mirror of
https://gh-proxy.org/https://github.com/klinsmaya/multica.git
synced 2026-11-08 01:16:45 +08:00
main
2388
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
|
7c186016a5 |
feat(core): multica 1.0 part 1 (MUL-8317) (#9635)
* feat(core): add app space model for rail navigation Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(layout): replace left sidebar with app rail and space panels Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * test(e2e): drive navigation through the app rail Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(layout): finish the rail and panel shell against the mock - The sidebar toggle always works: in Inbox, Chat and Settings it collapses the page's own list column instead of being disabled. - Move create workspace from the rail into the account menu, and complete the menu: profile, theme and language submenus, shortcuts, docs, change log, Discord, desktop download, log out. - Add an explicit New issue button to the Issues and My issues headers. - Autopilot panel gets a create button and template shortcuts; Team panel gets an invite-member button for owners and admins. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(desktop): draw tabs as pills on the title bar The active tab no longer merges into the content card: with the space panel inside the card, the merged tab landed over whatever column happened to sit below it. Every tab is now a pill on the title bar; the active one uses the card's own surface, inactive ones a lighter tint. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(layout): give list columns the space panel's width Inbox and Chat defaulted to 260px with their own saved widths, Settings was a fixed 240px, and the space panel used the sidebar width. They now share one ListColumn sized by --sidebar-width with the panel's resize edge, so the divider stays in place across spaces and one drag resizes them all. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(layout): return each rail space to where you left it Rail items always linked to the space's home, so switching spaces dropped the project, squad, chat session or settings tab you had open. The shell now remembers the last page (path and query) per space and workspace, and the rail links there; the current space's item still leads home, so a second click resets it. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(layout): animate the list column when the sidebar toggles Inbox, Chat and Settings unmounted their list column on toggle, so it snapped away while the space panel slid. ListColumn now stays mounted and eases its width like the panel (200ms, off under reduced motion and while dragging), which also keeps the list's scroll position across a collapse. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(layout): one title band for every left column The space panels used PanelHeader (title-sm, no rule), Inbox and Chat used the page header (body size, bottom border) and Settings its own 64px title. PanelHeader now takes a heading level, a leading slot and a meta slot, and Inbox, Chat and Settings render their list title with it. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(layout): collapse the whole sidebar with one toggle across spaces Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(layout): switch workspaces from one rail tile Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(layout): preview Inbox, Chat and Settings lists on rail hover Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(inbox): keep the list column's own loading skeleton Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(desktop): lay out the shell for the page on screen during rail switches (MUL-8317) The tab session URL moves before the router commits the next page (it renders in a transition), so for a few frames the shell already showed the incoming space's layout around the outgoing page: switching into Inbox or Chat dropped the Work panel and the old page's header icon jumped to the top-left. The router now publishes its committed pathname from a layout effect and the shell's panel and list columns follow it; space memory keeps following the navigation URL. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(views): flatten the settings nav into one level of section headings (MUL-8317) Scope header rows, sub-group headings and dividers stacked three levels into the 240px list column. Every section is now a peer with the space panels' section label and spacing; each page already names its scope in a badge. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(views): expand projects inline in the work panel (MUL-8317) Project rows in the Issues panel get a chevron that lists the project's open issues in place, so the sidebar can move between issues without opening the project page. The issue on screen reveals its project and stays lit; expansion persists per workspace. A dedicated open-issues cache follows realtime title patches and refetches only when status or project membership moves. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(views): sidebar tree settings, load more, expandable views and project order (MUL-8317) - Expanded projects and views list issues from the server (updated date, finished statuses hidden) with Load more / Show less. - One shared issue display popover on hovered expanded projects; a project display popover on the Projects header (ordering, statuses, only mine, counts); drag to reorder; collapsible section; no project cap. - Issue rows shimmer and show agent avatars while an agent works or waits. - Settings, order and expansion persist per workspace. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(views): show sidebar status filters as checkboxes (MUL-8317) Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(ui): auto-collapse the nav only below lg (MUL-8317) The rail now stays on screen beside the space panel, so laptop-sized windows (1024-1279px, just under the 1280 desktop default) have room for the panel; they no longer open Issues with it collapsed. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(views): one Projects display popover, higher-contrast filter chips (MUL-8317) Issue settings are shared by every project, so they move from each expanded row into the Projects header popover beside the project settings. Status filters go back to chips: a solid outline and tint when on, dashed and faded when off. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(server): list views across scopes and store sidebar view order Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(views): starred views sidebar, unified view header and All views page - Sidebar Views lists only starred views (built-in and saved, user-ordered) with an All views row; order lives in a per-user sidebar preference. - Issues, My issues and project pages share one header: view name menu, folding filter chips (+N), assignee-type / relation chip, Reset and Save. - Saved-view filters are fully editable; new views are created directly and renamed inline instead of through the save dialog. - New All views directory with tabs, search and star toggles. - Project page drops the breadcrumb and hides the redundant project chip. - Remove the view bar, save/manage dialogs and the old headers. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(inbox): switch between inbox and archive with header tabs Replace the Archived row that trailed the notification list (and the back row inside the archive) with Inbox | Archived tabs pinned under the list header. The unread count moves onto the Inbox tab, and an empty archive no longer says the inbox is empty. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * Revert "feat(inbox): switch between inbox and archive with header tabs" This reverts commit e8a8e1bcea1bf5b45c4e5830d654523a0b77daff. Co-authored-by: multica-agent <[email protected]> * fix(inbox): drop the count beside the no-agents-working filter The count widened the whole filter menu for a toggle whose result is visible in the list as soon as it is applied. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(inbox): switch to the archive from the page title The Archived entry trailed the notification list and scrolled away with it. The list title is now a menu (Inbox / Archived), so the switch stays in reach without adding a row, and the archive's back row is gone. An empty archive no longer says the inbox is empty. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(sidebar): keep a pinned issue lit after its URL turns canonical The issue route replaces /issues/<uuid> with /issues/MUL-123, but a pin only matched its own UUID href, so clicking a pinned issue lit the same issue under its project instead. Issue pins now match the open issue by id, a lit pin no longer auto-expands its project, and pinned issues show the running agent the same way tree rows do. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(sidebar): line pinned agent avatars up with the tree rows The avatar sat inside the gap reserved for the unpin button, so pinned avatars landed left of the project tree's and the label lost room. A pinned issue with agents now puts the avatar in the row's right edge and hands that slot to the unpin button on hover. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(sidebar): raise the contrast of the working-agent shimmer ShimmerText now reads --shimmer-text-base, --shimmer-text-highlight and --shimmer-text-spread, defaulting to the previous values. Sidebar issue rows (tree and Pinned) rest on faint-foreground and use a 22% highlight band so the sweep is visible at sidebar size; chat, the issue page and other shimmer hosts are unchanged. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(issues): keep the view header from overlapping on narrow widths The view-name button kept the Button base shrink-0, so once the h1 hit its min width the name painted over the add-filter trigger (visible on a 390px web viewport). The button now shrinks and truncates, the add-filter label becomes screen-reader-only in a narrow header, and FoldingFilterChips re-measures its reserved width when its leading or trailing controls resize instead of only when the chips change. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(issues): one Filter dropdown and a header that adapts to its width The view header no longer lays filters out as chips with a +N overflow. A single Filter button carries the active count and opens the filter menu, which now lists the dimensions already filtering first, each showing its values, with the scope axis (assignee type, relation to me) at the top. The split is taken when the menu opens so a row never moves mid-edit. The header measures itself: when the title side runs out of room it first folds the display controls to icons, then the agents chip and New issue, before the title truncates, and only releases a step once the row is wider than the width that forced it. This fixes the overlap in saved views with table controls and after toggling agents working. FoldingFilterChips and single-row-fit had no other users and are removed. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(sidebar): keep the highlight on the row that was clicked An issue that is both pinned and listed in the tree always lit its pin, so clicking it under a project moved the highlight away from the row just clicked. The work panel now remembers which section a click on an issue link came from: a click in the tree keeps the tree row lit and the pin unlit, a click on the pin keeps the pin. Arriving any other way (search, a link, a shortcut) still lights the pin. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(sidebar): let Views and Pinned collapse like Projects Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(sidebar): keep a dragged row at its own height while reordering (MUL-8354) (#9662) Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(sidebar): create an issue in a project from its row Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(issues): put the workflow beside the title and set the panel toggle apart The project workflow now sits on the left with the title and Filter, in the same borderless style. New issue matches Display's height, and a divider separates the right-sidebar toggle from the header actions. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(issues): fold the layout switcher into the Display panel One Display button whose icon follows the layout. The panel opens with the layouts as a segmented row, then grouping, ordering and sub-issues, then the current layout's own options. The table's grouping and column pickers move in from the header. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(issues): keep hovered Display chips from looking selected Light mode's muted and accent are the same grey, so a hovered chip matched a pressed one. Hover now only darkens the text and border, and a pressed chip fills with the stronger input tone. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(sidebar): give the Projects settings the Display panel's controls Both panels now share one chip and one sort-direction button. The sidebar popover drops its dividers and dashed off-chips, takes the same section headings, and its selects match the Display panel's width. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(issues): quiet the issue view header down to its title The header carried six pieces of text, half of which said nothing: the project's workflow name, "0 agents working" and "Display". - Workflow moves into the project title menu as a submenu; the properties panel keeps its row. - Filter becomes an icon beside the title, followed by Reset / Save, so the count sits next to the actions it causes. Its tooltip names the active filter fields. - Display and New issue become ghost icon buttons named by tooltips, with New issue showing its C shortcut. - The agents chip shows only while someone is working or its filter is on, as avatars and a number. With every control already an icon, the crowding steps are gone. MUL-8361 Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(issues): is / is not per filter, Relation to me, and a hover summary (MUL-8353) (#9663) * feat(issues): is / is not per filter, a Relation to me filter, and a hover summary (MUL-8353) Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(issues): match the Relation filter's Chinese labels to My issues (MUL-8353) Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(views): advanced filter mode, menu search, and due date filter Advanced mode turns the simple filters into a rule tree (match all/any, is/is not per condition, one level of groups) edited in a panel behind the same Filter trigger. Switching back is offered when the tree is something simple mode can say. The filter menu gets a search box that matches fields and values; a value hit reads as the condition it adds. Due date joins as a dimension with relative buckets (overdue, today, this week, next 7 days, none). The issue table query accepts filters.advanced, filters.due_dates and filters.today, compiled to SQL with the same IS NOT TRUE negation. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * test(e2e): menu search, due date and advanced filter Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> Co-authored-by: Cursor <[email protected]> * test(e2e): let menus settle before filter screenshots Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(views): advanced filter keeps status columns, property filters and negated totals - Columns follow a status the rule tree requires, and a status it names un-hides its column, so switching to advanced no longer drops rows. - Status branch totals drop the excluded keys for "is not" instead of falling back to the loaded count. - The advanced panel shows property filters as editable rows with an "Add property filter" entry, and Clear all clears them too. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(views): make custom properties ordinary advanced filter conditions Properties are fields in the advanced tree: they sit in the field picker after the built-in fields, can live in groups and follow all/any, and edit their values with the property's own options. The server compiles a property:<id> condition with the existing property predicate. Entering advanced mode moves property filters into the tree, so the separate property section and its Add property filter button are gone. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(views): typed filter operators and direct condition groups - Add condition group inserts an empty group and opens its own add menu, instead of asking for a field first. The group's header carries its match ("Match any of these"); rows only show the joining word. - A row's menu can turn it into a group, and a tree with groups reads back as one sentence under the rows. - Operators follow the field's type: is / is not, includes any / none of for labels and multi-value properties, contains for text, comparisons for numbers, before / after for dates, and is empty / is not empty for anything that can be unset. Text, number and date values are typed in the row. Every operator is a reading of the stored negated + values, so saved views and the server query are unchanged. - Actor value headings sit in a menu group, which Base UI requires. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(layout): New issue and search at the far end of the desktop title bar New issue lived only as a pen in the Issues panel header, so Inbox, Team, Chat and Auto had no way to start an issue but the C shortcut. Search sat in the rail between the workspace tile and the spaces. - Desktop: New issue and search are icon buttons at the right end of the title bar, styled like Back/Forward, with search last. The tabs give way to them. The rail drops search. - Web has no title bar, so its rail keeps both under the workspace tile, New issue first. - The button does what C does: a project page files the issue into that project. Both now share one helper. - The Issues panel header loses its pen; the page header keeps New issue. MUL-8359 Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(issues): remove horizontal scrollbar from the filter menu Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(sidebar): every view expands in the sidebar Saved views preview through the table query, which expresses every view filter (is not, relation, due date, rule trees, my-any). Built-in views expand too, scoped like their page and listed with the sidebar issue settings. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(sidebar): group expanded rows by spacing, folder icon for projects Issues under an expanded row drop to 26px rows with no gap and a 12px status icon, an open group ends with 8px of space, and the expanded row's name goes full-strength. Sections sit 20px apart. Sidebar projects show a folder icon instead of their emoji. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(sidebar): expandable rows give their label the full width The hidden chevron, an empty spacer and their gaps reserved about 90px at the end of every expandable row, so names truncated with space to spare. The chevron now takes the icon's place on hover. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(sidebar): tree row chevron swaps back to the icon immediately Drop the opacity fades, and show the chevron for keyboard focus only, so a clicked row no longer keeps it until focus moves. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(issues): drop the hint beside Switch to advanced filter The grey text sat where the menu shows current values, so it read as a setting rather than an explanation. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(layout): tighten the gap under panel titles, keep selected rows at regular weight The first section kept the 20px that separates sections, so it sat far below the panel title. Selected and expanded rows, desktop tabs, the Settings nav, agent tabs and the skills file tree no longer switch to medium weight: the wider glyphs made the label grow when it was clicked. Selection reads from the background and text colour. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(sidebar): keep stars while loading, leave view state to in-place clicks, page past 100 - Star, unstar and reorder write the stored order, not the resolved items, and wait until that order is known, so views still loading are kept. - Sidebar rows switch the active view only when the click navigates this page; a new tab takes its target from the URL. - Sidebar issue lists walk 100-row pages so Load more keeps growing. - Drop a trailing blank line in advanced-filter-panel. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(views): open the clicked view in desktop tabs, prepare view state only on in-place navigation - Desktop tabs carry their own ?view=: list pages adopt it on mount and on back/forward, write view changes back with replace, and the active-view map is swapped with the active tab. A saved view is part of tab identity, so cmd+clicking view B from a tab on view A opens B in a new tab. - AppLink and useRowLink call onNavigateInPlace right before they push, so sidebar and All views rows prepare view state exactly when the page navigates in place (desktop shift-click included, web shift-click not). - Rail items and the workspace tile use named radius tokens. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(views): one page system for lists and objects Every page now follows the Issues page: one 48px header, nothing stacked under it. - Lists entered from the sidebar (agents, squads, skills, runtimes, projects, views, autopilots, issue wakeups) title themselves with an icon and a scope menu, and keep only icon tools: filter with its count and Reset, display, and "+". Counts left the header; list search is removed. - Objects (agent, squad, skill, runtime, autopilot, member) open under a breadcrumb with at most one solid action and a ⋯ menu. Their tabs became body sections with actions on the section titles, in a 760px column, with a 300px properties panel the header toggles. - Analytics keeps its tabs, now in the header; settings, chat and the create-agent flow share the same header. - Shared CollectionPage* and ObjectPage* primitives in views/layout. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(views): one CollectionList for every list page Replace the per-page tables and ListGrid with a shared CollectionList in packages/views/layout: single-line 36px rows with inline grey descriptions, a shared column/grid template, header sort, row selection with shift-range, keyboard navigation (arrows, J/K, Enter, X, Cmd+A, Esc), right-click and kebab row menus, load-more, and a shared batch bar. - createCollectionViewStore factory in core; display menus are generated from the column definitions. - Agents, Squads, Skills, Projects, Autopilots, Runtimes (page and RuntimeList), Views and Wakeups migrate to it; Wakeups pages with load-more instead of prev/next. - The Issues batch toolbar uses the shared CollectionBatchBar. - Agents hides owner and access by default so the default columns fit at 1280px; persisted v1 defaults migrate, custom choices are kept. - Remove ListGrid and the now-unused per-page locale keys. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(layout): smooth the nav panel toggle - Ease the panel and list columns with a soft-start curve (220ms ease) instead of ease-out, whose first frame jumped a quarter of the width and, when that frame dropped, nearly half of it. - End-align the panel content and clip it while it grows, so it slides in with the edge instead of spilling under the page; the resize handle moves to a zero-width sibling so the clip does not cut it. - The floating chat stays mounted while closed and re-rendered on every frame of the toggle as its container resized; it now re-renders only when the clamped window size changes. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(issues): give My agents' work its own URL My agents' work shared /my-issues with My issues and was told apart only by a persisted global scope, so it could not open in its own tab, and switching one tab flipped every other. It now lives at /my-issues?scope=agents: the page, header and sidebar read the scope from the URL, desktop tab dedupe keys on it, and the tab gets its own title and icon. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(views): preserve My agents' work in copied links Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
e8ae72c9df |
MUL-8418: fix(usage): correct Cursor variant and OpenAI GPT-5.6 pricing (#9739)
* MUL-8418: fix(usage): price Cursor -fast / effort variants consistently Add vendor-published Fast rates Cursor's sheet omits (Opus 5.5/5, GPT-5.5, GPT-5.3 Codex, GPT-5.2), price gpt-5.1 and Cursor's deepseek-v41 spelling, bill Gemini 3.5/3.6 Flash at Cursor's sheet, and route provider=cursor rows in the llm_cost_usd metric through a Cursor-aware resolver that mirrors the frontend so Fast variants no longer borrow the standard rate. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * MUL-8418: fix(usage): update GPT-5.6 rates and pre-5.6 OpenAI cache writes GPT-5.6 Sol / Terra / Luna move to OpenAI's current standard rates (4/20, 2/12, 0.20/1.20) on both the dashboard and the llm_cost_usd metric. Backend pre-5.6 OpenAI rows billed cache writes at the cache-read rate; those tokens are ordinary input, matching the frontend. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * MUL-8418: fix(usage): price GPT-5.6 Sol at list rate, not promo Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
ac131c55cb |
MUL-8416 fix(comments): keep thread owner when root folds into a queued task (#9735)
A member reply merged into a still-queued task moves trigger_comment_id to the reply and leaves the thread root only in coalesced_comment_ids. ListConversationRootOwners matched the root against trigger_comment_id alone, so the thread lost its owner and later un-mentioned replies under the root were silently dropped. Treat a root in coalesced_comment_ids as owned, the same coverage rule taskCoversReplyParent and CancelAgentTasksByTriggerComment already use. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
53c4bbe5cd |
MUL-8395 fix(lark): distinguish bound bots from verified message reception (#9709)
* fix(lark): verify message reception separately from binding * fix(lark): preserve verification evidence during rollout * fix(lark): remove stray space in Chinese verification hint Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Multica Eve <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
3191de1930 |
MUL-8385 feat(wecom): show the run's steps and reasoning in the bubble while it works (#9696)
* feat(wecom): show the run's steps and reasoning in the bubble while it works The bubble used to open, spin for the whole run and close with the answer. It now fills in while the run is going: each tool call as a sentence in the reader's language, the agent's own reasoning under it, and how long it has been. Only the bot's principal sees any of it, and only in their own 1:1 — a group or anyone else gets the plain spinner, decided again for every round. - Steps come from task:message (the transcript), plus the daemon's first task:progress milestone, which is said as "starting the agent" in the bubble's language rather than the daemon's English. Only runs asked over WeCom (a WeCom channel_task_delivery row) show steps. - Progress reaches the replica that holds the round, whichever replica served the daemon's request. The bus subscriber only classifies the event and queues it on a per-run lane, so the daemon's request does no database read and no socket write. The holder reads the run's new transcript rows past a per-run watermark (ListTaskProgressSince), redacts, throttles and writes the frame. A replica without the round publishes a progress relay frame that carries ids and a seq, no text; it takes no claim and never enters the dispatcher's queues, so it cannot delay or crowd out an answer. Duplicates, replays and out-of-order frames read nothing new, and do not count as a sign of life for the rotation guard. - Credentials are redacted from steps and reasoning: labelled values, known key prefixes, long high-entropy runs, and the same checks on each part of a URL or of a word glued to shell punctuation, plus webhook URLs whose path is the secret. Nothing the renderer strips afterwards can join two pieces into a key the redactor never saw. Reasoning arrives in 500ms increments, so redaction and the </think> defence run over the joined buffer, where a key and its value are seen together, and a run that may still be the start of a key is held back until it has finished. - A bubble near the server's ten-minute window is handed over to a fresh stream; a hand-over the server refuses stops rotation, and any stream the server has disowned takes no further frame — every closer falls back to a plain message instead of spending another refusal. Every frame a step writes, and both frames of a hand-over, take the reply write guard. - The identity lookup is wired on the real boot path, and its absence logs a warning, since without it no bubble ever shows a step. A relay with no typing indicator attached warns once when progress frames arrive. * fix(wecom): show a step the throttle held back without waiting for the next report A step that landed inside the 1.5s refresh window was only shown when the run reported again. The step most worth showing is often the last before a long silence (a file read, then a test suite that runs for minutes), so the bubble sat on the earlier step for the whole wait. A held step now schedules one trailing refresh per feed for when the window ends. It goes through the run's lane like any report, is not counted as a sign of life for the rotation guard, is never published to other replicas, and finds nothing once the round has ended. Co-authored-by: multica-agent <[email protected]> * fix(wecom): keep the reasoning buffer across a bubble hand-over Redaction runs over the joined reasoning buffer so a label and its value split across two increments are seen together. A rotation started that buffer over, so `password=` at the end of the old bubble and `hunter2` at the start of the new one were judged apart and the value went out in the clear. The new bubble now starts with no step lines and a fresh throttle but the same raw reasoning buffer. The old bubble is sealed with the hand-over line, so the reasoning is not shown twice. Also gofmt the two-replica test. Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: multica-eve <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
77c17eb361 |
MUL-8410 fix(qoder): run chat turns on Qoder Cloud Agents (#9726)
* fix(qoder): run chat turns on Qoder Cloud Agents * fix(qoder): address review nits for Qoder chat (MUL-8410) - Tell the agent the chat audience (group/direct/unknown) like the local daemon's chat prompt, so an IM group reply is not treated as private. - Update the qoder-runtime README: chat turns and chat session continuation are supported. - Show conversation starters and the new-chat preview for Qoder agents, and keep starters when creating a Qoder agent. - PriorSessionID comment covers chats as well as issues. Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: 安辰 <[email protected]> Co-authored-by: Multica Eve <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
b77b069431 |
MUL-5491: feat(metrics): add MiniMax-M3 standard-tier pricing (#6128)
* feat(metrics): add MiniMax-M3 to pricing registry Map MiniMax-M3 in the backend pricing registry (input $0.60/M, output $2.40/M, cache read $0.12/M; cache write bills at the input rate as M3 has no separate cache-write rate) and add the minimax-m3 alias so its usage is priced instead of landing in the unpriced bucket. * fix(metrics): tighten MiniMax-M3 pricing alias * fix(metrics): price MiniMax-M3 cache writes * docs(metrics): clarify MiniMax-M3 pricing tier * fix(minimax): refresh text choices and M3 standard pricing * revert(views): drop MiniMax model picker filtering from this PR Keep this PR scoped to the MiniMax-M3 server-side pricing row. Hiding runtime-advertised models in the pickers is a separate product decision. Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: octo-patch <[email protected]> Co-authored-by: multica-eve <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
Xichang(Seacen) Zhao
|
908456c601 |
feat(wecom): tell the sender when the media budget gives up on an attachment, and count every failure (#9695)
When the Router's media budget ran out, the resolver had either never started or been cut off, so it saw no failure and the notice it owns never fired. The sender's photo was dropped in silence and the agent answered the words beside it as if there were no photo. engine.MediaAbandonNotifier is an optional extra a MediaResolver can implement. On that path the Router calls it on a goroutine of its own, with its own bounded context, started before database finalization and not waited on by it: the bind, the pending-media marker and the task promotion keep their five seconds whatever the chat write costs. Drain still waits for it. The call carries what only the Router knows: why the budget ended, whether ResolveMedia ran, and the refs it dropped. The WeCom resolver says the same "did not arrive" line it says for an unreadable attachment, and drops that line from its own notice when the budget cut it off, so the sender does not read it twice. A "too large" verdict reached before the deadline still goes. Its own notice also goes out on a goroutine now, so ResolveMedia no longer waits on the chat; the Router's Drain still waits for it, through the optional engine.MediaNoticeDrainer it calls once every media job is done. Both notices run on their own context through the language lookup, the write and the revoked-installation guard (guardReply). The two notice lines move into the copy pack and are written in the reader's language. media_failures_total counts each attachment that never reached the agent, once, by reason: unreadable, too_large, blocked_address (the SSRF guard refused where the host resolved), budget_expired, or shutdown. ResolveMedia counts what it failed on, with a deadline it hit counted as budget_expired rather than unreadable; NotifyMediaAbandoned counts the rest, every attachment when ResolveMedia never ran and the dropped refs when it did. It is counted even when there is no socket to tell the sender on. |
||
|
|
16f0d307dd |
fix(realtime): announce comments folded into a queued run (#9702)
A reply posted while the agent already has a queued run in the same thread is folded into that run on the server: the reply becomes the run's trigger and the earlier comment joins its covered set. The run's status does not change, so no task event fired and open issue pages kept the run under the earlier comment, leaving the new reply with no visible queue row. Publish task:coalesced after the merge (and after a delegated-failure merge). Web and desktop already refresh the issue's runs on any task: event, which moves the queued block under the newest comment it answers; mobile now listens for it too. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
3451e311a6 |
MUL-8401: fix(wecom): deliver a relayed seal to the replica holding the bubble (#9717)
* fix(wecom): deliver a relayed seal to the replica holding the bubble
A seal frame ends a round that has no words to send: a cancellation, or a
completion with nothing to say, raised on a replica that does not hold the
bubble. It never arrived. The frame carries no installation id, and the relay
addressed every frame by one:
- it was published under scope id "", and the shard reader drops a frame with
no scope id (ShardedStreamRelay.deliverMessage);
- perform's first check is whether this replica owns that installation's
socket, and ownsSocket("") is false everywhere.
So off-lease the bubble kept spinning for the rest of the stream's window.
A seal is for whoever holds the round, which neither the socket nor the claim
can tell. It is now published under its task id, and perform hands it to
every replica's deliverRelayed with no ownership gate and no claim. Every
replica but the holder finds no round and returns; a copy that reaches the
holder again finds the round already taken. A claim would be wrong as well as
unnecessary: taken once for the deployment, it can be won by a replica
without the round, and the holder would never close the bubble.
The relay test double now drops a frame with no scope id, as the shard reader
does. Delivering it anyway is what hid this.
* fix(wecom): dispatch relayed seals by task id, not the empty installation
A seal carries no installation id, so keying its dispatch queue by
installation put every seal in the deployment on one shard queue behind
one worker. A cancel-all burst larger than QueueDepth was shed, and those
bubbles kept spinning. A seal needs no ordering against other frames, so
it is now dispatched by task id.
Co-authored-by: multica-agent <[email protected]>
---------
Co-authored-by: Multica Eve <[email protected]>
Co-authored-by: multica-agent <[email protected]>
|
||
|
Xichang(Seacen) Zhao
|
53b3fba3d0 |
fix(wecom): read a long answer in a bubble to the end (#9692)
A closing stream frame is capped at 20480 bytes, the same as a message, and truncateStreamContent clips it to fit: an answer past the cap ends in "…" and the rest of it is nowhere the reader can reach. The plain-message path already splits a long answer into numbered pieces (#8344); the bubble path did not. The bubble now holds what one frame can and the rest follows as messages directly underneath it, split by splitForWire so the two paths break an answer at the same places. Both writers of an answer into a closing frame make the split: deliverAnswer on the replica that produced the completion, and deliverRelayed on the one that took a routed reply. The other two closers (sealRelayedRound, writeClosing) only ever write short fixed copy. The rest goes out under one hold of the chat's lock (sendPiecesCtx, which sendTextCtx now uses for a whole answer too), so nothing else pushed to the same chat lands between its pieces. Each piece is a reply frame and re-reads the installation first (guardReply, #9694). The rest is sent when the seal is on screen and also when its outcome is unknown: it has never been sent, so it cannot print twice, and holding it back leaves a bubble ending at (1/n). It runs on fallbackBudget, because an unknown seal has usually spent the turn's deadline. A piece under the bubble that fails stops the rest and the reply is counted by the seal's verdict, the way sendTextCtx reports a later piece failing (errPartiallySent): the reader has the head of the answer, and a drop would tell an operator to resend it. |
||
|
Multica Eveandmultica-agent
|
354243b459 |
MUL-8386: feat(cli): add workflow and project workflow commands (#9714)
* feat(cli): add workflow and project workflow commands Wrap the existing /api/issue-workflows endpoints and the project workflow switch so agents and scripts can read and change workflows without the web UI. Permission checks stay on the server. Refs #9698 Co-authored-by: multica-agent <[email protected]> * docs(skill): teach the workflow CLI in the multica-platform skill Workflow edits are rare, so the commands live in the projects reference instead of the per-turn brief. Co-authored-by: multica-agent <[email protected]> * fix(cli): keep the mapping plan when a workflow write needs status mappings On a 409 workflow_status_mapping_required, project workflow set and workflow update now print the server's code and full plan to stdout (JSON, or a table with --output table) and the message to stderr, and still exit non-zero. The default formatter kept only the first message. Also fix the zh CLI reference wording for tasks vs runs. Co-authored-by: multica-agent <[email protected]> * fix(cli): keep error bodies up to 64 KiB so structured errors parse newHTTPError kept only 4 KiB of an error response. A workflow's 409 status mapping plan lists one row per status in use and passes 4 KiB at around 45 statuses; the cut body no longer parsed, so the CLI fell back to a generic message and dropped the plan. Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: multica-agent <[email protected]> |
||
|
Xichang(Seacen) Zhao
|
0249e8609e |
fix(wecom): upload a file again when WeCom replaces the connection mid-upload (#9715)
The attachment worker resolved the installation's sender once and carried every file on it. When WeCom dropped the socket under a chunk and had already subscribed a new connection, the chunk's second offer went to the closed socket, the upload failed, and the user was told the file could not be sent while a live connection sat in the registry. A failed upload never produced a media_id, so nothing reached the chat and starting over cannot duplicate anything. sendAttachment now uploads the file once more when the registry holds a different sender than the one the upload used, unless the failure is one a new connection would repeat: a stated refusal, a withdrawn permission, an empty or oversize file, or a spent budget. A failure after the file message went out is still unknown and still never retried. The restart runs under the same context, so the write guard is asked before each of its frames as before. Each file also starts on the sender the registry holds when its turn comes, rather than the one resolved before the first file. |
||
|
|
cd768e9793 |
MUL-8397: fix(daemon): parse OpenCode 2.x nested MCP server config (#9713)
* fix(daemon): parse OpenCode 2.x nested MCP server config OpenCode 2.x nests user MCP servers under mcp.servers; 1.x keeps them flat under mcp. Reading both as the flat mcp map turned a 2.x config into a single literal "servers" entry with no url or command, so the OpenCode 2.x MCP support check refused every run that carried user MCP servers. Resolve the user server map through a provider dispatcher: OpenCode detects the shape from the content (mcp.servers if present, else the flat mcp map); every other provider keeps the existing dotted-key lookup. Applied to both the task-local merge and the agent capabilities inventory. Co-authored-by: multica-agent <[email protected]> * fix(daemon): respect disabled OpenCode MCP servers (MUL-8397) Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: multica-agent <[email protected]> Co-authored-by: Sol-Boy <[email protected]> |
||
|
Xichang(Seacen) Zhao
|
4157f5b26e |
fix(selfhost): pass MULTICA_WECOM_DEFAULT_LOCALE into the backend container (#9691)
router.go reads MULTICA_WECOM_DEFAULT_LOCALE to pick the language the bot writes its own copy in for readers it cannot look a language up for (a group chat, anyone not linked to a Multica account). Neither .env.example nor docker-compose.selfhost.yml named it, so a self-hosted operator had no way to find the knob, and setting it on the host did nothing: compose forwards only what its environment: block lists, and the deployment silently stayed on zh-Hans. Add it to both, and a test that parses router.go for every MULTICA_WECOM_* literal it reads and fails when either file is missing one. |
||
|
Bohan Jiangandmultica-agent
|
fe23bd1161 |
MUL-8392: feat(properties): support up to 200 select options (#9712)
* feat(properties): support up to 200 select options Co-authored-by: multica-agent <[email protected]> * fix(properties): cap filter values and streamline picker tests Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: multica-agent <[email protected]> |
||
|
|
799f5614a7 |
fix(lark): apply pong heartbeat updates and close failed sessions (MUL-8394) (#9708)
* fix(lark): apply pong heartbeat updates and close failed sessions * fix(lark): keep read deadline above ping cadence and accept updates when ping loop is unarmed - Raise the effective read deadline to 3x the active ping interval so a long Lark-requested cadence cannot trip the idle read timeout. - Let pingLoop receive interval updates even when the initial interval is non-positive, so Run never blocks delivering a pong config update. Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Multica Eve <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
Bohan Jiangandmultica-agent
|
374b6f3db4 |
fix: respect omp profiles when discovering runtime skills (#9711)
Co-authored-by: multica-agent <[email protected]> |
||
|
Xichang(Seacen) Zhao
|
bf8970aa33 |
fix(wecom): ask again before a queued reply is written to a revoked installation (#9694)
The installation's status is read once, when a reply is addressed. The reply then waits for the chat's turn and the socket's writer, and a revocation in between was written straight through: the answer, the closing frame of its bubble, a failed run's notice, an inbox push. replyWriteGuard puts the check on each of those frames, the way attachmentWriteGuard already does for files. Stream frames now consult the guard too, so a bubble is held to the same check as the message it would fall back to. A revoked installation is a skip (installation_inactive), not a drop; a permission that cannot be read is a drop on the direct path and another offer on the relay. A retried closing frame that the guard refuses was written nowhere, which says nothing about the first attempt; seal now keeps the first attempt's unknown instead of reporting the bubble provably empty. |
||
|
|
c066f7ea61 |
MUL-8382: fix(wecom): write the unreadable-message receipt in the sender's language (#9693)
* fix(wecom): write the unreadable-message receipt in the sender's language The one line the read loop sends back for a message it cannot read (a location card, an image callback with no url) was a zh-Hans literal in wecom_channel.go, so a sender whose Multica profile says English got every other notice in English and this one in Chinese. It was the last copy #8831 left as a literal; the lint's pending list counted it. It is now copyPack.UnsupportedMsgType, resolved the way the Replier resolves its notices: the sender's profile language in a 1:1, the deployment's in a room. The read loop gets the lookup through ChannelDeps.Languages, which router.go sets, and RegisterWecom warns when it is nil, the way NewOutboundReplier does, so the boot-path wiring test covers both. The lookup runs only for a message nothing here can read; readable messages take no extra query. * fix(wecom): resolve the receipt copy after the dedup claim A redelivery stopped at the claim no longer pays for the language lookup. Also hoist the logger fallback in RegisterWecom so both warnings share it. Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Multica Eve <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
2083970b51 |
MUL-8374: fix: keep private chat content to the session owner (realtime + REST) (#9677)
* MUL-8374: fix(realtime): deliver private chat content only to the session owner Chat events and chat-task transcript (task:message) frames were fanned out to every connection in the workspace. Route every chat:* event, and task:message for chat tasks, to the chat session creator's user scope instead. Producers stamp the owner on the bus event (from the same chat_session row they already read); an event without an owner is dropped rather than widened. Chat-task lifecycle events keep workspace fanout for agent presence. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(server): keep chat-task content private on REST task reads The transcript endpoint now 404s chat tasks for anyone but the chat session's creator, and the agent run list and workspace task snapshot keep foreign chat-task rows but blank their result and error. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(realtime): scope private chat delivery to the owner's workspace connections User-scope delivery reached the chat creator's connections to every workspace. Those are authorized against their own workspace only, so a creator removed from the chat's workspace kept receiving its transcript and reply through a socket to another workspace. Chat content now goes to a workspace_member scope each connection joins at connect time; it is re-authorized against the chat's workspace on every write. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
Multica Eveandmultica-agent
|
f2a6398d6e |
MUL-7257: refactor(server): remove unused compatibility queries and capabilities (#9671)
* refactor(comments): drop redundant RFC3339 fallback parse RFC3339Nano parsing already accepts timestamps without fractional seconds, so the second time.Parse(time.RFC3339) attempt can never succeed where the first failed. Keep the existing 400 responses and add a test pinning second-precision since and the invalid since/before cases. Co-authored-by: multica-agent <[email protected]> * refactor(daemon): stop advertising unread plugin capabilities execution-manifest-v1, agent-skill-v1 and remote-mcp-v1 have had no server-side reader since the Plugin V1 gates were removed in #7144. Drop the constants and the daemon advertisement. Co-authored-by: multica-agent <[email protected]> * refactor(autopilot): remove legacy no-reservation terminal queries UpdateAutopilotRunCompleted, UpdateAutopilotRunFailed and UpdateAutopilotRunSkippedWithResult have no production callers; terminal transitions go through UpdateAutopilotRunTerminalWithQuota. Their generated Params structs were still used as service argument types, so replace them with service-local structs. The reconciler test now seeds its crash window with an explicit fixture UPDATE. Co-authored-by: multica-agent <[email protected]> * refactor(tasks): remove unused StartAgentTask query Production start paths use StartAgentTaskWithSupplement. Move the seven test fixtures to the same query with both capability flags off, which performs the same running transition. The empty-body start handler branch is unchanged. Co-authored-by: multica-agent <[email protected]> * refactor(channels): remove unused pending_fresh queries MarkChannelChatSessionPendingFresh, LockChannelChatSessionPendingFresh and ClearChannelChatSessionPendingFresh have no callers. The context generation path and ClearChannelChatSessionPendingFreshForRevision remain. Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: multica-agent <[email protected]> |
||
|
Basuotianandmultica-agent
|
e7dfbe83bd |
MUL-7995: fix(openclaw): report actionable state ownership lock error (#9222)
* fix(daemon): give OpenClaw tasks a private per-agent state dir (INFRA-70) Co-authored-by: multica-agent <[email protected]> * fix(openclaw): report actionable state ownership lock error (INFRA-70) * fix(openclaw): handle resident gateway local refusal (INFRA-70) --------- Co-authored-by: multica-agent <[email protected]> |
||
|
Bohan Jiangandmultica-agent
|
46d4e26712 |
MUL-8347: fix Hermes compatibility with gateway-backed and older ACP (#9666)
* fix(hermes): support gateway-backed and in-process ACP Co-authored-by: multica-agent <[email protected]> * fix(hermes): interrupt active task gateways during cleanup Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: multica-agent <[email protected]> |
||
|
|
375b23247b |
fix(codex): price GPT-6.1 Terra and add it to the fallback catalog (#9668)
gpt-6.1-terra runs through Codex but has no row in either price table, so its usage estimates at $0 and shows up as an unmapped model. Add OpenAI's standard short-context rates ($1 input, $0.05 cached input, $1.25 cache write, $5 output per 1M) to the backend metrics table and the frontend estimate table. The model is live-only on codex-cli 0.164.0-alpha.11.1 (absent from `codex debug models --bundled`), so also add it to the static fallback catalog with the observed `low` default and low..ultra efforts, and normalize its label alongside the rest of the GPT-6 family. Co-authored-by: Claude Opus 5.5 <[email protected]> Co-authored-by: Multica Eve <[email protected]> |
||
|
|
ec72a00ac3 |
MUL-8342 fix(daemon): reject legacy DSH profiles (ZIC-374) (#9649)
* fix(daemon): reject legacy DSH runtime profiles Co-authored-by: multica-agent <[email protected]> * fix(dsh): show bridge migration steps in refusal notices Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: multica-agent <[email protected]> Co-authored-by: Sol-Boy <[email protected]> |
||
|
|
07221fbc64 |
docs(skills): keep theme text tokens off fixed light fills in html blocks (#9664)
Co-authored-by: Multica Eve <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
inoisyandEvgeniy Yakutov
|
828e0c927e |
MUL-8346: fix(agent): handle Claude Code session and weekly limits (#9652)
* fix(taskfailure): classify Claude Code session and weekly limits as provider quota Claude Code reports a spent subscription window as "You've hit your session limit · resets 2pm (UTC)" or "You've hit your weekly limit · resets <date>". Neither matched the quota patterns, so these runs were classified as agent_error.unknown and quota-aware handling never saw them. * fix(agent): stop waiting for claude after an error result With stdin closed, the Claude Code CLI stays alive while any background task it started is running, then opens a new turn when the task ends. After an error result (a spent session limit, an API failure) that turn cannot change the outcome, yet the daemon kept the run open for the background task's lifetime. Terminate the process group 10s after an error result, the same way cancellation does, so the run fails with the result's error. --------- Co-authored-by: Evgeniy Yakutov <[email protected]> |
||
|
|
f0c361ebca |
feat(ui): Multica icons, roomier menus and typography pass (MUL-8334, MUL-8338) (#9640)
* refactor(ui): add @multica/ui/icons as the single icon entry point (MUL-8334) Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * refactor: import icons from @multica/ui/icons instead of lucide-react (MUL-8334) Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * chore(lint): ban direct lucide-react imports outside @multica/ui/icons (MUL-8334) Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(ui): draw Multica icons and enlarge primitive icon sizes (MUL-8334) Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(ui): roomier menu and sidebar rhythm (MUL-8334) Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(ui): mark body and detail layers in icon glyphs (MUL-8334) Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(ui): Inter optical sizing, lighter titles, CJK leading and zh-TW font stack (MUL-8338) Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(ui): 13px for in-place reading text, regular weight for descriptive labels (MUL-8338) Issue detail timeline, comment headers, properties panel, wakeup and PR rows move from 12px to 13px; sub-issue identifiers match list/board (12px regular); sidebar group labels drop to regular weight; board card titles get more leading under a CJK UI via a new cjk: variant. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(ui): tint the active sidebar icon's body with a light fill (MUL-8334) Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(ui): draw every remaining icon and drop lucide-react (MUL-8334) All 286 icon names now render Multica drawings. The drawings, their construction helpers and the generator live in packages/ui/icons/drawings; glyphs.ts and index.ts are generated and a test fails when they drift. Search loses its lens highlight and Inbox its arrow. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * chore: guard against icon libraries and document drawing new icons (MUL-8334) Lint bans the common React icon libraries, AGENTS.md points to a new multica-icons skill, route icons must mark a body for the active tint, and comments no longer describe product icons as Lucide. Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(ui): fill the active sidebar icon solid with its details knocked out (MUL-8334) Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix(lint): ban icon library subpaths too (MUL-8334) Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Cursor <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
bb3747186e |
feat: let agents steer running turns they could start (MUL-8285) (#9630)
* feat(server): let agents steer running turns they could start (MUL-8285) An agent comment can now be added to another agent's running turn, the way a member steers from the composer: exactly when the comment could start that agent (canInvokeAgent, judged on the human at the top of the chain), never its own turn. - steer_task_ids now work for agent authors; steer: true asks the server to add the comment to the current running turn of every recipient without a named turn. - Agent-authored bindings are capped per turn (10) and per authoring agent (3). The task row is locked in its own statement before the bind, so the cap count sees what an earlier lock holder committed. - An agent comment that steered a turn but was never read is replayed on completion like the mention it replaced. Member receipts keep today's behaviour (the member sees the failure and can retry or resend). - The supplement claim returns author_type and the agent's name. A daemon that does not send task-supplement-author-v1 gets the attribution in the content, so an agent message never reads as a human's. - can_supplement on task-runs is computed for an agent caller by the same rule. Co-authored-by: Tim Smart <[email protected]> Co-authored-by: WitMiao <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(daemon): present an agent's added message as an agent's (MUL-8285) Claims send the task-supplement-author-v1 capability and render agent messages as peer input that cannot replace the objective or override humans. Co-authored-by: Tim Smart <[email protected]> Co-authored-by: multica-agent <[email protected]> * feat(cli): add --steer and --steer-active to issue comment add (MUL-8285) Docs: CLI help, CLI_AND_DAEMON.md, the CLI docs in every locale, the squad leader briefing, and a steering section in the multica-platform mentions reference. Co-authored-by: Tim Smart <[email protected]> Co-authored-by: multica-agent <[email protected]> * fix: close the review gaps in agent steering (MUL-8285) - Replay an unread agent steer even when it reached the turn through an implicit route (a worker's reply to its leader): a comment bound to the completing run counts as planned input for the replay filter. - Drop the receipt of an agent steer once completion has replayed it, so no client (old or new) offers to resend what the follow-up already handles. - Never hand an agent's message to a daemon without task-supplement-author-v1: it would frame it as a human's. The message stays pending, settles when the turn ends, and is replayed as a mention. - Tell a squad leader about --steer only when its daemon advertises cli-comment-steer-v1, since an older CLI rejects the flag; the skill reference says to drop the flag on 'unknown flag'. Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Tim Smart <[email protected]> Co-authored-by: WitMiao <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
4a0c147c50 |
fix(storage): support GCS S3 XML API signing (MUL-8232) (#9513)
* fix(storage): omit unsigned headers from SigV4 to support GCS S3 XML API Google Cloud Storage through its S3-compatible XML API mutates the Accept-Encoding request header in transit (identity -> identity,gzip(gfe)). Because aws-sdk-go-v2 signs Accept-Encoding into SignedHeaders by default, this mutation causes HTTP 403 SignatureDoesNotMatch on storage calls. This commit: - Automatically excludes accept-encoding from SigV4 signing when AWS_ENDPOINT_URL points to Google Cloud Storage (storage.googleapis.com). - Adds S3_UNSIGNED_HEADERS to configure additional unsigned headers, preserving accept-encoding when targeting GCS. - Strips unsigned headers prior to SigV4 signing while restoring captured headers (including Accept-Encoding: identity) on the wire via an HTTP client wrapper. - Documents S3_UNSIGNED_HEADERS across all 5 documentation translations (en, zh, ja, ko, fr) with an integrity protection note. - Adds regression tests for SigV4 omission, wire transmission, and presigned X-Amz-SignedHeaders. Closes #9509 * fix(storage): scope GCS signing workaround to Accept-Encoding (MUL-8232) Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Sol-Boy <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
1eb4b5d1f8 |
MUL-8203: record permission-blocked mentions as system replies (#9624)
* fix(comments): leave system replies for denied mentions (MUL-8203) Co-authored-by: multica-agent <[email protected]> * fix(comments): address refusal notice review nits (MUL-8203) Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Yushen <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
b20cea33ad |
fix(desktop): wait for daemon shutdown before quitting (MUL-8290) (#9590)
* fix(desktop): wait for daemon shutdown before quitting * fix(desktop): clarify daemon quit progress and dialogs * fix(desktop): preserve stopped daemon after cancelling quit * fix(desktop): suppress deferred restart after cancelled quit * fix(daemon): require explicit cleanup confirmation before quit Record a profile-scoped, PID-matched shutdown result before releasing the daemon lock. Keep cleanup timeouts and provider cleanup_unconfirmed results sticky, and reject absent, malformed, stale or negative results in the forced stop path. Co-authored-by: multica-agent <[email protected]> * fix(daemon): keep unconfirmed cleanup sticky across repeated stops When a stopped daemon left no matching shutdown result (for example a daemon started by an older CLI, or a stale record from an earlier daemon), the first forced stop correctly failed, but a repeated stop saw an idle profile with no result file and reported success. Record the unconfirmed outcome on failure so cancelling the quit dialog and quitting again still requires an explicit ignore. A profile that never ran a daemon still stops as an idle no-op, and the next daemon start replaces the record. Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Bohan-J <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
7d02596fc3 |
fix(agent): MUL-8307 drive OpenCode over ACP so long tools and subagents are not force-stopped (#9619)
* fix(agent): drive OpenCode over ACP so long tools and subagents are not force-stopped `opencode run --format json` reports a tool only when it finishes, with the call and its result in one event. The daemon never saw a tool in flight, so a long silent tool or a delegated subagent was judged by the 10-minute OpenCode idle window instead of the tool budget and force-stopped (GH #9614). Built-in OpenCode 1.15+ and 2.x runtimes now run `opencode acp`, which on both majors announces a tool when it starts executing. MessageToolUse is emitted from that in_progress frame, so the daemon applies AgentToolWatchdog for the whole tool. One code path covers both majors: - model via session/set_config_option (2.x has no session/set_model); effort via the advertised `effort` option, splitting 2.x `model#variant` - resume via session/load; a missing session is reported as ResumeRejected - cancel via session/cancel, which aborts the running tool on both majors - MCP: 1.x keeps OPENCODE_CONFIG_CONTENT; 2.x gets servers in session/new, which lifts the 2.x MCP refusal on this transport - 2.x subagent text (`_meta["opencode/child-session"]`) stays out of the parent's output; its tool calls are kept - usage summed from the session export over this turn's assistant messages, since the prompt result's usage is not per-turn on either major The run transport stays as the fallback for custom profiles, older or unknown versions, agents with custom_args, and MULTICA_OPENCODE_TRANSPORT=run. Also fix the run transport on 2.x: tool_use parts carry the call id as `id` (1.x: `callID`), so every 2.x tool message had an empty CallID. Co-authored-by: multica-agent <[email protected]> * fix(agent): bound OpenCode ACP teardown when the runtime stops reading A runtime that stops reading stdin mid-prompt left the prompt write blocked holding the client's write lock. The cancel path then waited on that lock to send session/cancel, and the cancel grace and the process-group kill only ran after the write, so the process could outlive the task indefinitely. - session/cancel is sent only once the prompt is fully delivered, from its own goroutine, so the grace starts immediately and a stalled pipe cannot hold it - a turn OpenCode never answered (or a cancelled/timed-out setup) is torn down forcibly: stdin closed first to release a blocked write, then SIGTERM to the group and SIGKILL after the grace; an answered turn still exits on EOF Regression tests cover a stalled prompt write under a 1s timeout and a runtime that ignores session/cancel. Co-authored-by: multica-agent <[email protected]> * fix(agent): bound OpenCode ACP teardown when a setup write stalls Setup requests (initialize, session/new, session/load, set_config_option) are synchronous stdin writes that check their context only after the write returns. A runtime that stops reading after initialize, with a large MCP payload in session/new, left the run goroutine blocked in that write, so the deferred teardown never ran and the process outlived the timeout. A watcher now force-stops the process as soon as the run ends before the prompt is fully delivered; closing stdin releases the blocked write. Once the prompt is delivered the existing session/cancel path owns shutdown. Regression test: 256 KiB MCP env, runtime stalls after initialize, 1s timeout; returns in ~1s (60s before the fix). Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Multica Eve <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
Multica Eveandmultica-agent
|
55be334864 |
fix(taskfailure): classify Kiro dispatch failure as retryable provider_network (MUL-8314) (#9621)
Co-authored-by: multica-agent <[email protected]> |
||
|
FrezcandFrezc
|
a8e7561421 |
[MUL-8300] fix(lark): resolve outbound references to workspace links (#9604)
* fix(lark): resolve outbound references to workspace links * test(wecom): use normal budget for no-holder state assertion * fix(channel): preserve Markdown reference source boundaries * fix(lark): format references with one workspace lookup * fix(deps): update x/net for HTML parser advisory --------- Co-authored-by: Frezc <[email protected]> |
||
|
|
ca593b3103 |
MUL-8302: fix(daemon): preserve task environment in Hermes tool shells (#9607)
* fix(daemon): preserve task environment in Hermes tool shells * fix(daemon): complete Hermes task env and preserve derived isolation (MUL-8302) Co-authored-by: multica-agent <[email protected]> * fix(deps): patch x/net HTML vulnerability (MUL-8302) Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Sol-Boy <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
mrlonely
|
84e56b6b00 |
MUL-8256 fix(runtime): recover stalled HTTP/2 connections (#9551)
* fix(daemon): recover stalled HTTP/2 control-plane connections * fix(daemon): recover stalled skill bundle connections * fix(runtime): share HTTP/2 recovery across persistent clients |
||
|
|
6388272a8a |
fix(github): skip PR merge status move when the project workflow lacks the target (#9598)
A merged PR moved its issue to the workspace's pr_merge_status even when the issue's project workflow does not list that status, bypassing the check a manual status change goes through. The decision now loads the project workflow and returns not_in_workflow instead; the issue page renders nothing for it, so the issue silently stays where it is. Default workflows and issues outside any project are unchanged. Co-authored-by: Multica Eve <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
Bohan Jiangandmultica-agent
|
d7d1831fcf |
feat(cli): add run-targeted retries with a fresh option (#9592)
Co-authored-by: multica-agent <[email protected]> |
||
|
Multica Eveandmultica-agent
|
a008d893dd |
MUL-8292: refactor(brief): move Available Commands detail behind --help (#9596)
* MUL-8292: refactor(brief): move Available Commands detail behind --help pointers Each bullet keeps its signature and the behaviour needed at the call site; wakeup condition catalogue, PR snapshot mechanics, actor filters and the repeated file rules now live only in --help and the multica-platform references. Issue brief 17,300 -> 16,372 bytes (-928); chat and autopilot briefs drop by the same amount. New TestAvailableCommandsKeepsCompressedContracts pins the kept contracts and guards the moved detail from drifting back. Co-authored-by: multica-agent <[email protected]> * MUL-8292: fix(brief): keep Available Commands rules self-contained in every kind Review follow-up on #9596: - comment list / issue create / comment add pointed at Workflow step 2 and Comment Formatting, which chat and autopilot briefs do not have; the bounded-read recipe and the workdir file rule are inline again. - repo checkout restores both keep conditions and the clean-checkout case. - wakeup scopes last_error to missing PR links/snapshots; a pending check or unmet condition records nothing. Contract test now runs every kind x provider, and a new test fails when the shared section references a heading or workflow step the kind lacks. Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: multica-agent <[email protected]> |
||
|
|
188971168d |
fix(runtimes): spell out Qoder Cloud Agents instead of QCA (MUL-8282) (#9581)
* fix(runtimes): spell out Qoder Cloud Agents instead of QCA The UI, docs, landing copy and server messages called the product "QCA" in labels such as "QCA Agent", "QCA runtime" and API errors like "QCA token is required". Use the full name Qoder Cloud Agents everywhere a person reads it, and add it to the brand list in the naming conventions so the abbreviation does not come back. Co-Authored-By: Claude Opus 5.5 <[email protected]> * fix(runtimes): refine Qoder Cloud Agents copy (MUL-8282) Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: 安辰 <[email protected]> Co-authored-by: Claude Opus 5.5 <[email protected]> Co-authored-by: Sol-Boy <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
ac673fbf19 |
MUL-7384: fix(onboarding): require confirmed daemon setup targets (#8907)
* fix(onboarding): show self-host setup command when daemon URLs configured (fixes #8425) * fix(onboarding): fail closed until daemon setup target is confirmed (MUL-7384) Co-authored-by: multica-agent <[email protected]> * fix(onboarding): preserve legacy self-host setup commands (MUL-7384) Co-authored-by: multica-agent <[email protected]> * fix(i18n): restore Traditional Chinese locale parity (MUL-7384) Co-authored-by: multica-agent <[email protected]> * test(daemon): pin Pi upstream cancellation handoff (MUL-7384) Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Sol-Boy <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
c11b3375f5 |
MUL-8269: fix(runtimes): price Bedrock-served Anthropic ids and DeepSeek-V4.1-Flash (#9561)
* fix(runtimes): price Bedrock-served Anthropic ids and DeepSeek-V4.1-Flash Bedrock reports Anthropic models as `<region>.anthropic.<id>` with an inference-profile version on snapshots (`us.anthropic.claude-sonnet-5`, `eu.anthropic.claude-sonnet-4-5-20250929-v1:0`). Neither resolver accepted that shape: the backend alias rules anchor on `/`, `:` or start-of-string, and the frontend `canonicalCandidates` only strips `/` and `:` routing prefixes, so the vendor and region prefixes survived into the lookup. Every Bedrock-served turn therefore missed the rate table and rendered as $0.00 plus an "unmapped models" warning, while the tokens were still counted in `task_usage` (cost_usd_ticks is provider-reported and NULL for Anthropic, so nothing else was pricing them either). Strip the Bedrock vendor/region prefix and the `-v<digit>:<digit>` profile version on both sides, before the existing dot↔dash normalisation. Only `anthropic.`-prefixed ids are rewritten, so non-Anthropic Bedrock models (`us.amazon.nova-*`) stay unmapped rather than borrowing a Claude rate. DeepSeek-V4.1-Flash is also missing: the catalog lists v4-flash and v4-pro only, and there is deliberately no family fallback, so `deepseek-v4.1-flash` and OpenRouter's dated `deepseek-v4.1-flash-1017` both went unpriced. Add the official `deepseek-flash` sheet rate ($0.15/$0.60, cache read $0.003, cache write mirrors input) and one alias rule that admits the dated snapshot — DeepSeek documents that id as the model actually served for the undated slug, so it is the same SKU at the same rate, not a second tier. Both tables are updated together (server/internal/metrics/pricing.go drives the Prometheus counters, packages/views/runtimes/utils.ts drives the dashboard) and both suites pin the new ids, including the negative cases: a later SKU, a `-preview` qualifier or a non-Anthropic Bedrock id must stay unmapped. * ci: re-run (previous backend-tests failure was in channel-media reconciler timing tests, unrelated) * fix(runtimes): align Flash pricing and constrain Bedrock aliases (MUL-8269) Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Sol-Boy <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
16a067f30a |
MUL-8270: fix(db): keep lease claims within query limits (#9563)
* fix(storage): claim one cleanup object per statement * fix(tasks): bound stale dispatch recovery candidates * fix(db): restore English lease claim documentation Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Sol-Boy <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
73f5664ca4 |
MUL-8263 fix(agents): reject reserved custom arguments on save (#9560)
* fix(agents): reject reserved custom arguments on save * fix(agents): validate OMP session arguments before saving Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Sol-Boy <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
|
bcf40040e0 |
MUL-8233: fix Pi cancellation and guard unconfirmed cleanup (#9557)
* fix(agent): join Pi cancellation cleanup and guard continuation Co-authored-by: multica-agent <[email protected]> * fix(agent): preserve Pi completion and negotiate cleanup Co-authored-by: multica-agent <[email protected]> * perf(agent): avoid Pi exit delay and index cleanup guards Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Multica Eve <[email protected]> Co-authored-by: multica-agent <[email protected]> |
||
|
小七
|
ff6d3207c8 |
fix(agent): keep per-run usage when Claude resume baseline capture fails (#9567)
When a resume's baseline capture fails, the result event's modelUsage was stored as the task's usage as-is. On newer Claude Code a resumed session restores its cumulative counters, so that reported total spans every prior task in the session: the #8781 inflation returned for every baseline-unavailable mode. The reported trigger is a custom runtime profile whose command is a wrapper script that sets CLAUDE_CONFIG_DIR internally - the daemon resolves the transcript from its own env (os.Environ() + agent custom_env) and never sees it, so every resume logged 'claude usage baseline unavailable' and stored the running session total (observed ~10x on a session's 7th task). The code already distinguishes a transcript that exists but has no cost-state (older Claude Code reports per-run totals there; behavior preserved, pinned by existing tests) from one that cannot be found at all. For the latter the reported modelUsage is now treated as non-authoritative: usage keeps the stream-derived per-run totals, and the reported totals are only recorded for the resume-rejected recapture where the run is known to have started a fresh session of its own. Affected modes now degrade to a conservative per-run approximation instead of silently storing a cumulative total; the WRN stays, reworded to say what actually happens. Out of scope, per the adjudication on the issue: per-profile env vars (schema/feature) and expanding $HOME/~ in custom_env values (makes the manual workaround portable but leaves the default inflation path live). Ablation: reverting only server/pkg/agent/claude.go on this branch makes both new tests fail again with the cumulative total stored (700/30/800/80 instead of the per-run 70/0/80/8). Fixes #9564 Generated-by: GLM-5.3-Flash (ZCode) |
||
|
Multica Eveandmultica-agent
|
7c50af7b66 |
MUL-8094: feat(wakeups): accept blocked in --until-issue-state (#9553)
* feat(wakeups): accept blocked in --until-issue-state Lets a coordinator wait for a specific child to reach blocked, alongside the existing in_review handoff, without waiting for the child to close. Exact status key match, same as in_review. Refs #9344 Co-authored-by: multica-agent <[email protected]> * fix(i18n): keep blocked state key in lowercase English in zh wakeup copy Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: multica-agent <[email protected]> |
||
|
Frezc
|
e30ae35301 | fix(search): treat out-of-range issue numbers as text (#9501) |