The compact view was the original list layout and is what users expect
on this page; the post-#2840 default of comfortable changed long-standing
behavior. Reset the unpersisted default (and the cross-workspace fallback
in `merge`) back to compact. Updates the view-store tests accordingly.
MUL-2464
Co-authored-by: multica-agent <github@multica.ai>
* feat(comments): thread-internal pagination via --tail + reply cursor (MUL-2421)
Long threads inside a single issue still forced agents to read every reply
once they used --thread, even after MUL-2387 fixed cross-thread noise. This
adds reply-level paging so a 200-reply thread can be navigated tail-first
without dragging the whole conversation into prompt context.
- New SQL query ListThreadCommentsForIssuePaged: same recursive root walk
as the legacy thread query, but caps reply count and supports an
(created_at, id) composite cursor. Root is unconditional — even tail=0
emits it so the reader keeps the "what is this thread about" context.
- Handler ListComments: parses `tail` (non-negative, ThreadTailSet flag
preserves the tail=0 intent), threads it through to the paged query,
and re-uses X-Multica-Next-Before / X-Multica-Next-Before-Id for the
reply cursor. Cursor's meaning is now context-dependent: thread cursor
under --recent, reply cursor under --thread + --tail.
- CLI: new --tail flag (only valid with --thread; mutually exclusive
with --recent), reply-cursor semantics for --before / --before-id when
paired with --thread + --tail, stderr label flips to "Next reply cursor"
so an operator copy-pasting the cursor knows which scope it scrolls.
- Tests cover the new contract: tail=N keeps newest N + root, tail=0 is
root-only, anchor on a nested reply still walks up, reply cursor
scrolls older replies page-by-page, since combined with tail filters
after the cut, and the negative-flag-combination matrix.
Out of scope: prompt template update to hint at `--thread <id> --tail 30`
on long threads — separate follow-up per the issue.
Co-authored-by: multica-agent <github@multica.ai>
* fix(comments): only emit reply cursor when older reply exists (MUL-2421)
The thread-tail path emitted `X-Multica-Next-Before` whenever the page
filled to exactly the requested reply count, even when there was nothing
older to scroll to. So `--thread <root> --tail 3` on a thread with
exactly 3 replies sent a cursor that, when followed, returned just the
root — a wasted round-trip that surfaced as a phantom "older replies"
affordance in the agent prompt.
Switch to a `reply_limit + 1` probe: ask the SQL for one extra row, trim
the oldest overflow before responding, and only emit the cursor when an
older reply actually existed. The exact-boundary case (replyCount ==
tail with no overflow) now returns no cursor.
Also documents `--thread/--tail/--recent/--before` and the cursor
semantics in CLI_AND_DAEMON.md, which was the second must-fix in the
MUL-2421 review.
Co-authored-by: multica-agent <github@multica.ai>
* fix(comments): suppress reply cursor when --since covers older replies (MUL-2421)
In the thread + tail + since path the server still emitted a reply cursor
whenever there was an older reply on disk, regardless of `since`. If the
oldest retained reply on the page was already `<= since`, every older
reply was guaranteed to be filtered out too, so the next page only ever
returned the root — wasting round-trips until the agent walked the whole
pre-`since` history. Mirror the recent + since suppression: when
`replies[0].CreatedAt <= since`, drop the cursor.
Test covers the exact case from Elon's review: tail=2 overflow, body
keeps a fresher reply, but the cursor target (oldest retained reply) is
already past `since` — header must be empty.
Co-authored-by: multica-agent <github@multica.ai>
* feat(prompt): default comment-trigger reads to --thread --tail 30 (MUL-2421)
Comment-triggered agents previously defaulted the trigger-thread read to
the unbounded `--thread <id> --output json`, which dumps the full thread
into the prompt — exactly the kind of context bloat MUL-2387 fixed at the
cross-thread layer but never bounded inside a single thread.
Use the new `--tail` flag landed earlier in this PR (server + CLI) as the
default for both the per-turn prompt and the runtime-config Workflow:
- `--thread <trigger-id> --tail 30 --output json` is the new default.
Root is always included so "what is this about" context survives.
- If 30 replies aren't enough, the prompt now spells out the reply
cursor: re-feed the stderr `Next reply cursor: --before <ts>
--before-id <reply-id>` pair back to walk older replies.
- `--recent 20` stays as the cross-thread background fallback, with an
explicit callout that the same `--before` / `--before-id` flags walk
*threads* (not replies) in that mode.
- Available Commands core line now surfaces `--tail N` and both stderr
cursor labels so non-workflow callers also discover the flag.
- `--since` callouts reflect the post-MUL-2421 combinable mode names
(`--thread --tail` / `--recent`).
Tests (`prompt_test.go`, `execenv_test.go`) pin the new defaults and add
a regression guard against the unbounded `--thread` recipe sneaking back
in.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
The post-#2946 onmessage guard logs the raw event.data alongside the
warning. A malformed or rogue server can stream arbitrarily large
garbage and bloat the renderer / desktop main-process log buffers, so
cap the logged payload to the first 200 chars and append a
"(truncated, N chars total)" suffix when truncation occurs.
MUL-2490
Co-authored-by: multica-agent <github@multica.ai>
* fix(cli): accept slug + short UUID prefix in workspace get/update/member (MUL-2385)
`workspace list` shows the 8-char short UUID prefix, name, and slug by
default; `workspace get`/`update`/`member list` only accepted full UUIDs.
That broke the natural list -> get flow: every value the user could copy
from list output was rejected. They had to either rerun list with
`--full-id` or parse the JSON output -- both implementation-detail level
operations.
Extend `resolveWorkspaceByIDOrSlug` with a short UUID prefix fallback
(>=4 hex chars, ambiguous matches return all candidates), introduce
`resolveWorkspaceRef`/`resolveWorkspaceArg` helpers that fetch the
caller's accessible workspaces and resolve UUID/slug/prefix in one call,
and wire them into get/update/member list (switch already used the same
list-then-resolve pattern). Full UUIDs short-circuit the extra
`/api/workspaces` round trip; access control remains on the downstream
endpoint.
Also add a one-line tip after `workspace list` table output pointing
users at get/update/switch with the same identifier columns, and
broaden the command Use strings to `<id|slug|prefix>` so help reflects
the new behavior.
Refs https://github.com/multica-ai/multica/issues/2750
Co-authored-by: multica-agent <github@multica.ai>
* chore(cli): include prefix hint in workspace list footer
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
The model dropdown already exposes a "Default (provider)" option meaning
"follow the CLI's current selection". Tagging the runtime's preferred
model with a small "default" chip created two competing notions of
"default" in the same UI and confused users. Remove the chip from both
the create-agent ModelDropdown and the inspector ModelPicker; keep the
underlying RuntimeModel.default flag intact since thinking-prop-row
still uses it as a fallback heuristic.
Co-authored-by: multica-agent <github@multica.ai>
Replaces the plain "Loading..." text fallback in SquadDetailPage with a
skeleton that mirrors the loaded page's two-column layout (left inspector
+ right tabs panel), matching the SquadsListSkeleton work shipped in #2890.
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* fix(comments): allow agent self-mention to enqueue cross-issue handoff
The @mention path in CreateComment unconditionally skipped any
self-mention. That dropped the child→parent handoff between issues
assigned to the same agent: the child run posted `@J` on the parent
issue, the guard tripped, and the parent's J was never woken — the chain
silently broke.
Drop the self-trigger `continue` in the agent mention branch. Runtime
ready / private-agent gate / HasPendingTaskForIssueAndAgent dedup all
remain, so a same-issue self-mention while a queued or dispatched task
exists is still deduped; a running task no longer pre-empts a new
follow-up (the existing queue coalescing handles that).
Three regression tests:
- cross-issue self-mention enqueues a task on the target issue
- same-issue self-mention while running queues a follow-up
- same-issue self-mention with a pre-existing queued/dispatched task
is deduped
MUL-2338
Co-authored-by: multica-agent <github@multica.ai>
* test(handler): assign per-workspace issue number in self-mention fixture
The fixture inserts two issues in the same test workspace; without an
explicit number both default to 0 and the second insert violates
uq_issue_workspace_number, taking the backend CI job down on PR #2928.
Mirror the workspace-counter advancement pattern from
issue_scheduled_test.go so each fixture issue gets a unique number.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* feat(squads): skeleton loader + AlertDialog archive confirm (MUL-2437)
- Replace `Loading...` text on the squads list with a Skeleton placeholder
matching the SquadCard shape (avatar + title + subtitle), aligning with
the Agents / Dashboard pattern.
- Replace the native `confirm()` on the squad detail Archive button with
the project's AlertDialog (destructive variant, pending-disabled, i18n
copy interpolating the squad name).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* fix(squads): drop misleading restore copy from archive confirm (MUL-2437)
Archive is irreversible — there is no unarchive command (see
apps/docs/content/docs/squads.mdx:113). Aligns dialog copy with
docs: tells the user the action can't be undone and to create a
new squad if they need the routing back.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* feat(issues): surface "agent working" on board + add Working filter (MUL-2452)
Adds a brand-color "agent working" badge to board cards / list rows so
users can see at a glance which issues have an active agent task, plus a
new "Working" toggle on the `/issues` and `/my-issues` headers (next to
the existing scope segmented control) that filters to those issues. The
toggle shows an avatar stack of the agents currently active on the
current surface + scope. Pure frontend: re-shapes the existing
workspace-wide `agentTaskSnapshot` cache via two new selectors
(`activeTasksByIssueOptions` / `workingIssueIdsOptions`), no new SQL,
endpoint, or DB field; WS `task:*` events already invalidate the
snapshot so the badge / filter update in realtime.
Project detail page keeps the per-card badge but intentionally omits the
header toggle (`showWorkingToggle={false}`) to leave the project
surface's filter dimensions unchanged.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* fix(issues): working filter column header reflects filtered count (MUL-2452)
Assignee-grouped board column headers kept showing the unfiltered cache
total when Working was on, because `PaginatedAssigneeBoardColumn` passed
`useLoadMoreByAssigneeGroup`'s cache-derived `total` straight to
`BoardColumn`. The hook still needs the cache total for hasMore, but the
displayed count must follow the visible-after-filter set.
Split the two: when Working is active the column header now uses
`group.totalCount` (set by applyWorkingFilterToGroups) for the assignee
path, and `issueIds.length` for the status path. Load-more keeps reading
from cache so paginated columns still see the full server total.
Regression tests cover applyWorkingFilterToGroups (total rewrite +
empty-group preservation), filterIssues workingOnly combinations, and an
end-to-end assertion via IssuesPage that proves the column header equals
the filtered count, not the cached value.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
* feat(runtime): teach agents the parent/sub-issue protocol (MUL-2338)
Adds a Parent / Sub-issue Protocol section to the runtime brief built by
`buildMetaSkillContent`, emitted whenever the agent is running on a real
Multica issue (assignment- or comment-triggered). Two behaviors are now
documented for every issue-bound agent:
- A. When wrapping up a child issue, post the final result and switch to
`in_review` on this issue first, then post a single top-level comment
on the parent. Mention the parent assignee only when it is another
agent on a still-open parent — never self-mention, never @ member /
squad, never re-trigger a `done` / `cancelled` parent.
- B. When creating sub-issues, choose `--status backlog` for sub-issues
that must wait and `--status todo` for the one to start immediately;
promote with `multica issue status <id> todo` when its turn comes.
The signal is explicitly framed as best-effort — no server-side state
sync, no claim of a guaranteed handshake. The section is skipped for
chat, quick-create, and run-only autopilot runs, which have no
parent/child semantics.
Tests in runtime_config_test.go assert that the section is present in
both issue workflows, absent in the three non-issue modes, and that the
wording does not introduce a non-existent `multica issue list --parent`
command or promise a reliable handshake.
Co-authored-by: multica-agent <github@multica.ai>
* fix(runtime): split Step A of parent/sub-issue protocol by trigger type (MUL-2338)
Comment-triggered runs were inheriting an unconditional
`multica issue status <this-issue-id> in_review` from Step A, which
conflicts with the comment-triggered workflow rule "Do NOT change the
issue status unless the comment explicitly asks for it" (Elon's blocking
review on PR #2918). Step A now branches on trigger type:
- Assignment-triggered: keep "post final results + flip in_review".
- Comment-triggered: complete the reply per the existing workflow rule,
only flip status when the triggering comment asked for it, and gate
the parent-notification steps on actually closing out child work.
Tests lock the boundary: comment-triggered briefs must not contain the
unconditional in_review command, must echo the existing status
guardrail inside Step A, and must spell out the "closing out" gate.
Assignment-triggered briefs still carry the unconditional flip.
Co-authored-by: multica-agent <github@multica.ai>
* fix(runtime): simplify parent/sub-issue mention rule to always @ parent assignee (MUL-2338)
Per Bohan's directive on PR #2918: the per-case mention table (same agent /
member / squad / closed parent) is overkill prompt complexity. Replace it
with a single rule: always @mention the parent's assignee using the URL
that matches assignee_type. The platform's existing run dedup handles
re-triggers, and a single rule is easier for agents to follow predictably.
Preserves the existing comment-triggered boundary (Step A still does NOT
add an unconditional in_review flip on comment-triggered runs).
Co-authored-by: multica-agent <github@multica.ai>
* refactor(runtime): compress parent/sub-issue protocol to 3-rule convention (MUL-2338)
Drop the spec-flavored A/B sub-headings and per-case mention table; keep
three numbered rules (close out child, notify parent, pick backlog vs
todo) plus a one-line best-effort preamble. The comment-triggered
branch still re-asserts the "do not change status unless asked"
guardrail and gates parent notification on actually closing out child
work; the assignment-triggered branch still flips to `in_review`.
Section is now 7 lines instead of 29. A new TestParentSubIssueProtocolIsCompact
guards the ≤10-line ceiling so this stays a convention, not a spec.
Co-authored-by: multica-agent <github@multica.ai>
* fix(runtime): make sub-issue creation rule unconditional in parent/sub-issue protocol (MUL-2338)
Elon's review on PR #2918: the preamble previously gated all three
rules on the current issue having `parent_issue_id`, but rule 3
(creating sub-issues) needs to reach top-level parents that have no
parent themselves — that is exactly where the `todo` vs `backlog`
decision matters most. Move the gate from the preamble onto rules 1
and 2 per-rule; rule 3 now applies to any issue-bound run. Section
stays at 7 newlines (≤10).
Co-authored-by: multica-agent <github@multica.ai>
* refactor(runtime): unify parent/sub-issue protocol as mechanism description (MUL-2338)
Drop the if/else split between assignment- and comment-triggered runs in
the Parent / Sub-issue Protocol section: both runs now read the same
two-rule description of how the parent/child mechanism works. The
comment-triggered workflow rule "Do NOT change the issue status unless
the comment explicitly asks for it" naturally short-circuits the parent
notification (no status flip → not closing out the child → skip), so the
protocol no longer needs to branch on TriggerCommentID.
Tests collapse the two trigger-specific cases into one parameterized
test, and the assignment vs comment status-flip invariants are now
anchored on the real workflow command (with substituted issue id)
instead of the protocol's removed `<this-issue-id>` placeholder.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
Follow-up to #2919 review nits — comments still described the empty
thinking_level as "use runtime default" and claimed ThinkingPicker callers
guaranteed non-empty levels. Both were stale after the semantics changed:
- packages/core/types/agent.ts: clarify that "" clears the override and
the local CLI config / built-in default decides at runtime.
- thinking-picker.tsx: document that the stale-orphan clear path in
ThinkingPropRow mounts the picker with an empty levels list plus a
persisted value, so callers do not guarantee non-empty levels.
Co-authored-by: multica-agent <github@multica.ai>
* fix(issues): retry button targets the row's agent, not the assignee (MUL-2457)
The execution log retry button used to re-fire the issue's current
assignee instead of the agent that actually ran the clicked row. After
a reassignment, or for squad workers / @-mention agents, the rerun
landed on the wrong agent.
POST /api/issues/{id}/rerun now accepts an optional task_id: when set,
the rerun targets that task's agent (and reuses its leader/worker
role). An empty body keeps the assignee-driven CLI/API contract.
The execution-log retry button passes task.id, so per-row retry always
fires the correct agent. enqueueMentionTask gained a forceFreshSession
parameter so the new mention-path rerun keeps the same fresh-session
contract as the assignee path.
Co-authored-by: multica-agent <github@multica.ai>
* fix(issues): inherit trigger provenance + fix cross-issue test (MUL-2457)
Address review feedback on PR #2921:
1. RerunIssue now inherits TriggerCommentID from the source task when
sourceTaskID is valid. Without this, a per-row rerun of a comment-
or mention-triggered task degrades into a generic issue run because
the daemon's buildCommentPrompt path keys on TriggerCommentID. The
inherited summary is rebuilt naturally inside the enqueue helpers
(buildCommentTriggerSummary derives it from the comment ID).
2. The new cross-issue rejection test inserted a second issue without
`number`, hitting uq_issue_workspace_number on a same-workspace
collision with the fixture's issue. Both inserts now claim the next
available per-workspace number (MAX(number)+1) — matching the
pattern used by notification_listeners_test.
Added TestRerunIssueInheritsTriggerCommentFromSourceTask to lock the
trigger provenance contract.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* polish(agent-inspector): optimistic updates + picker layout + thinking-default semantics
Round of cleanup on the agent inspector pickers after using them end-to-end:
1. **Optimistic updates** (`agent-detail-page.tsx`)
The `handleUpdate` callback that backs every inspector picker
(thinking / model / visibility / concurrency / runtime / name /
description / avatar) was strictly sequential:
`await api.updateAgent → invalidateQueries → toast.success`. Each pick
waited 0.5-2s for the network round trip before the trigger chip
updated, which read as visible UI lag.
Snapshot the cached agent list, patch the matching agent
synchronously via `setQueryData`, then run the network request in
the background. On error roll back to the snapshot before the toast
surfaces the cause. All inspector pickers now respond instantly.
2. **Block-in-inline fix in Model + Thinking pickers**
`PickerItem` wraps its children in a flex `<span>`. The picker
bodies had `<div>` children, which is block-in-inline (invalid
HTML5) and triggers a browser layout quirk that off-aligns
descendants — model IDs floated to the center under their labels
in ModelPicker, descriptions indented unevenly under levels in
ThinkingPicker. Replace the inner `<div>`s with `<span block
text-left>` so the layout is deterministic across rows.
3. **Visual polish in Thinking picker**
Label was `font-medium` at the parent's default `text-sm` (14px),
chunky next to the 10px description. Drop to `text-[13px]`, bump
description to `text-[11px] leading-snug` with `mt-0.5` so the
contrast between rows feels less jarring.
4. **Match Model picker's row typography to Thinking's**
Same `text-[13px]` for label + `text-[10px] mt-0.5` for the model
ID. Both pickers now read as the same component family.
5. **"Default" semantics: follow CLI config, not model factory default**
The chip displayed "Default" / "default" badge when no
`thinking_level` was set, alongside a `[default]` chip on the
model's factory-advertised default option in the menu. That was
misleading: when Multica omits `--effort` (because picker is
unset), it's the user's *local CLI config* (claude/codex) that
decides the reasoning level — not the model's factory default.
Showing "medium [default]" while the user has xhigh in their CLI
config lies about what actually fires at the API.
- Trigger label: "Default" → "Follow CLI config" (zh: "跟随 CLI 配置")
- Footer clear button: "Use model default" → "Follow CLI config"
- Footer tooltip: explicitly mentions claude/codex CLI config
- Inline `[default]` badge on the factory-default option: removed
- `defaultLevel` prop chain (picker + prop-row + test): cleaned up
as now-dead code
6. **Stop hiding the Thinking row while discovery loads**
`if (levels.length === 0 && !value) return null` hid the row
while the runtime-models query was still in flight, which
subscribed-then-unsubscribed from useQuery in such a way that
the discovery only fired when the user manually opened the Model
picker. Gate the early return on `!isLoading && !isFetching` so
ThinkingPropRow stays mounted (and thus its useQuery keeps
subscribed) until discovery returns; row appears as soon as
data arrives, no Model-picker tap required.
7. **Drop the inline tooltip on Thinking picker items**
The same description was rendered both inline under the label
(always visible) and as a hover tooltip (overlapping the next
row). The hover bubble was redundant — removed.
Tests
- `pnpm --filter @multica/views test thinking-picker` → 7/7 pass after
renaming the "Default" assertion + clearing the unused defaultLevel
test prop.
- `pnpm --filter @multica/views typecheck` clean.
* fix(test): align thinking-prop-row tests with renamed copy + loading-aware row gate
CI surfaced 3 broken assertions in `thinking-prop-row.test.tsx` —
all consequences of the polish PR's behaviour changes that the test
file hadn't tracked:
- "hides the row when ... no thinking levels and nothing is persisted"
The row now stays mounted while runtime-models discovery is in
flight (so the useQuery subscription actually survives long enough
to issue the request — fixes the bug where Thinking only appeared
after manually opening the Model picker). The assertion asserted
absence only after `initiate` was called, but loading is still in
progress at that point. Wrap the absence assertion in `waitFor`
so it waits for the row to disappear after the query settles.
- "clears the orphan value via the picker footer"
Tooltip copy changed from "Clear and fall back to this model's
default reasoning level" → "Clear the override and let the local
CLI config decide the reasoning level". Update the regex.
- "renders the row with \"Default\" when value is empty"
Trigger label changed from "Default" → "Follow CLI config" to
reflect that Multica omits --effort and the local CLI config
decides. Update the assertion + test name.
`pnpm --filter @multica/views test` → 701/701 pass.
* fix(agent-inspector): drop loading-row gate + per-field optimistic rollback (MUL-2339)
Addressing review feedback on #2919:
- ThinkingPropRow no longer keeps the row visible during discovery.
The previous explanation ("early return null aborts the useQuery
subscription") was wrong — React doesn't unmount a component that
returns null, so hooks (and their subscriptions) stay live. The
loading-aware gate only succeeded in showing an empty "Follow CLI
config" row that opened to an empty menu before discovery settled.
Restore the simple `levels empty && !value -> null` behavior; the
sibling ModelPicker mounts unconditionally and keeps the shared
runtime-models query active regardless.
- AgentDetailPage.handleUpdate now rolls back only the fields the
failing PATCH wrote, instead of restoring a whole-list snapshot.
A whole-list snapshot rollback discards any concurrent successful
inspector mutation that landed between snapshot and rollback. Per-
field rollback + a final invalidate converges the cache on server
truth without clobbering unrelated optimistic writes.
- Sync the now-stale "use model/runtime default" wording in the
thinking-related JSDoc and type comments: empty thinking_level is a
"no override" sentinel — the backend omits --effort and the upstream
CLI config decides — not a Multica-known default level.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
- Capture `brew tap` output and print the same diagnostic tail on
failure that `brew install` already prints, so #2867-style "no
signal" reports are gone from both Homebrew failure paths.
- Add a `brew tap` failure regression case to `scripts/install.test.sh`
and refactor the test runner to share sandbox/curl-stub setup; both
cases now also assert the diagnostic tail is emitted.
- Move the shell installer test out of the heavy backend job into a
dedicated `installer` matrix job that runs on `ubuntu-latest` and
`macos-latest`, since the installer targets macOS/Homebrew and BSD vs
GNU `tar` / `sed` / `mktemp` differences are the next likely break.
- Surface `MULTICA_INSTALL_DIR`, `MULTICA_BIN_DIR`, and
`MULTICA_SELFHOST_REF` in `install.sh --help` so `MULTICA_BIN_DIR`
stops looking like a test-only knob.
Co-authored-by: multica-agent <github@multica.ai>
* feat(desktop): pin tab — keep parked tabs anchored across navigations (MUL-2449)
Adds tab pinning to the desktop tab bar. Pinned tabs render as icon-only at
the left, suppress the X close button, and intercept any `navigation.push()`
that would change their pathname — those are redirected into a new tab so
the pinned tab stays parked on its original route. Search/hash/back/forward
stay in-tab so pinned filter and drawer state still work.
Implements the FINAL combo from the MUL-2449 RFC §4: right-click menu +
⌘⇧P shortcut (D1 a+c), icon-only visual (D1v i), pathname-change → new tab
with same-path-allowed (D2a/b A), back / refresh allowed (D2c/d A), pinned
auto-cluster left and persist (D3a/b A), pinned can't be X-closed (D3c A),
dedupe respected (D4a A), default Issues tab pinnable (D4b A), drag clamped
to its zone (D4c A), deep link prefers pinned (D4e A).
Store changes:
- Tab.pinned added; togglePin maintains the "pinned first" invariant by
inserting at the zone boundary.
- moveTab clamps cross-zone drags so dnd-kit can't violate the ordering.
- Persistence bumped v2 → v3 with a defaulting migration (pinned=false).
Rehydrate sorts pinned-first as a defensive net.
Navigation:
- tryRouteToPinnedNewTab compares the active tab router's live pathname
to the target. Same-pathname push (query / hash / sub-router) falls
through to the router; different pathname → openTab + setActiveTab
(foreground; respects dedupe).
UI:
- Tab bar wraps each tab in a shadcn ContextMenu with Pin/Unpin + Close
(Close disabled for pinned or last-remaining tab).
- Pinned tabs use a narrower icon-only layout with an accent left border
and a divider between the pinned and unpinned groups.
- Global keydown listener registers ⌘⇧P / Ctrl+Shift+P to toggle pin on
the active tab.
Tests: - tab-store: togglePin ordering, moveTab boundary clamping, v2→v3
migration.
- navigation: pinned push → new foreground tab; same-pathname push stays
in tab; cross-workspace still wins over pin.
Co-authored-by: multica-agent <github@multica.ai>
* test(desktop): cover TabNavigationProvider.push pin interception (MUL-2449)
Add pathname-diff / same-pathname cases for the per-tab navigation
adapter. Existing tests only exercised the root-level
DesktopNavigationProvider, but in-tab AppLink / page clicks flow
through TabNavigationProvider — so a future refactor that drops the
pin check from that provider would silently regress.
Co-authored-by: multica-agent <github@multica.ai>
* refactor(desktop): pin tab — hover button, full title, drop ⌘⇧P (MUL-2449)
Jiayuan's interactive review of PR #2914 surfaced three changes to the
RFC's D1 (entry / visual) decisions:
1. Drop the ⌘⇧P global shortcut — it added a keybinding for a
low-frequency action and crowded the shortcut namespace.
2. Reveal a Pin / Unpin button on tab hover instead of relying on the
right-click menu as the primary entry; right-click remains as a
fallback (and for Close).
3. Pinned tabs keep their full title and width. The only weak visual
differences vs. unpinned tabs are the accent left border and the
suppressed X close button.
Removes the global keydown listener (no other doc / handler referenced
it). Adds a hover-only Pin / Unpin span next to the existing close
affordance, both gated by group-hover. Drops the icon-only width /
hidden-title styling for pinned tabs.
Tests: new tab-bar.test.tsx covers Pin / Unpin button rendering, click
handlers (togglePin), the hidden-X invariant on pinned tabs, and the
full-title rendering. 146 passed, typecheck clean.
Co-authored-by: multica-agent <github@multica.ai>
* refactor(desktop): pin tab — drop accent left border, swap leading icon to Pin (MUL-2449)
Jiayuan reported that the accent left border on pinned tabs reads as a
heavy black edge in light mode and looks unrefined. Replace it with a
quieter identifier: pinned tabs swap their route icon for a Pin glyph
in the leading slot (same size, no extra horizontal space). The hidden
X close button stays as the secondary cue. RFC §3 D1v moves from
iii FINAL to iv FINAL; iii is demoted to v2 FINAL → v3 REMOVED.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Lambda <lambda@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
* feat(agent): inspector picker for thinking_level (MUL-2339)
PR1 (#2865) shipped the backend — column, daemon-side discovery,
Claude/Codex injection, API validation — but the agent detail inspector
had no UI to set the value. Users could only configure thinking_level
via custom_env / API. This wires up the picker so it lives next to
Runtime and Model where everything else editable already lives.
Picker is per-(runtime, model): it reuses the same `runtimeModelsOptions`
query the Model picker already runs (60s cache, no extra round-trip)
and reads the active model's `thinking.supported_levels`. When the list
is empty — every provider except Claude/Codex today, or a Claude model
that doesn't expose `--effort` — the entire PropRow is hidden, not just
rendered inert. The picker never gets to invent value/label pairs
itself; they come verbatim from each CLI's own catalog (`Low`,
`Extra high`, …) so the user sees exactly what `claude --effort` /
`/effort` and Codex's TUI show.
The `default_level` from the catalog is badged inside the popover so
the user knows which value `""` (the persisted "use model default"
sentinel) maps to. The clear footer sends `""` explicitly, which the
backend already understands as the tri-state "explicit clear" branch
of UpdateAgent. Invalid combinations (e.g. picking a value not in the
target provider's enum after a runtime swap in the same PATCH) hit
the existing 400 path on the server and surface as a toast via the
inspector's standard `onUpdate` error handler — no extra client-side
guard needed.
Exports `RuntimeModelThinking` and `RuntimeModelThinkingLevel` from
`@multica/core/types` so views consumers can refer to them by name.
i18n keys added in EN and zh-Hans (parity test green).
Co-authored-by: multica-agent <github@multica.ai>
* fix(agent): preserve unknown thinking_level in picker label
Stale persisted values (model swap, CLI catalog shrink) used to render
as 'Default' even though the backend would still ship the orphaned
token. Fall back to the raw value when no entry matches so the user
sees what's actually saved and can clear it.
Co-authored-by: multica-agent <github@multica.ai>
* test(agent): unit tests for thinking-picker label + clear flow
Covers the default-vs-set trigger label, the unknown-token preservation
path added in 3452fae3f, the read-only display, picking and re-picking
into onChange, and the clear footer's empty-string emission.
Co-authored-by: multica-agent <github@multica.ai>
* fix(agent): keep Thinking row visible when value is stale (MUL-2339)
Inspector was hiding the row whenever the active model had no
supported_levels, which also hid persisted orphan tokens (model swap
into a non-thinking runtime, or a CLI catalog that shrank). PR1's
per-model invalid behavior is daemon-side warn/drop, not a synchronous
DB clear, so the frontend has to surface the raw value and let the
user explicit-clear it via the picker footer.
Render the row when levels are empty AND value is empty; otherwise
keep it. Extract ThinkingPropRow into its own file so the row-level
logic is unit-testable.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* feat(agent): persist thinking_level per agent (MUL-2339)
Adds a nullable `thinking_level` column to the `agent` table so the
backend can route a runtime-native reasoning/effort token (e.g. Claude's
`xhigh`, Codex's `minimal`) through to the agent CLI on every dispatch.
The column is intentionally TEXT rather than an enum — Claude and Codex
publish overlapping but distinct vocabularies and we want the persisted
value to round-trip exactly through whichever CLI receives it. NULL is
the "use runtime default" sentinel that every downstream consumer reads
as "do not inject --effort / reasoning_effort".
This commit is just the storage layer (migration + sqlc); subsequent
commits wire it through the API, daemon, and agent backends.
Co-authored-by: multica-agent <github@multica.ai>
* feat(agent-backend): inject reasoning effort for claude + codex (MUL-2339)
Extends ExecOptions with a runtime-native ThinkingLevel string and wires
it into the Claude and Codex backends. Discovery is driven by the local
CLI so the daemon advertises whatever the host install supports rather
than a hand-maintained list that goes stale.
Per Elon's PR1 review:
- Claude: parses `claude --help` to learn the `--effort` superset and
projects through a per-model allow-list (xhigh is Opus-only; max is
session-only on the smaller models). Falls back to a conservative
static list when the binary is missing or help drift hides the line.
- Codex: drives `codex debug models --output json` so per-model
reasoning subsets and the documented default come directly from the
CLI. The older config-error probe trick is gone — the JSON path is
stable and doesn't pollute stderr with an intentional misconfig.
- Cache key includes (provider, executablePath, cliVersion) so a CLI
upgrade invalidates entries that referenced the older help / catalog.
Per Trump's PR1 constraint, all three Codex injection points
(thread/start.config, thread/resume.config, turn/start.effort) flow
through one helper (`applyCodexReasoningEffort`) so they cannot drift
independently. The shared `codexReasoningCases` fixture in
`thinking_test.go` asserts the same value→{shape, key} contract at
each site for every level the runtimes know about.
Claude's `--effort` is also added to `claudeBlockedArgs` so a user
custom_args entry can't silently outvote the daemon-injected value.
Co-authored-by: multica-agent <github@multica.ai>
* feat(api): wire thinking_level through API + daemon contract (MUL-2339)
End-to-end plumbing for the per-agent reasoning/effort setting:
- AgentResponse / TaskAgentData now carry `thinking_level`; the daemon's
claim response includes it and the daemon's executor passes it through
to agent.ExecOptions, where the Claude and Codex backends already know
what to do with it.
- ModelEntry on the runtime-models wire format gains a `thinking` block
carrying `supported_levels` + `default_level` per model so the UI can
render a runtime-aware picker without the server having to know about
the local CLI install. `handleModelList` projects the agent-package
catalog (including the new Thinking field) into the wire shape.
- CreateAgent / UpdateAgent gate the field with a synchronous provider
enum check (claude / codex only today). UpdateAgent is tri-state:
field omitted = no change, "" = explicit clear (new
`ClearAgentThinkingLevel` query, mirrors the existing mcp_config null
pattern), non-empty = validate then set.
Per Trump's PR1 review, the API NEVER auto-clears on a runtime/model
swap and ALWAYS returns 400 on an unknown literal value — same shape
across CreateAgent, UpdateAgent, and combined patches that move
runtime + level in one request. Per-model combination failures (e.g.
`xhigh` against a model that only supports up to `high`) surface as a
daemon-side task error, not a silent server-side rewrite.
TS types follow the same shape: `Agent.thinking_level`,
`CreateAgentRequest`/`UpdateAgentRequest` add the field, `RuntimeModel`
grows a `thinking` block. Older backends omit the field, which the
front-end treats as "no picker for this model" — installed desktop
builds keep working.
Co-authored-by: multica-agent <github@multica.ai>
* fix(agent): correct codex debug models argv + pin via runner test (MUL-2339)
`codex debug models --output json` is rejected by codex-cli 0.131.0 —
the subcommand emits JSON on stdout by default and has no `--output`
flag. Drop the flag and add `--bundled` to skip the network refresh
discovery doesn't need. Move the argv to a package-level var and add
a test that runs a fake `codex` to assert the binary actually
receives exactly `debug models --bundled`, so the contract can't
silently drift on the next refactor.
Also teach ValidateThinkingLevel to resolve an empty model to the
provider's default model entry. Without this, every default-model
task with a persisted thinking_level would be misjudged "unknown
model" by the daemon guard.
Co-authored-by: multica-agent <github@multica.ai>
* fix(api): reject runtime switch that would leave invalid thinking_level (MUL-2339)
A PATCH that changed `runtime_id` without touching `thinking_level`
used to silently keep the existing value, so a Claude agent storing
`max` could land on a Codex runtime where `max` is not a recognised
token at all, and the daemon would receive a literal-invalid level.
Hold the same "always 400 on literal-invalid, never silent coerce"
rule on this implicit path. When runtime_id changes and the existing
value is not in the new provider's enum, return 400 with the
recovery options (clear via `thinking_level=""` or re-set in the
same PATCH).
Add coverage for both the kept-when-still-valid and the rejected
cases, plus the two recovery paths (clear and replace).
Co-authored-by: multica-agent <github@multica.ai>
* fix(daemon): guard runTask with per-model thinking_level validator (MUL-2339)
ValidateThinkingLevel existed but had no call site — `task.Agent.
ThinkingLevel` flowed straight into ExecOptions, so `xhigh` configured
on a non-Opus Claude model, or API-side stale values that escaped the
provider enum gate, would be injected anyway.
Run the validator before building ExecOptions. Invalid combinations
log a warning and drop the level instead of failing the task: the
agent still runs, just at the runtime's default reasoning effort.
Discovery errors fail open (keep the level, let the CLI surface any
objection) so a transient `claude --help` failure can't strand work.
Empty model is forwarded as-is; the validator resolves it to the
provider's default model internally per the cross-package contract.
Co-authored-by: multica-agent <github@multica.ai>
* chore(agent): drop stale `--output json` comments + unused scanner (MUL-2339)
Codex CLI's `debug models` subcommand emits JSON without an `--output`
flag, and `parseCodexDebugModels` never read from the bufio.Scanner.
Sync the comments with the actual invocation and remove the dead init.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
* fix(runtimes): use official Gemini spark icon (MUL-2447)
Gemini provider was falling through to the default Monitor icon in the
runtime list. Add the official 4-point spark mark with Google's
blue → purple → pink gradient, matching the SVG style/sizing of the
other provider icons.
Co-authored-by: multica-agent <github@multica.ai>
* fix(runtimes): use current Gemini multicolor spark gradient (MUL-2447)
Per review on PR #2904: the previous 3-stop blue/purple/pink gradient
was the legacy Bard-era Gemini spark. Update to the 5-stop cyan → blue
→ purple → pink → orange gradient used by the current Gemini app/web
multicolor mark.
Co-authored-by: multica-agent <github@multica.ai>
* fix(runtimes): switch Gemini icon to aurora multicolor treatment (MUL-2447)
Co-authored-by: multica-agent <github@multica.ai>
* fix(runtimes): align Gemini aurora color positions and smooth spark path
Swap yellow/green radial gradient anchors so colors land at the official
positions: top red / right blue / left yellow / bottom green, matching
gemini.google.com's current aurora spark. Replace the arc-based 4-point
spark outline with a cubic-bezier version normalized to the 24-viewBox
so the inset between tips is smoother and closer to the gstatic source.
Co-authored-by: multica-agent <github@multica.ai>
* fix(runtimes): use Simple Icons Google Gemini mark (MUL-2447)
Drop the hand-crafted aurora gradient approximation and inline the
canonical "Google Gemini" path from Simple Icons (CC0 1.0), rendered
in the Simple Icons brand color (#8E75B2). This matches the pattern
used by the other provider marks in this file (Claude/Codex from
Bootstrap Icons, etc.) instead of trying to manually approximate the
official multicolor wash from gemini.google.com (which paints via a
clipPath over an embedded raster).
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Lambda <lambda@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
The Start button lives in `DaemonRuntimeActions`, which is rendered in
the per-machine detail pane and only when the selected machine is
flagged `isCurrent`. After the user manually stopped the daemon,
`status.daemonId` went back to undefined, so no machine could be
matched as `isCurrent` — the local row either disappeared (when the
server-side runtime had been GC'd) or moved into the "remote" section
(when it was still present but unmatched). Either way the Start button
was unreachable until the app was restarted.
Two-part fix:
- `DesktopRuntimesPage` now caches the last-known daemonId/deviceName
so the local match keeps working while the runtime is still on the
server (recently_lost / offline window).
- `buildRuntimeMachines` accepts an `ensureLocalMachine` flag; when no
real runtime matches, a placeholder local row is synthesized so the
Start button still has a home. Desktop opts in via a new
`hasLocalMachine` prop on `RuntimesPage`. The empty state is also
suppressed when this prop is set so the placeholder row isn't hidden
behind the "register a runtime" hint on first launch.
Co-authored-by: Lambda <lambda@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
* feat(autopilot): support assigning autopilot to a squad (MUL-2429)
Path A (Squad-as-Leader) from the RFC: when an autopilot's assignee is a
squad, dispatch resolves to squad.leader_id and executes against the
leader's runtime — semantics match a human manually assigning the issue
to that squad, no fan-out.
Backend scope only; frontend picker change is a follow-up PR.
Changes:
- 096_autopilot_squad_assignee migration: drop agent FK on
autopilot.assignee_id, add assignee_type column (default 'agent'),
add autopilot_run.squad_id attribution column.
- service.AgentReadiness: single source of truth for archived /
runtime-bound / runtime-online checks. Shared by autopilot
admission gate, run_only dispatch, and isSquadLeaderReady.
- service.resolveAutopilotLeader: translates assignee_type/id to the
agent that actually runs the work.
- dispatchCreateIssue: stamps issue with assignee_type='squad' for
squad autopilots and enqueues via EnqueueTaskForSquadLeader.
- dispatchRunOnly: belt-and-braces readiness re-check after resolving
squad → leader so a leader that went offline between admission and
dispatch produces a clean failure instead of a doomed task.
- handler.CreateAutopilot / UpdateAutopilot: accept assignee_type with
squad/agent existence + leader-archived validation. Backward-compatible
default of "agent" preserves the contract for older clients.
- Analytics: AutopilotRunStarted/Completed/Failed events carry
assignee_type and squad_id; PostHog can now group autopilot runs by
squad without joining back to the autopilot row.
Co-authored-by: multica-agent <github@multica.ai>
* fix(autopilot): reject archived squads, route post-admission skips, cleanup dangling-agent autopilots (MUL-2429)
Addresses three review findings on PR #2888:
1. Archived squad handling: validateAutopilotAssignee now rejects squads
with archived_at set; resolveAutopilotLeader returns errSquadArchived
so the admission gate fails closed; DeleteSquad now mirrors the issue
transfer for autopilot rows (TransferSquadAutopilotsToLeader) so
surviving autopilots flip to assignee_type='agent' (leader) instead
of dangling at the archived squad.
2. dispatchRunOnly post-admission readiness: introduces errDispatchSkipped
sentinel, recognised by DispatchAutopilot via handleDispatchSkip so
the run is recorded as `skipped` (not `failed`). Manual triggers no
longer 500 when the leader's runtime goes offline between admission
and task creation. New TestManualTriggerDoesNotErrorOnPostAdmissionSkip
locks the behaviour in.
3. Dangling agent assignee after migration 096 dropped the FK:
shouldSkipDispatch now distinguishes pgx.ErrNoRows / errSquadArchived
(hard skip — retrying won't help) from transient DB errors
(fail-open). DeleteAgentRuntime pauses autopilots that target agents
about to be hard-deleted (ListArchivedAgentIDsByRuntime +
PauseAutopilotsByAgentAssignees) so the breakage surfaces as a paused
row in the UI instead of a quiet skip-burning loop.
Unit tests cover the sentinel unwrap contract and errSquadArchived
errors.Is behaviour. Integration test
TestAutopilotDispatchSkipsWhenRuntimeOffline re-verified against a fresh
DB with migration 096 applied.
Co-authored-by: multica-agent <github@multica.ai>
* fix(autopilot): bump last_run_at on post-admission skip (MUL-2429)
Match recordSkippedRun (pre-flight skip) and the success path so the
scheduler / "last seen" UI both reflect that this tick evaluated the
trigger, even when the post-admission readiness gate caught a late
regression.
Addresses Emacs review caveat #1 on PR #2888.
Co-authored-by: multica-agent <github@multica.ai>
* feat(autopilot): mixed agent/squad assignee picker in dialog (MUL-2429)
End-to-end UI for assigning an autopilot to a squad. Closes the PR #2888
backend gap: the squad-as-assignee feature was already wired in Go (Path A,
RFC §4) but the desktop dialog never offered the choice.
- core/types/autopilot: add `AutopilotAssigneeType`, surface
`assignee_type` on `Autopilot` + Create/Update request payloads.
- views/autopilots/pickers/agent-picker: switch to a polymorphic
AssigneeSelection (`{type, id}`); render agents and squads as two
grouped sections with shared pinyin search.
- views/autopilots/autopilot-dialog: maintain `assigneeType` state, send
it on create/update, render the trigger avatar / hover dot with
`assignee.type`.
- views/autopilots/autopilots-page + autopilot-detail-page: render the
assignee row using `autopilot.assignee_type` so squad-typed autopilots
show the squad avatar + name, not a broken agent lookup.
- locales: add `agents_group` / `squads_group` / `select_assignee` keys
(en + zh-Hans), keep legacy `select_agent` for callers that still
reference it.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Lambda <lambda@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
Wires the frontend half of the read-only RFC. The Settings → GitHub tab
now always issues the installation list query for any workspace member
(the backend gates it via `RequireWorkspaceMember` after PR #2886) and
gets `can_manage` straight from the API response. The render matrix
covers the six cases the RFC calls out:
- configured + connected + admin → Disconnect + (optional) Connected by
- configured + connected + member → read-only "Connected to" + read_only_hint
- configured + not connected + admin → Connect button + dev description
- configured + not connected + member → contact_admin_to_connect hint
- not configured + admin → operator banner + disabled Connect
- not configured + member → contact_admin_to_connect hint
New i18n keys (en + zh-Hans): read_only_hint, connected_by, contact_admin_to_connect.
The unused github.manage_hint string is removed (its non-admin branch
now resolves to one of the two new hints depending on connection state).
GitHubInstallation gains an optional `connected_by` display name so the
UI can render the "Connected by {name}" line without further changes
once the backend exposes the field.
Co-authored-by: Lambda <lambda@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
* feat(github): expose read-only installation list to workspace members (MUL-2413)
Relax `GET /api/workspaces/{id}/github/installations` from owner/admin-only
to any workspace member so the Settings → Integrations tab no longer renders
blank for non-admins (the original symptom of MUL-2413).
The handler now reads the caller's role from the workspace middleware:
- owner / admin keep the full row including the numeric `installation_id`
(the connect / disconnect handle) and receive `can_manage: true`.
- every other role (member / guest) receives rows with `installation_id`
omitted and `can_manage: false`, giving them visibility into "is GitHub
wired up?" without the management handle.
`GET /github/connect` and `DELETE /github/installations/{id}` stay under
the admin/owner middleware group — this PR only relaxes the read path.
Tests: `TestListGitHubInstallations_RoleGating` exercises admin, owner,
member, and guest paths against the real DB-backed handler fixture and
asserts the field stripping + `can_manage` contract.
Refs: MUL-2413
Co-authored-by: multica-agent <github@multica.ai>
* fix(github): redact installation_id from realtime broadcasts (MUL-2413)
GET /github/installations strips the numeric installation_id for non-admin
members, but the github_installation:created / uninstall / suspend WS
events were still publishing it, so the same handle was reachable from
any workspace client subscribed to the workspace scope. Broadcast both
payload variants without it — the frontend uses these events only to
invalidate the installations query, so admins re-query the list endpoint
to recover the management handle.
Also adds a router-level test that mounts the production middleware split
(member-visible list vs. owner/admin connect+delete) so a future routing
change can't silently widen the write surface.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Lambda <lambda@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
In the attachment preview modal, image and video previews used
`max-h-full max-w-full`, which let small assets render at their
natural size and leave the modal mostly empty. Switch to
`h-full w-full` so the preview always occupies the modal viewport,
relying on `object-contain` to preserve aspect ratio without
upscaling beyond the intrinsic bounds.
Only touches `packages/views/editor/attachment-preview-modal.tsx`
for the image (line 355) and video (line 373) branches; pdf, audio,
markdown, html, and text branches keep their existing layout.
Co-authored-by: multica-agent <github@multica.ai>
Tiptap's stock ListItem keymap binds Enter only to splitListItem. When the
cursor sits in an empty top-level list item, splitListItem returns false
(without dispatching) with a code comment saying "let next command handle
lifting" — but no next command is chained. Enter then falls through to
ProseMirror's baseKeymap which inserts another empty paragraph inside the
list item, trapping the user.
Replace StarterKit's ListItem with PatchedListItem whose Enter binding
chains splitListItem → liftListItem via commands.first. The lift fallback
only runs when splitListItem returns false (top-level empty case),
restoring the standard "double-Enter exits the list" behaviour seen in
every other rich-text editor. Non-empty and nested-empty items are
unaffected because splitListItem already handles them correctly.
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
## Summary
Adds per-user `profile_description` so coding agents have cheap, durable context about who is asking. v1 per the brief Xeon locked in on [MUL-2406](mention://issue/63a7247c-4f6a-42cf-90d1-7c746e77158a):
- **DB** — `user.profile_description TEXT NOT NULL DEFAULT ''` (migration 096). 2000-rune cap enforced server-side. No nullable / privacy state to manage.
- **API** — `PATCH /api/me` accepts the field; `UserResponse` always emits it. Client wraps `updateMe` in a lenient `UserSchema` + `EMPTY_USER` fallback per CLAUDE.md API Response Compatibility.
- **UI** — Settings → Account gains an "About you" textarea with live `n/2000` counter, `maxLength` guard, and a localized too-long error (EN + zh-Hans).
- **CLI** — `multica user profile get` / `multica user profile update` with `--description / --description-stdin / --description-file / --clear`, mirroring the existing `issue comment add` input-mode menu.
- **Daemon injection** — claim handler resolves the runtime owner and stamps `requesting_user_name` + `requesting_user_profile_description` on the task. `buildMetaSkillContent` emits `## Requesting User` between `## Agent Identity` and `## Available Commands`, blockquoted and framed as background context. The block is omitted entirely when the description is empty (no token cost when unused).
Brief is written **once per task** via `CLAUDE.md` / `AGENTS.md`, not the per-turn prompt — same path the agent already reads for identity, so no extra per-turn cost.
## Test plan
- [x] `go build ./...`, `go vet ./...`, `go test ./internal/cli/ ./internal/daemon/ ./internal/daemon/execenv/ ./cmd/multica/`
- [x] New brief tests: `TestBuildMetaSkillContentEmitsRequestingUser`, `TestBuildMetaSkillContentOmitsRequestingUserWhenEmpty`
- [x] `pnpm typecheck`, `pnpm lint`, `pnpm test` (74 files, 644 tests pass)
- [ ] Handler DB tests (`TestUpdateMe*`) require a migrated test DB — not runnable in this sandbox
- [ ] Manual: open Settings → Account, set a description, confirm the next daemon-run agent's `CLAUDE.md` shows `## Requesting User`
* feat(issues): show project segment in issue breadcrumb (MUL-2422)
Render the issue's project (when present) between the workspace and any
parent-issue segment. Segment reflects the issue's own `project_id` so
the same URL produces the same breadcrumb from every entry point.
Failed/missing project queries fall back to an "Unknown project"
placeholder; loading shows a skeleton to avoid layout shift.
Co-authored-by: multica-agent <github@multica.ai>
* fix(issues): cap project breadcrumb width to preserve title precedence
Constrain Project crumb to max-w-72 (matching ProjectChip) and add
min-w-0 to the title span so the flex compression order matches RFC
§5/§9: Project/Parent shrink before the current Issue title.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Lambda <lambda@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
* refactor(runtimes): declutter the runtimes page (MUL-2407)
Cuts visual noise on the Runtimes detail view without removing real
information:
- MachineDetail: drop the 4-card metric grid (RUNTIMES / HEALTH /
WORKLOAD / CLI) and replace it with a single inline meta strip. The
cards repeated what the title chip and runtime rows already show.
- PageHeaderBar: remove the inline tagline + "Learn more" link. The
header is now icon + title + count + connect button.
- VisibilityBadge: only render the Public chip. Private is the default,
so a row of `🔒 Private` badges was pure noise.
- CliCell: drop the per-row "Desktop" managed badge — the same string on
every desktop row carried near-zero information.
- MachineSidebar row: hide the truncated daemon-id subtitle. The id is
still available on hover via `title` and remains visible in the
detail header.
Co-authored-by: multica-agent <github@multica.ai>
* fix(runtimes): address review feedback on inline meta and hover title
- Inline meta now reads "6 runtimes · 5 online" instead of "6 6 online"
by using runtime_count for the total label.
- Sidebar machine title hover now shows full daemon id (with subtitle
fallback) so the daemon id is recoverable after the sub-row was hidden.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
320px was too cramped for typical rendered HTML (charts, dashboards,
formatted documents). Matches the existing HTML attachment preview
height for visual consistency across both iframe surfaces.
Co-authored-by: Lambda <lambda@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
* refactor(onboarding): remove starter-content kit, unify install-runtime issue across mark-onboarded paths (MUL-2438)
Drops the post-onboarding ImportStarterContent / DismissStarterContent
flow (handler + routes + StarterContentPrompt + templates + locale
strings + analytics event). The bug — web onboarding seeding 6+ starter
issues without a runtime — only existed through that path; with it gone
the source disappears.
The "install a runtime" issue from BootstrapOnboardingNoRuntime is now
the canonical no-runtime onboarding seed. The title/description and a
LockAndFindActiveDuplicate-deduped seeder move to
handler/no_runtime_issue.go, and CompleteOnboarding / CreateWorkspace /
AcceptInvitation seed it whenever the workspace has no runtime yet, so
every mark-onboarded entry point lands the user on a concrete next
step.
starter_content_state column is kept and continues to be claimed as
'imported' in all five entry points so older desktop builds (which
still render the legacy dialog on NULL) don't surface it to accounts
created after this change.
Co-authored-by: multica-agent <github@multica.ai>
* fix(onboarding): backfill starter_content_state for in-window NULL users (MUL-2438)
054 only covered pre-feature users. Anyone onboarded between then and the
starter-content kit removal could still sit at NULL, and old desktop
clients gate the legacy StarterContentPrompt on `starter_content_state
IS NULL`. The import/dismiss routes are gone, so leaving these rows NULL
would surface a dialog whose buttons 404. Mark them 'imported' to match
the new helper's claim semantics.
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: Lambda <lambda@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
Add optional `opts.activate` to NavigationAdapter.openInNewTab. Default
stays `false` so cmd/ctrl+click on links/mentions keeps browser-style
background semantics. The two explicit toolbar entry points
(attachment-preview-modal, html-attachment-preview) opt in with
`{ activate: true }` so the new tab gains focus after the modal closes.
Both desktop providers (root + per-tab) now use the tab id returned by
`store.openTab` to call `setActiveTab` only when `activate` is true.
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
- Drop `workspace current`; `workspace get` (no args) already prints the
current default workspace, so the two were doing the same thing.
- Rename `workspace members` to `workspace member list` to free up the
`member` namespace for future `add` / `remove` subcommands and align
with the rest of the CLI's `<resource> <verb>` shape.
- Add `--full-id` to `workspace list`, matching `project list`,
`autopilot list`, and friends.
Docs and the daemon prompt are updated to match.
Co-authored-by: multica-agent <github@multica.ai>
* fix(openclaw): parse whole buffer instead of line-by-line scanner
Follow-up to c87d7676 (WOR-10). The stdout/stderr swap fixed the dominant
case but `processOutput` still scanned line-by-line and only attempted a
whole-buffer parse from a fragile fallback path. Pretty-printed JSON
(openclaw 2026.5.x emits the result blob indented across many lines) made
every individual line unparseable on its own — `{`, ` "payloads": [`,
` {`, etc. — so the success path hinged entirely on the fallback
joining `rawLines` and re-trying.
Under load (daemon restarts racing the close-on-cancel goroutine, partial
chunked reads when stdout closes mid-flight) the line scanner could see
truncated input that never reassembled into valid JSON, surfacing
"openclaw returned no parseable output" against runs where the agent had
in fact completed the work and posted comments. Roughly 30–40% of recent
runs in v0.2.27 logs hit this path; multica still wrote a `task_failed`
inbox row for each one even though the underlying issue had moved to
`in_review` or `done`.
The fix:
- processOutput now reads the full stdout buffer with `io.ReadAll` first.
- A new `parseWholeBufferOpenclawResult` helper attempts a single
`json.Unmarshal` against the entire buffer (after trimming, and after
optionally stripping leading non-JSON log lines). When it matches, we
build the result and return — the line scanner never runs.
- If the whole-buffer parse fails, we fall through to the existing NDJSON
line-by-line scanner. This preserves streaming-event support (kept for
forward compatibility and other backends) without leaving openclaw's
dominant pretty-printed shape at the mercy of timing.
- The failure path now emits a `(got N bytes; preview: ...)` suffix on
the canonical "no parseable output" error so future debugging isn't
blind. The exact canonical phrase is preserved for empty buffers so
existing dashboards / log-grep tooling keep matching.
Tests:
- TestOpenclawProcessOutputWholeBufferPrettyJSON: feeds a hand-crafted
multi-line indented blob (multiple payloads, nested agentMeta, usage
map) and asserts every field round-trips through the whole-buffer fast
path.
- TestOpenclawProcessOutputDeeplyIndentedFixture: re-runs the recorded
openclaw 2026.5.5 stdout fixture (1070 lines) directly through
parseWholeBufferOpenclawResult, asserting the bug-shape parses cleanly
on the first attempt without falling through to NDJSON scanning.
- TestOpenclawProcessOutputEmptyBufferErrorIncludesByteCount: tightens
the empty-buffer failure path, asserts the canonical phrase survives so
observability tooling keeps working.
All existing tests in the openclaw + buildOpenclawArgs suites stay green
(streaming NDJSON event tests, lifecycle tests, structured-error tests,
usage-field-variant tests). The two pre-existing flaky timeout-tight
codex tests (TestCodexExecuteSemanticInactivityAllowsContinuous*) fail on
both this branch and on c87d7676 baseline; they are unrelated and out of
scope here.
Co-authored-by: multica-agent <github@multica.ai>
* fix(openclaw): drop dead preview branch, document streaming regression
Rebase + review-fix follow-up on top of f27df2d9b.
processOutput's preview branch was unreachable: openclawNoParseableOutputError
was only called from the `!gotEvents && trimmed == ""` path, which by
construction means the entire scanned buffer collapsed to whitespace, so the
`(got N bytes; preview: ...)` formatter could never fire on a non-empty buffer.
Replace the helper with a single canonical-string constant (callsite is now
inline) and update the test name to match what it actually asserts (the
canonical empty-buffer error string is preserved for external log-grep /
dashboard consumers).
Also document on processOutput that the line-scanner path is no longer
truly streaming after the io.ReadAll switch: events accumulate until
stdout closes. OpenClaw 2026.5.x does not emit streaming events so this
regression is invisible today, but flag it for the next backend that
might.
Misc: switch the scanner's input source from
`strings.NewReader(string(buf))` to `bytes.NewReader(buf)` to drop one
unnecessary byte/string round-trip.
MUL-1908
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
Co-authored-by: J (Multica agent) <j@multica.local>
* feat(projects): scheduled-only Gantt data source + WS reactivity (MUL-1881)
Project Gantt now fetches its own scheduled-only data instead of riding the
Board/List pagination cache. The Unscheduled drawer and pagination warning
banner are gone, and any WS-driven issue change (create / update / delete)
invalidates the new cache so the timeline stays live.
- Backend: `GET /api/issues?scheduled=true` adds an
`(i.start_date IS NOT NULL OR i.due_date IS NOT NULL)` predicate on both
ListIssues and CountIssues. New SQL filter is plumbed through sqlc + handler.
- Frontend: new `projectGanttIssuesOptions(wsId, projectId)` issues a single
fetch and lives under its own cache key. WS handlers and mutations
invalidate the prefix on create/update/delete so the bar reacts to
start_date / due_date changes from other tabs and from this tab without
waiting on the WS round-trip.
- GanttView: drops the Unscheduled section, the pagination warning banner,
and the load-all button; renders only scheduled rows.
- Removes now-dead `useLoadAllRemaining`, `myIssueListPaginationOptions`,
`summarizeIssueListPagination`, and the gantt locale strings that
supported the old plumbing.
Co-authored-by: multica-agent <github@multica.ai>
* fix(projects): page through Gantt fetch and isolate per-view data sources
- Walk paginated `scheduled=true` issues until total is reached so projects
with more than 500 scheduled bars no longer silently truncate.
- Gantt mode disables the bucketed Board/List query and reads its own
scheduled cache for the project empty-state check, so the page never
short-circuits Gantt with a Board-derived "no issues" CTA.
- `onIssueLabelsChanged` patches matching rows in the Project Gantt cache
in-place, keeping label filters consistent after attach/detach from
other tabs or agents.
MUL-1881
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: multica-agent <github@multica.ai>
Earlier the unification commit dragged in a Tailwind override stack
(ring, rounded-md, transition-shadow, bg-background/95, button hover
classes) "to make standalone surfaces work without .rich-text-editor
scope". Because the legacy CSS rules were not removed, both layers
applied in the editor, producing a visible double-stroke selection
ring and a light-theme hover on top of the dark-glass toolbar.
This commit reverts the styling churn:
- ImageAttachmentView now emits the same span-only DOM as the original
ReadonlyImage: <span.image-node> > <span.image-figure> > <img.image-content>
+ <span.image-toolbar> with naked <button> children. No Tailwind tax.
- The `.image-*` rules in content-editor.css are de-scoped from
`.rich-text-editor` so the single set of styles also drives chat /
AttachmentList renders. Editor-only behavior (640px cap, NodeView
centering) stays under the `.rich-text-editor` scope.
- A `data-clickable` attribute carries the "this image is clickable
to preview" hint that the readonly cursor rule used to key off the
`.rich-text-editor.readonly` scope.
- ImageView NodeViewWrapper no longer adds its own `image-node` class
because `<Attachment>` already emits one; the duplicate was harmless
but redundant.
Visual: editor + readonly comments render identical to before. Chat /
AttachmentList previously rendered a gray file card for images (the
P0 fix in the parent commit) and now match the editor visual without
the heavy-handed Tailwind detour.
Tests: 98 attachment-related tests pass; full `pnpm typecheck` + `pnpm
test` (652 tests) green.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two independent root causes made "Open in new tab" on a desktop
attachment-preview modal feel like "the popup is still there and the
current tab got replaced":
1. `AttachmentPreviewModal.handleOpenInNewTab` never called `onClose()`,
so the modal stayed mounted over the new tab.
2. Both `DesktopNavigationProvider.openInNewTab` and
`TabNavigationProvider.openInNewTab` called
`store.setActiveTab(tabId)` after `store.openTab(...)`, which stole
focus to the new tab — violating the type contract
("Desktop only: open a path in a new background tab") and matching
neither Chrome's cmd+click default nor the user's expectation.
Fixes:
- Modal: always call `onClose()` after dispatching the navigation
(desktop adapter path and web `window.open` fallback path).
- Desktop navigation: drop the post-`openTab` `setActiveTab` call in both
providers. `openTab` already preserves `activeTabId` for new paths and
switches to the existing tab when the path is already open, which is
exactly the background-tab semantics the type contract advertises.
Tests:
- `attachment-preview-modal.test.tsx`: assert `onClose` is invoked on
both the desktop and web fallback branches.
- `pageview-tracker.test.tsx`: rename the "openInNewTab / addTab" case
so the comment no longer claims `openInNewTab` activates the new tab.
- New `apps/desktop/.../platform/navigation.test.tsx`: assert that
`openInNewTab` on both providers calls `openTab` and never
`setActiveTab` for same-workspace paths, and routes cross-workspace
paths through `switchWorkspace`.
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
HTML attachment previews mount the document inside a sandboxed
`<iframe srcdoc>` deliberately WITHOUT `allow-same-origin` — uploads are
untrusted user content. Chromium treats fragment-link clicks inside such an
opaque-origin srcdoc iframe as cross-origin frame navigation and silently
rejects them, so clicking a TOC entry never scrolls.
Append a tiny shim script to the srcdoc that intercepts `<a href="#...">`
clicks inside the iframe and calls `scrollIntoView` directly. The shim runs
in the iframe's own opaque origin under `allow-scripts` — no new
capabilities, no sandbox token changes; it cannot reach parent / cookies /
localStorage.
All three HTML attachment surfaces share the same helper:
- inline 480px card (html-attachment-preview.tsx)
- full-screen modal (attachment-preview-modal.tsx)
- full-page route (attachment-preview-page.tsx)
References: whatwg/html#3537, crbug 40191760.
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: multica-agent <github@multica.ai>
When the renderer crashes hard enough to leave a white window (React
boundary unrecoverable, syntax error during initial mount, preload
script throw), DevTools can't be opened and the only signal in the
`make dev` terminal is the daemon-manager 5s polling complaint
("Render frame was disposed before WebFrameMain could be accessed").
That's a downstream symptom — the actual JS error is unreachable, so
the user has no path to diagnose without restarting the renderer
(which loses the failure mode entirely).
Add four webContents listeners on the main BrowserWindow, gated by
`is.dev` so packaged builds keep their stderr clean:
- `console-message`: forwards every renderer `console.*` to main's
stderr with file:line. React error boundaries, `window.onerror`, and
unhandled-rejection handlers all surface here.
- `render-process-gone`: serialises the GoneDetails (`crashed` / `oom`
/ `killed` / `launch-failed`) so the user sees *why* the renderer
died, not just that it did.
- `did-fail-load`: catches loadURL/loadFile failures. Skip
`errorCode === -3 (ABORTED)` because that's the normal HMR-induced
navigation abort.
- `preload-error`: the one error class DevTools can never show, because
preload runs before the window owns a console. Without this listener
preload throws are invisible.
All output is prefixed with `[renderer <tag>]` so it's easy to grep
distinct from main's own logs.
No behavioural change in production: the entire block is inside an
`is.dev` guard. Packaged builds keep their existing stderr.
