Problem
Each agent rendered as a fixed 3-line block; a 12-agent team produced ~47 lines of mostly-truncated text, making it hard to scan which agent is active and what each is configured to do — especially when several sub-agents run in parallel.
Proposed solution
- Render the current agent as a focus card in place (wrapped description, full model, thinking gauge + value).
- Render the other agents as compact two-line rows (name + right-aligned effort gauge, then the full model id), separated by a blank line.
- A width-aware degradation ladder and left-truncated model tails.
- Harness-backed agents (e.g. claude-code) show their harness type instead of a blank model.
- Rework the click-zone mapping to explicit per-line ownership so clicks keep resolving to the right agent.
Scope
pkg/tui/components/sidebar plus a small harness-type accessor on the agent/team.
Problem
Each agent rendered as a fixed 3-line block; a 12-agent team produced ~47 lines of mostly-truncated text, making it hard to scan which agent is active and what each is configured to do — especially when several sub-agents run in parallel.
Proposed solution
Scope
pkg/tui/components/sidebarplus a small harness-type accessor on the agent/team.