[Feature]: Inspect sub-agent prompts and live activity from the Agents panel

[Luzivog]

Problem

When using sub-agents in T3 Code, I can see token usage and recent commands, but I want to understand the work each agent was actually assigned and follow its progress. A token count and the latest command do not tell me whether the orchestrator gave it the right task, whether it is staying on scope, or what it has found so far.

Requested behavior

Let me open a sub-agent from the Agents panel and inspect:

• The exact task prompt the orchestrator gave it at spawn time.
• Any subsequent instructions or follow-up messages from the orchestrator.
• A live, scrollable transcript of the agent's visible messages, tool calls, and results, where exposed by the provider.
• Its current status and final response to the orchestrator, with the history still available after it finishes.

Keep this detail in an expandable view or separate panel so it is easy to return to the parent conversation.

Example

The orchestrator delegates an investigation to one agent and implementation to another. I want to open either agent, read its assignment, and see what it is checking or changing without asking the parent to relay a progress report.

Smallest useful scope

A read-only detail view showing the original delegation prompt and available activity history would already be valuable. Direct messaging or controlling sub-agents can be separate work. Where a provider does not expose a transcript, make that limitation clear.

Related discussions

• [Feature]: visual interface for subagents #6858 requests a visual interface for sub-agents.
• Subagent support as nested threads #6689 discusses sub-agents as nested threads.
• [Bug]: Spawned subagent rows aren't clickable — can't drill into a subagent's work #8137 is a closed issue about opening sub-agent details.

This request focuses on inspecting the actual assignment and subsequent work, beyond roster visibility, token counts, and a recent-command summary.

[juliusmarminge]

Closing as a duplicate of #6689 — that thread is the consolidation point for opening an individual subagent's work/conversation (see also #6858). Please add any extra detail there.