Problem

During the v0.8.43 prep run, the Agents/Tasks/Transcript UI exposed internal plumbing instead of a coherent agent team:

• raw child ids like agent_58ca1c8b and temporary names like harvest-1911
• agent_eval shown as if it were another worker instead of a parent action
• prompt sizes such as <912 chars> shown without the actual user-facing assignment
• transcript rows like delegate done plus truncated JSON that make the loop hard to follow

This makes CodeWhale feel powerful but not understandable. A user should be able to glance at the Agents panel and know who is doing what.

Scope

For v0.8.44, keep this focused on naming and display semantics, not deeper orchestration changes.

• Generate human-readable child labels from the assignment/role, with a stable short id only as secondary detail.
• Use a small work-role vocabulary for common child sessions: scout/recon, builder, reviewer, verifier, release-prep, docs, triage.
• Render agent_eval in the main UI as a parent action: review result, wait for agent, or fetch transcript.
• Keep raw tool names and ids available in expanded details/logs for debugging.
• Avoid whimsical names that hide the actual assignment; the user-facing label must still say what the child is doing.

Acceptance criteria

• Agents panel does not primarily display opaque names like agent_<hash>.
• Recent/Live tools does not make agent_eval look like a separate worker.
• Transcript live cards summarize delegated work in human terms.
• Existing logs/debug views still expose raw ids/tool names for support.
• Add focused snapshot or rendering tests for the naming/rendering path.

Related tracker: #1878
Related architecture issue: #1896