Summary

• add a Sidebar info mode selector to /settings, with the existing compact roster as the default and a detailed per-agent view as an opt-in
• keep the compact mode identical to the current two-line agent roster
• retain the full six-cell effort gauge while explicitly labeling effort, context, and cost in detailed mode
• aggregate cumulative cost per agent without counting repeated session snapshots twice
• restore historical per-agent costs from persisted session trees
• add per-agent cost to the Agent Inspector and a By Agent section to /cost

Settings

/settings → Appearance now includes:

Sidebar info mode                         ‹ Compact ›

The selection previews immediately across tabs. Apply persists it as settings.layout.sidebar_info_mode: detailed; Cancel restores the previous mode. Compact is omitted from the config because it is the default.

Preview

The previews below use ANSI-stripped output.

Compact mode, default sidebar width

Agents ────────────────────────────────

▶ root                        ▰▰▰▰▱▱ ^1
  anthropic/claude-opus-4-8         30%

  helper                      ▱▱▱▱▱▱ ^2
  openai/gpt-5.4-mini               91%

  budget                      ◉ 8.2K ^3
  openai/gpt-4o

  plain                              ^4
  google/gemini-flash

Detailed mode, default sidebar width

Agents ────────────────────────────────

▶ root                               ^1
  anthropic/claude-opus-4-8
  Effort ▰▰▰▰▱▱ high
  Context 30% · Cost $0.13

  helper                             ^2
  openai/gpt-5.4-mini
  Effort ▱▱▱▱▱▱ off
  Context 91% · Cost $0.02

  budget                             ^3
  openai/gpt-4o
  Effort ◉ 8.2K · Context — · Cost —

  plain                              ^4
  google/gemini-flash
  Context — · Cost —

Detailed mode, minimum sidebar width

At the 20-column minimum width, the complete effort gauge remains visible and metrics flow onto separate lines.

Agents ────────────

▶ root           ^1
  …/claude-opus-4-8
  Effort ▰▰▰▰▱▱
  Ctx 30%
  Cost $0.13

  helper         ^2
  …nai/gpt-5.4-mini
  Effort ▱▱▱▱▱▱ off
  Ctx 91%
  Cost $0.02

  budget         ^3
  openai/gpt-4o
  Effort ◉ 8.2K
  Ctx — · Cost —

  plain          ^4
  …gle/gemini-flash
  Ctx — · Cost —

At wider sidebar sizes, all three detailed metrics share one line:

▶ root                                             ^1
  anthropic/claude-opus-4-8
  Effort ▰▰▰▰▱▱ high · Context 30% · Cost $0.13

Behavior

Validation

Proxy variables were unset for the full test and lint runs because the local proxy intercepted private-address SSRF tests and Go module downloads.

Known limitations

• live cost attribution after an in-session handoff follows the latest agent because cumulative session events do not expose an exact per-agent split
• historical per-agent costs cannot be reconstructed for remote sessions that do not persist per-message attribution