Agent visualiser: unclear parent-child relationships and opaque sub-agent activity

Summary

Two related UX gaps in the agent visualiser were observed when dispatching multiple sub-agents in parallel. Neither problem is a crash — both are gaps in the information displayed while agents are running.

---

Problem 1: No parent-child linkage

Description:
When sub-agents are spawned in parallel, the visualiser lists them at a flat level with no indication of which parent agent created them. If multiple agents are running simultaneously (or if a parent agent is also running alongside its children), there is no way to tell the hierarchy.

Expected behaviour:
A visual grouping, indented tree, or other clear indicator connecting each sub-agent to the parent agent that spawned it. For example, an indented list or a labelled section like "Sub-agents of [parent task]".

---

Problem 2: Opaque sub-agent activity display

Description:
While a sub-agent is running, the visualiser only shows the tool name and a 0 counter (e.g., Glob 0, Grep 0). There is no description of what the agent is doing or what parameters it was invoked with.

Expected behaviour:
Show at minimum:

• The agent's description string (passed as the description parameter when the Agent tool is called), so the user knows the intent of the agent.
• Optionally, the current tool call's meaningful label (e.g., Glob: ~/Documents/Research/** rather than Glob 0).

---

Steps to reproduce

1. In a Claude Code session, dispatch 2 or more background agents in a single message (e.g., three parallel Explore agents).
2. Open the agent visualiser while the agents are running.
3. Observe that:

• All agents appear at the same visual level, with no indication of parent-child relationships.
• Each sub-agent shows only a tool name and a 0 counter, with no description or meaningful activity label.

---

Environment

• Claude Code version: (run claude --version to check)
• OS: macOS (Darwin 25.3.0)

---

Impact

These gaps make it difficult to understand what is happening during multi-agent runs, reducing confidence in the system and making it harder to debug or reason about agent behaviour while tasks are in flight.