Prompt Opinion A2A times out after receiving and rendering valid completed task

## Summary

Prompt Opinion successfully calls our external A2A agent, receives HTTP 200, and visibly renders the returned completed task payload in the
browser transcript.

However, after the `SendA2AMessage` tool call succeeds, the Prompt Opinion chat stream still emits:

> The LLM took too long to respond and the operation was cancelled

So the remaining issue appears to be Prompt Opinion post-tool LLM synthesis after a successful external A2A response, not public routing or
our A2A runtime response.

## External A2A setup

Agent name:

```text
external A2A orchestrator

Prompt Opinion calls this endpoint shape:

POST /message:send/v1/message:send

No authentication headers or secrets are required for this test endpoint.

## Request body Prompt Opinion sends

Prompt Opinion sends a compact A2A message body shaped like this:

{
"message": {
"role": "ROLE_USER",
"parts": [
{
"text": "is this patient safe to discharge today?"
}
],
"messageId": "..."
}
}

## Runtime response summary

Our runtime returns:

{
"http_status": 200,
"response_bytes": 5356,
"content_type": "application/json; charset=utf-8",
"has_jsonrpc_2_0": true,
"has_result_task": true,
"has_top_level_task_alias": true,
"task_state": "TASK_STATE_COMPLETED",
"has_status_message_parts_text": true,
"has_artifact_text": true,
"includes_final_verdict_not_ready": true,
"includes_required_evidence_anchors": true,
"both_downstream_services_hit": true
}

## Response shape

The response is a JSON-RPC 2.0 envelope with result.task and a top-level task alias:

{
"jsonrpc": "2.0",
"id": "...",
"result": {
"task": {
"id": "...",
"contextId": "...",
"status": {
"state": "TASK_STATE_COMPLETED",
"message": {
"role": "ROLE_AGENT",
"parts": [
{
"text": "Care Transitions Command result:\nFinal verdict: not_ready.\nStructured baseline: ready.\nHidden-risk result:
hidden_risk_present.\n\nWhy the answer changed:\nThe structured chart looked discharge-ready at rest, but narrative evidence shows
exertional oxygen desaturation and unsafe home setup tonight.\n\nEvidence:\n- Nursing Note 2026-04-18 20:40: SpO2 dropped to 82% after
walking/stairs with dyspnea.\n- Case Management Addendum 2026-04-18 20:55: home oxygen delivery delayed until tomorrow; daughter cannot stay
overnight.\n\nImmediate blockers:\n- clinical_stability\n- equipment_and_transport\n- home_support_and_services\n\nRequired before
discharge:\nHold discharge today; reassess exertional oxygen needs; confirm oxygen delivery; confirm overnight support/transport plan;
update clinician handoff."
}
]
}
},
"artifacts": [
{
"name": "Care Transitions Command fused response",
"parts": [
{
"text": "Care Transitions Command result:\nFinal verdict: not_ready.\nStructured baseline: ready.\nHidden-risk result:
hidden_risk_present.\n\nWhy the answer changed:\nThe structured chart looked discharge-ready at rest, but narrative evidence shows
exertional oxygen desaturation and unsafe home setup tonight.\n\nEvidence:\n- Nursing Note 2026-04-18 20:40: SpO2 dropped to 82% after
walking/stairs with dyspnea.\n- Case Management Addendum 2026-04-18 20:55: home oxygen delivery delayed until tomorrow; daughter cannot stay
overnight.\n\nImmediate blockers:\n- clinical_stability\n- equipment_and_transport\n- home_support_and_services\n\nRequired before
discharge:\nHold discharge today; reassess exertional oxygen needs; confirm oxygen delivery; confirm overnight support/transport plan;
update clinician handoff."
}
]
}
],
"metadata": {
"final_verdict": "not_ready",
"hidden_risk_result": "hidden_risk_present",
"narrative_source_count": 3,
"both_mcps_hit": true
}
}
},
"task": {
"...": "same task object as result.task"
}
}

## What renders in the browser

Prompt Opinion visibly renders the returned task content in the transcript. The visible browser transcript includes both STATUS_MESSAGE and
ARTIFACT_MESSAGES, for example:

STATUS_MESSAGE: Care Transitions Command result:
Final verdict: not_ready.
Structured baseline: ready.
Hidden-risk result: hidden_risk_present.

Why the answer changed:
The structured chart looked discharge-ready at rest, but narrative evidence shows exertional oxygen desaturation and unsafe home setup
tonight.

Evidence:
Nursing Note 2026-04-18 20:40: SpO2 dropped to 82% after walking/stairs with dyspnea.
Case Management Addendum 2026-04-18 20:55: home oxygen delivery delayed until tomorrow; daughter cannot stay overnight.

So Prompt Opinion is receiving and displaying the completed task payload.

## Remaining error

After that successful external A2A result, the Prompt Opinion chat stream still emits:

The LLM took too long to respond and the operation was cancelled

This appears to happen after the external A2A runtime has already returned HTTP 200 with the completed task.

## What seems ruled out

This does not appear to be:

- Public endpoint routing failure
- Runtime POST failure
- Non-200 A2A response
- Missing result.task
- Missing completed task state
- Missing visible task text
- Oversized payload
- Missing evidence text
- Downstream service failure

The response is compact, about 5.3 KB, and renders visibly in the transcript.

## Question

Is there a different response shape or convention Prompt Opinion expects from SendA2AMessage so that the chat model stops cleanly after the
external A2A task completes?

In particular:

1. Should the final answer be in result.task.status.message.parts[0].text, result.task.artifacts[0].parts[0].text, or somewhere else?
2. Should the response include a separate result.message in addition to result.task?
3. Is the top-level task alias acceptable, ignored, or potentially harmful?
4. Should HTTP+JSON message endpoints return application/json or application/a2a+json?
5. Is there a known Prompt Opinion post-tool synthesis timeout after successful SendA2AMessage responses?
6. Is there a way for an external A2A agent response to tell Prompt Opinion “this task is complete; render this text as the final answer
without further LLM synthesis”?

## Expected behavior

Once Prompt Opinion receives the completed A2A task with visible text and TASK_STATE_COMPLETED, it should render the returned text and
complete the chat turn without a timeout/cancelled error.

## Actual behavior

Prompt Opinion renders the returned task text, but the chat turn still ends with:

The LLM took too long to respond and the operation was cancelled