# Claude after runtime probe

Source e7417481e5 (before main merge); actual isolated web dev server port6361/14401, Linux fs-28, Claude Opus5.5. Verified persisted child messages via read-only .t3 database. All 3 runs completed. Subsequent main merge triggered additional empty parent wakes, which were cancelled by the server; the original probe and all child tasks were complete before that merge.

The new Claude Opus 5.5 reviewer kept the conversation across all three rounds. In round 2 it gave back the marker even though my prompt never included it, and in round 3 it correctly described round 2.

I found no conflict with higher-priority instructions. The orchestration rules say to use `t3_thread_send` on `childThreadId` for follow-ups that need the same child's conversation, which is what this probe does. I used `delegate_task` instead of the native Agent tool because `orchestrator_capabilities` reported inherited options that weren't empty (`fastMode: false`). The rules say native subagents can't reliably apply those options.

**Child:** All three rounds came back as `providerInstanceId: claudeAgent`, `model: claude-opus-5-5`, `options: [{fastMode: false}]`. This is a new child, not the earlier Codex one.

Shared childThreadId: [Sentence review memory probe (Claude)](t3-thread://v1/thread:delegated-task:command%3Amcp%3Aaa7ac19d-60af-4d42-b94d-36ffc09240e9%3Adelegate-task%3Aprobe19821-claude-same-conversation-r1-20261022)

```text
thread:delegated-task:command%3Amcp%3Aaa7ac19d-60af-4d42-b94d-36ffc09240e9%3Adelegate-task%3Aprobe19821-claude-same-conversation-r1-20261022
```

**Tools used:** `mcp__t3-code__orchestrator_capabilities`, `mcp__t3-code__delegate_task` (round 1, `mode: wait`), `mcp__t3-code__t3_thread_send` (rounds 2 and 3, `mode: queue`), `mcp__t3-code__t3_thread_wait`, and `mcp__t3-code__task_status`.

| Round | Tool | Child run | Exact response |
|---|---|---|---|
| 1 | `delegate_task` | ordinal 1 | `READY` |
| 2 | `t3_thread_send` | ordinal 2 | `PLUM19821CLAUDE` |
| 3 | `t3_thread_send` | ordinal 3 | `In round 2 you asked me to recall the marker from round 1 and reply with nothing but that marker. I replied with exactly: PLUM19821CLAUDE` |

Task IDs, in round order:

```text
node:delegated-task:command%3Amcp%3Aaa7ac19d-60af-4d42-b94d-36ffc09240e9%3Adelegate-task%3Aprobe19821-claude-same-conversation-r1-20261022
node:delegated-task:command%3Amcp%3Aaa7ac19d-60af-4d42-b94d-36ffc09240e9%3Athread-send%3Aprobe19821-claude-same-conversation-r2-20261022
node:delegated-task:command%3Amcp%3Aaa7ac19d-60af-4d42-b94d-36ffc09240e9%3Athread-send%3Aprobe19821-claude-same-conversation-r3-20261022
```

Every task finished with `status: completed` and `hasPendingChildRuns: false`. No files were read or edited, no forge was contacted, and no top-level threads were created.

Persisted synthetic-only messages:

```json
{
  "childThreadId": "thread:delegated-task:command%3Amcp%3Aaa7ac19d-60af-4d42-b94d-36ffc09240e9%3Adelegate-task%3Aprobe19821-claude-same-conversation-r1-20261022",
  "messages": [
    {
      "role": "user",
      "text": "Act as the review sub-agent for this task.\n\nConversation-continuity probe. Do not read or edit any files, run commands, or call any forge. You are a reviewer for the sentence 'The result is ready.' Remember this marker for later in this conversation: PLUM19821CLAUDE. For this round, respond with exactly: READY",
      "messageId": "message:delegated-task:command%3Amcp%3Aaa7ac19d-60af-4d42-b94d-36ffc09240e9%3Adelegate-task%3Aprobe19821-claude-same-conversation-r1-20261022",
      "runId": "run:thread:thread%3Adelegated-task%3Acommand%253Amcp%253Aaa7ac19d-60af-4d42-b94d-36ffc09240e9%253Adelegate-task%253Aprobe19821-claude-same-conversation-r1-20261022:ordinal:1"
    },
    {
      "role": "assistant",
      "text": "READY",
      "messageId": "message:provider:claudeAgent:native-item:c77f081f-cbfd-40d7-b46f-b907a2fdfda6",
      "runId": "run:thread:thread%3Adelegated-task%3Acommand%253Amcp%253Aaa7ac19d-60af-4d42-b94d-36ffc09240e9%253Adelegate-task%253Aprobe19821-claude-same-conversation-r1-20261022:ordinal:1"
    },
    {
      "role": "user",
      "text": "Round 2 of the same review conversation for 'The result is ready.' Still no file reads, commands, or forge calls. Recall the marker I asked you to remember in round 1 and reply with only that marker.",
      "messageId": "message:mcp:aa7ac19d-60af-4d42-b94d-36ffc09240e9:thread-send:probe19821-claude-same-conversation-r2-20261022",
      "runId": "run:thread:thread%3Adelegated-task%3Acommand%253Amcp%253Aaa7ac19d-60af-4d42-b94d-36ffc09240e9%253Adelegate-task%253Aprobe19821-claude-same-conversation-r1-20261022:ordinal:2"
    },
    {
      "role": "assistant",
      "text": "PLUM19821CLAUDE",
      "messageId": "message:provider:claudeAgent:native-item:4ed04dae-eb88-43d3-b0c1-406ae9de56c2",
      "runId": "run:thread:thread%3Adelegated-task%3Acommand%253Amcp%253Aaa7ac19d-60af-4d42-b94d-36ffc09240e9%253Adelegate-task%253Aprobe19821-claude-same-conversation-r1-20261022:ordinal:2"
    },
    {
      "role": "user",
      "text": "Round 3 of the same review conversation for 'The result is ready.' Still no file reads, commands, or forge calls. Briefly describe what happened in round 2: what I asked you and exactly what you replied.",
      "messageId": "message:mcp:aa7ac19d-60af-4d42-b94d-36ffc09240e9:thread-send:probe19821-claude-same-conversation-r3-20261022",
      "runId": "run:thread:thread%3Adelegated-task%3Acommand%253Amcp%253Aaa7ac19d-60af-4d42-b94d-36ffc09240e9%253Adelegate-task%253Aprobe19821-claude-same-conversation-r1-20261022:ordinal:3"
    },
    {
      "role": "assistant",
      "text": "In round 2 you asked me to recall the marker from round 1 and reply with nothing but that marker. I replied with exactly: PLUM19821CLAUDE",
      "messageId": "message:provider:claudeAgent:native-item:7c36c1a4-1f0b-4359-bef9-ad181c3c05b5",
      "runId": "run:thread:thread%3Adelegated-task%3Acommand%253Amcp%253Aaa7ac19d-60af-4d42-b94d-36ffc09240e9%253Adelegate-task%253Aprobe19821-claude-same-conversation-r1-20261022:ordinal:3"
    }
  ],
  "runs": [
    [
      1,
      "completed"
    ],
    [
      2,
      "completed"
    ],
    [
      3,
      "completed"
    ]
  ]
}
```