Resumed subagent's transcript/output is not shown in the Claude Desktop UI — only the raw resume JSON

What happened

When I spawn a subagent (via the Agent/Task tool), let it finish, and then send it another message to resume it (SendMessage), the Claude Desktop app does not display the resumed subagent's transcript or its returned output. The chat only shows the raw tool-result JSON from the resume call, e.g.:

{"success":true,"message":"Agent \"<agent-id>\" had no active task; resumed from transcript in the background with your message. You'll be notified when it finishes. Output: C:\Users\<user>\AppData\Local\Temp\claude\<project>\<session-id>\tasks\<agent-id>.output","resumedAgentId":"<agent-id>","pin":{"id":"<agent-id>","name":"<agent-id>","ref":"<ref>"}}

The completion notification later confirms the agent finished, but the actual work/transcript the resumed agent produced is only written to the .output file on disk — it's never surfaced inline in the desktop UI the way a freshly-spawned agent's transcript is.

Expected behavior

Resuming a completed subagent should show that agent's transcript/output in the UI (or at least an expandable view of the .output file contents), the same as when the agent is first spawned. Seeing only the resume-acknowledgement JSON gives no visibility into what the agent actually did.

Steps to reproduce

1. Spawn a subagent and let it run to completion.
2. After it finishes, send it a follow-up message to resume it (SendMessage to the agent id/name).
3. Observe: the chat shows only the {"success":true,"message":"...resumed from transcript in the background...","resumedAgentId":...} JSON blob.
4. When the resumed agent completes, its transcript/output is not rendered inline — only the completion notification appears; the details live in the on-disk .output file.

Environment

• Claude Desktop app (Windows 11)
• Claude Code within the desktop app; subagents spawned via the Agent/Task tool, resumed via SendMessage.

Impact

No visibility into a resumed subagent's actual output without manually opening the temp .output file, which breaks the normal transcript view and makes multi-turn subagent workflows hard to follow.

<img width="329" height="341" alt="Image" src="https://github.com/user-attachments/assets/0bb6bae8-ff77-45ef-8e56-75504500136e" />

https://github.com/MuhammadShaharyarNaeem/blackgoat-agentskills