Label Agent tool calls with the dispatched agent's name

Scenario

When a session dispatches multiple agents in parallel (review swarms, research fan-out, or skill-orchestrated sub-tasks), every tool-call row in the transcript starts with the same literal Agent(...) prefix. Concrete failure modes:

• Telling which agent is which means reading the parenthesised description on every row, both while watching a session run and when auditing a long one afterwards. A code-quality review and a general-purpose researcher look identical until you do.
• Correlating transcript rows with --output-format stream-json events requires matching on description text, because the TUI doesn't expose the subagent_type that the JSON emits.

Proposal

In the interactive TUI transcript, render Agent tool calls using the subagent_type argument that is already present on every call:

code-quality(Review GDScript diff)
Explore(Find hyprlock submit path)
general-purpose(Check Linear for stale tickets)

(Built-in agent casing as currently rendered; the field is subagent_type on the Agent tool input.)

Scope: interactive terminal transcript only. No schema change required, since --output-format json / stream-json already carries subagent_type in the tool input.

Related

• #39147
• open, asks for subagent_type on the Claude App remote-control view. Same data, different surface.
• #43553
• stream-json subagent events.
• #14859
• agent hierarchy in hook events.
• #46253
• subagent observability gap.
• #16529
• closed via autoclose; about user-prompt attribution from inside an agent.
• #26151
• closed stale; hook-based rendering override, an alternative the maintainer might prefer.