The checkbox list was capped at 12em and scrolled, and the box shrinks the
list to its widest label, so the scrollbar sat in the middle of the popover
with the lower permissions hidden. The permissions are a short fixed list:
MultiselectFilterOptions takes scrollable={false} and the permissions box
uses it. Filter lists that can grow long keep the scroll area.
A new Skeleton (a shimmering shape, still where motion is off) and a region
that tells assistive tech once that something is loading. The agent card's
header is split into AgentCardLayout, a layout with a slot for each part: the
card fills the slots with the real icon, badges and controls, its new
AgentCardSkeleton twin fills the same slots with placeholder shapes, so the
card that replaces it lands in the same place. The agents page uses the twin in
place of its "loading…" text.
usePolled takes a cache key: it saves each answer in localStorage and starts
from the saved one, so opening the agents, alerts, builds, review runs or hives
view shows the last data at once and the refresh spinner marks the update. A
saved answer is dropped after a day, when it was saved by another build of the
UI, at log out (a new onLogout of the shared user menu) and when the gateway
ends the login (a new onSessionEnded of the session guard). Pages with an
editor (schedules, notification flows, broadcast) do not use it.
usePolled reports when a request is in flight (through a new onBusy of the
poller) and holds it on for at least 0.4 seconds, so a quick answer still
reads as a refresh. The picker swaps its clock for the spinner, in the same
box so its width stays, and the spinner stands still where motion is off.
Every page with a picker passes it; a page with several polled requests shows
it while any of them runs. The Jobs page fetches through shared components
and has none yet.
The two controls sat among the pane's header badges as a labelled badge and a
cross glyph. They are now icon badges in the top right corner, through a
`trailing` slot of the terminal preview, so the header on the left keeps the
agent, split, subagents and state badges and a narrow pane wraps less.
The menu on an agent's terminal page used to open the subagent on a tab of
its own. It now starts a group of this terminal and the subagent's, split to
the right; in a group's pane it splits that pane. A pane's layout key is the
agent, or `agent/subagent` for a subagent's pane, so the layout operations
stay as they are and a destroyed agent takes its subagent panes with it. A
subagent already in the group shows a marker and does nothing when chosen,
and maximize opens a subagent pane's terminal on its own.
tuwunel 1.9.4 exports no metrics. The host collector gets an httpcheck
receiver on the gateway's /_matrix/client/versions (new
services.hyperhive.otel.httpChecks, declared by hive-matrix), and a timer in
the hive-matrix container pushes the database directory's size through the
forwarder's OTLP port, as the attic cache-size unit does. The size step
dereferences the state directory's symlink and pushes nothing for a result
under 4096 bytes.
The builder pushes its build metrics (builds by kind and outcome, build
duration, queue depth and wait, reply publish failures) to its forwarder's
OTLP receiver, which relays them to the swarm collector.
It also registers the standard store client-certificate gauges
(bao.client_cert.*, cert=builder) on the same exporter. They are the
builder's liveness signal: exported every minute whether or not a build
runs. The cert-metrics table in docs/scheduler/observability.md gains the
builder row.
The per-agent /stats page read these from its sqlite; the swarm board had
none of them. hive-agent now counts the tools a turn called, the commands
bash tasks ran (from the bash daemon's own table, newer rows only),
reminders scheduled and delivered, and records the first turn of a fresh
session; the turns tab shows them, with fresh sessions per agent. Claude's
own export has no tool-call counter, so the docs no longer say it does.
The per-agent /stats page shows what the swarm agents board lacked about
turns. The new tab reads the turn count and duration the harness already
records: average duration, cache hit rate, tokens per turn, turns per hour
by agent and by model, duration percentiles, and the wake source and result
mixes. The duration histogram used the SDK default buckets, which stop at
10 s while a turn averages 24 s, so every p95 read a flat 10 s; it now has
explicit bounds up to an hour.
