e"
    ],
    "callers": [
      "projectedSubagentsToRuntime",
      "ThreadRelationshipsPanel",
      "react_stack_bottom_frame"
    ]
  },
  {
    "title": "bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb...",
    "prompt": "bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb Reply exactly ASCII_OK. Do not use tools or create subagents.",
    "explicitTitle": null,
    "model": "gpt-6.1-sol",
    "status": "completed",
    "result": "ASCII_OK",
    "childThreadId": "thread:provider:codex:native-thread:01a1a3ca-e60e-7892-82e3-9c66c05d7410",
    "wellFormed": true,
    "length": 80,
    "tailCodeUnits": [
      "62",
      "62",
      "62",
      "62",
      "62",
      "2e",
      "2e",
      "2e"
    ],
    "callers": [
      "projectedSubagentsToRuntime",
      "ThreadRelationshipsPanel",
      "react_stack_bottom_frame"
    ]
  }
]

Parent returned PR_PARENT_OK.
Clicked Previous agents, opened boundary child, saw full emoji prompt and CHILD_OK.
Returned using Open parent; opened short emoji child, saw SHORT_OK; returned again.
Reloading preserved child records and successful results.

Checks through blacksmith-check:
vp test run packages/client-runtime/src/state/subagentRuntime.test.ts: 12 passed.
vp run check:changed aee76a8b473694878e820ed642b4b4018991f378: passed.
Format, lint, both knip passes, seven package typechecks, and 12 tests passed.
The shallow worker needed git fetch --depth=3 of pull/30390/head for comparison.
Hosted PR checks at inspection: 30 successful, 0 failing, 0 pending.

Limits:
Matching-data screenshots show the actual app; web clips the title before the boundary.
Live client debugger output establishes the Unicode difference, not the screenshots.
Native mobile Agents sheet, VoiceOver, Claude workflows, native desktop shells,
and remote/relay/tunnel connections were not exercised on this Linux VM.
Claude was unauthenticated. No macOS/iOS/Android device testing.
No source changes or pushes were necessary.