[katulevskiy]

Mirror of zeronsh#647, opened on this fork to run CI. Not for merging.

Problem

The Explorer's Subagents list mixes everything in one column. Once a few subagents have finished they crowd out the ones still running, and nothing anywhere else in the app says a chat has subagents running: the sidebar shows the same working spinner whether the parent is thinking or ten subagents are.

Change

Subagents section

• Running subagents stay on top as plain rows (the order from Explorer: list running subagents first, longest-running on top zeronsh/zeron#638).
• Everything finished folds into a Finished (N) dropdown, closed by default. Inside are Completed (n) and Failed (n) lists, each collapsible and each paging at ten rows with "Show more". An empty list is not drawn.
• Rows still open the subagent exactly as before. Nothing is hidden or removed: the list is read from the transcript's spawn chips, so there is no discard or hide.

Running pill (● N)

• Sidebar chat rows, beside the row's own activity indicator rather than replacing it: in compact rows it takes the time stamp's place while subagents run, and in regular rows it leads the status label. It also shows when the parent has settled but subagents are still running in the background.
• On the titlebar's explorer button, so running subagents can be found while the panel is closed: the button's face breathes in the activity colour and a small count badge sits on its corner. Both are drawn over the button, so the titlebar layout does not move.
• Beside the Subagents title, which stays visible when the section is collapsed.

To badge chats that are not open, the session row now carries running_subagents: the engine publishes it as subagent sinks open and settle, and clears it when the run ends. It is #[serde(default)] on the wire, written in both the workspace and registry docs, and read through view::running_subagents, which counts a row that has not been refreshed within SESSION_STALE_MS as zero, like the working indicator. The UI folds the staleness-checked count into its session presence key so a stale count clears without a new frame.

Screenshots

Rendered with GPUI from a fake transcript (three running, two completed, one failed). Left: Finished closed, the default. Right: Finished opened.

The explorer button with the panel closed (its face pulses):

Sidebar, compact (left) and regular (right): the first two rows have 3 and 2 subagents running, the fourth has 12 while its parent is idle.

Counts of one, two and three characters ("99+" is the cap). The badge and the sidebar pill draw their digits in Geist Mono, whose metrics centre the digits on the line (checked from the font tables). Both are painted rather than laid out, so the digits are placed exactly: in the pill the dot is centred on the digits' middle from the same snapped baseline (0.0 device pixels apart on a 1.875x capture), and the badge's digits are centred on its real middle. The padding after the "+" of a capped count is trimmed by a pixel, since its thin arms read as extra space.

Every single digit in the badge, captured at 1.875x. Measured against the badge on 1.25x and 1.875x captures, each digit is centred to within half a device pixel, and exactly for most.

The images live on a separate pr-assets-subagent-groups branch of my fork, so they are not part of this diff.

Tests

• files::subagents: grouping, default open state, slot counts as groups open and close, paging.
• view::running_subagents: counts while fresh, zero once stale or absent.
• workspace doc round trip carries the count.
• e2e::session_row_counts_running_subagents_until_the_run_ends: a scripted run with two subagents reports 2, then 1 as one settles, then 0 when the run ends.
• cargo test -p zeron-ui --lib passes (1361). Run in parallel it occasionally fails one of the text-selection tests (a different one each time; they share a global selection); it passes with --test-threads=1.

Notes

• Collapsing or opening a group changes the section's height without animating it.

---

[contribution-manager-katulevskiy]

Contribution Manager

needs triage · medium risk · untriaged

Open review workspace · Revision fdb0d691

Next actions:

• Direction decision needed
• Scope, risk, and platform triage needed
• Author has marked this PR as draft
• Independent code review needed
• Required check: core-tests (missing)

This summary is maintained by the review app. Decisions and test evidence are recorded in the workspace; the app does not merge PRs.