All four packages move from preact 10.29.8 to 11.2.0. The new types need four
small changes in swarm-ui and the shared dropdown: a ref that may be empty is
typed RefObject<T | null>, HTMLAttributes is imported from preact instead of
read from the JSX namespace, and TextField's type is the five kinds it is
used with, asserted once for the input's new union. npm-deps-hash follows the
lockfile.
durationLabel moves to util.ts, termPaths to agent/terminal/term-paths.ts,
StateBadges and tokenTotal to agent/state, and TryOutResult to
pages/review-common, so the builds, runs and agent pages no longer import
across page folders. SectionHeading is the body-size section heading the
create form and the egress card each styled; Preformatted is the scrolling
text block of the builds tail and the AGENTS.md view, which now share one
chrome and are both keyboard-scrollable. The components page shows both.
useSubmit holds the idle, sending, done and error state and the request
flow that six forms each wrote out; FormResult shows the success line and
the problem panel; FormStack is the column the link, claude-login,
inference and limits forms each styled by hand. The create-agent, link
account, backend-env and agent-input forms use the submit flow; the link,
claude-login, inference and limits forms use FormStack.
Hint, ActionRow, Toolbar, EmptyState and ExternalLink replace the muted
text, button row, refresh bar, empty line and _blank anchor that each page
restyled under its own class. ExpandableCard (with a loading twin) holds the
builds and recent-runs cards, which had identical head and body CSS. The
create form's raw checkbox becomes CheckboxField. The components page shows
each new piece.
BadgeSkeleton is the twin of Badge: the real Badge frame, as tall as an
interactive one (the settings values are buttons), with a placeholder for its
value. Permissions, resource caps, internet and allowed domains use it in place of
their "loading…" text, the details popover shows lines of text, and the
AGENTS.md editor a block, in place of theirs. SkeletonLines is a few stacked
lines for such blocks.
The globe of the egress card and of the allowed-domains box showed as a
plain monochrome glyph where the bee and the bell were emoji: a font with a
text-style glyph for the globe won over the emoji font. The two globes now carry
the emoji variation selector, and a panel's icon asks for emoji presentation
(font-variant-emoji), so another icon cannot fall the same way.
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.
The roster loses its second column and its table view. Selecting a card opens
its details under it as four folded sections: overview, settings, accounts and
the terminal. The status, message and wanted rows, which the card header
already shows, are gone from the details. The terminal page's details dialog
keeps the flat list.
The small preview in the detail panel keeps only the connection badge and
expand, and no longer follows the agent's state stream; the full terminal
tab keeps all its badges. The table gets a state row with the turn state, a
paused badge, and the context and billed tokens (not the model, which the
inference row shows). The preview's top padding drops to one badge row, and
the showHeaderBadges prop goes since the full-tab mount is now the only one
with badges.
The detail panel had editors for inference, instructions, resource caps and
accounts, but nothing for the permissions the swarm grants, which could only
be set at create time or with swarmctl. A permissions row shows the grant
(the default set, none, or the list) and an edit dialog ticks the catalogue
from GET /api/permissions and saves with PUT /api/agents/{agent}/permissions.
Ticking exactly the default set sends the DELETE, as the create form sends no
list for it.
When the server answers an event stream with an error status the browser
gives up and the badge stayed on disconnected with nothing to press. The
terminal, its state header and the messages page now offer a reconnect,
and the one rule for the state after an error moved into a tested helper.
The resource caps panel gets a retry beside its error, and its loading
state announces itself.
Inference, resource caps and linked accounts showed a failed read as a
badge with the reason in a hover title, which keyboard and touch users
cannot read. They now use ApiErrorPanel, and a list that already loaded
stays on screen under a failed refetch.
The success line of both link forms read the live text field, so editing
the name after a link changed the sentence. The result now carries the
name that was sent.
The matrix form held the typed token and password after a failed submit,
and kept them across a switch of credential mode. It now drops them when
the request ends and when the mode changes.
The linked-accounts list and the delete dialog said forgejo where the
link menu and forms say forge; they now say forge.
The instructions row of the agent detail panel gets an edit button that
opens the agent's AGENTS.md, as the harness last published it, in a text
box. Save sends a unified diff against the revision that was loaded; when
the agent changed the file meanwhile the editor keeps the operator's text,
shows the agent's version and offers to save over it or take it. An unsaved
edit survives closing the dialog.
The diff is built in the browser (LCS over lines, three lines of context,
git's no-newline marker) and was checked against git apply on 600 random
before/after pairs.
create_agent puts back the caps and inference settings it replaced if a
later store or the job insert fails, so a refused creation leaves nothing
behind for the name. The shared field components take an id prefix so the
create dialog and an open editor never share a DOM id. The queue check now
follows the hive roster check, the 422 is documented, and the copy says
empty rather than closed and that destroy keeps stored settings.
POST /api/agents takes optional limits and inference objects, checked by
the same rules as the two PUT routes before anything is queued and stored
ahead of the job graph. The create form gets collapsed sections for both,
reusing the fields of the detail-panel editors.
docs/swarm/inference.md had a table-row suffix pasted before the H1 by a
bad substitution, and the PUT row had lost it; the text is back in the
row.
A failed read of the agent-inference key already applied no cap (an
unreachable swarm must not block subagent work) and warned, but the
warning did not say which agent. It now names the agent and the error,
a test pins that, and subagent.md says a failed read starts anyway.
The max-subagents field stays a text field with a digit pattern: the
review-settings page's numeric field (concurrent reviews) is a TextField
with pattern="[0-9]+" and validation, and no field in swarm-ui uses
type="number". The new field gets the same pattern.
Nothing capped how many subagents an agent runs at once; the operator
wants to set that per agent from swarm-ui (#5234, shape (b)).
AgentInference gains `max_subagents: Option<u32>`. An absent field
decodes as None, so every stored value still parses and means "no cap",
which is today's behaviour.
swarm-controller accepts 1..=32 on PUT. 0 would refuse every start; 32
is far more concurrent claude sessions than one agent container has
memory for, so anything above it is a typo, not a cap.
The subagent daemon already read the agent's agent-inference key for
subagent_models; it now reads it on every start and continue (not only
those naming a model) and refuses a new run once the cap's worth of
names is tracked, counted under the same lock as the name reservation.
Every tracked slot counts: running, starting, a goal run between turns,
and an interrupted one still exiting. A goal re-prompt never goes
through the reservation, so it is never refused. A failed read applies
no cap, as it falls back for models.
swarmctl's `inference set` does GET, merge, PUT; its mirror type now
carries the field so a set from the CLI keeps a cap set in swarm-ui
instead of dropping it, and `get` prints it.
The hive reads the caps about every 30 seconds and a running agent takes a
changed cap at its next spawn or rebuild, so the dialog and the ui doc no
longer say the hive applies a change within seconds.
A resource caps row in the agent details reads GET /api/agents/{name}/limits
and shows the cpu quota and memory max, each reading as the hive deciding
when the swarm sets none. An edit dialog PUTs the filled fields, which
replaces the stored value whole, and a clear action DELETEs it. A refusal,
such as a percentage with a leading zero, shows as the route's reason under
the fields and leaves them as typed.
The details open under the agent badge without dimming or blocking the page; a click outside or Escape closes it. New ui/popover primitive (anchored, portal, no backdrop) with a sample on the components page. A confirm or link dialog opened from inside it stays modal, and the popover ignores clicks and Escape that belong to that dialog. AgentDetailDialog becomes AgentDetailPopover.
pages/agents held the roster page, the terminal page and every component under them. It keeps the two pages; the components move to src/agent, by concern: roster (card, wanted menu, actions), detail (panel body, dialog, inference, linked accounts, link menu) and terminal (preview, input, subagents, stream hooks, favicon), with the shared row type and icon at the top. Moves and import rewrites only.
