Repository navigation
Conversation
Output of /release-plan (full mode) over v0.9.5..origin/dev (rc3, 15e3502): payload census across both trackers, release-floor (ent#649) scoring, five seam scans, the work order (6 MUST, 4 SHOULD), and draft Upgrade-notes + Known-limitations blocks for /release. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ze plan Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…6dde) One enterprise commit: Abilityai/trinity-enterprise#853, the enterprise half of Abilityai/trinity-enterprise#849 (the SSO setting "keep email-code sign-in available for users"). The OSS half, #3428, is already on dev; this bump ships the pair. Related to Abilityai/trinity-enterprise#849 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… (ent#856)
tabOrderFrom takes the arrival snapshot (Main first, recency, archived rows
ranked not pruned); agentChatTabs({order}) draws the strip in that frozen
order with born chats after Main by created_at; agentLanding gains arm 0
(most recent unread chat, via agentUnreadLanding) ahead of the last-open
arm; forgetLastOpen drops a closed chat from the session memory.
Requirement §5.43 added first. Shell wiring follows (A2).
Refs abilityai/trinity-enterprise#856
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…last-open (ent#856)
One watcher on the derived stageAgent (+ threadListTrusted) is the only
writer of the arrival snapshot tabOrder: take on a new agent, clear off
stage, retake once when an untaken (cold-link) snapshot becomes trusted.
The strip, cycleChat and closeChat's close-next all read the same order.
The /a/:name route watcher now passes lastOpen (landing arm 1) and
replaces the history entry it lands from, so Back does not loop.
forgetLastOpen runs on close and on a thread-missing 404.
Existing pins updated to the 2026-10-13 ruling: a route-driven landing
from a thread of the same agent returns to that thread (was: fresh), and
the openThread/landOnAgent signature pins carry the replace option.
Mutation: reverting openThread(row, { replace }) turns the Remount spec red.
Refs abilityai/trinity-enterprise#856
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… off stage (ent#856) portalChatCycle: a reply that re-sorts the thread list mid-walk causes no skip and no repeat; a snapshot for another agent is ignored. portalAgentLandingStage: the snapshot is retaken on agent -> room -> agent (sidebar door and by URL) and agent -> Inbox -> agent, held on a same-agent click, and a cold link is untaken until trusted, retaken once, then held through a later poll and a failed-then-recovered poll. The room-by-URL case found a gap in A2: Back/forward or a link to /workspace/r/:id never writes activeRoomId, so stageAgent kept the agent on stage. stageAgent now also reads activeRoomIdFromRoute. Mutations: dropping order from cycleChat, dropping the route-room term, and retaking on every trusted flip each turn one of these red. Refs abilityai/trinity-enterprise#856 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ing note (ent#856) workspace-chat-tabs-and-titles.md gains the section: the one stage-key writer, the comparator, the unpruned snapshot, landing arms 0/1/fresh, replace-on-route-landing, and the test map. §5.43 records the intended arm-1 behaviour change (Back / the agent row from one of its chats returns to the chat left open in this session — the 2026-10-13 landing ruling) and that a room reached by URL is off stage. Refs abilityai/trinity-enterprise#856 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…itch, T1/T2/T3/T5 recorded (ent#856) Requirements first (RoE 1): the Action row of the §5.40 table reads Newest by default with Urgency on the control; the new 'Action's sort' block carries AC B1-B7 and the orchestrator rulings — T1 Action only, T2 the operator queue left as is, T3 the rail follows the default explicitly, T5 the three inconsistencies recorded (iii is a public bug for handoff). Refs abilityai/trinity-enterprise#856 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…equest (pure layer) (ent#856)
portalInbox.js: INBOX_SORTS / INBOX_SORT_DEFAULT ('newest'),
normalizeInboxSort (allowlist), askRecencyCompare (created_at desc, id
desc — byAtDesc's rule), inboxSortFor(tab, query, stored) (URL, then the
viewer's memory, then Newest; null off Action), inboxSortKey /
loadInboxSort / saveInboxSort (identity-bucketed, try/catch, fail-open),
and actionItems(openAsks, now, sort). The urgency order (§3g C1) is
unchanged and now on the switch.
portalRailAsks.js: railAskItems passes INBOX_SORT_DEFAULT explicitly (T3).
RED first: portalInboxSort.spec.js (new) and the rail's T3 case failed on
the missing API and the urgency order. Existing pins moved to the ruling:
portalInboxActionSort passes 'urgency'; portalInbox.spec.js's default case
reads newest-first; the deep-linked-ask hold pin selects the older ask so
it still proves the second slot holds. Mutation: the rail passing
'urgency' turns two portalRailAsks cases red.
Refs abilityai/trinity-enterprise#856
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…r memory (ent#856)
PortalInbox.vue: sort = inboxSortFor(tab, ?sort=, remembered) — the URL,
then the viewer's memory (bucketed by resolveLayoutIdentity), then Newest;
null off Action. actionItems takes it; visitKey carries it, so a change is
a new visit (rows re-sort on the click, then hold still). onSort writes
both the URL (replace) and the memory. A stray or unknown ?sort= leaves
the URL, like a stray ?from=. The control is BaseSelect variant="ghost"
("Sort asks": Newest / Urgency) in the list head line, Action only (T1).
PortalInboxList.vue: the head line is one fixed 44px box (the ghost
select's height) with a #head-end slot; the loading reserve draws the
same box, so neither a tab switch nor landing data moves the list. The
line grows from ~36px to 44px once, on every tab and state alike.
Mount pins (RED first, 7 failing): Newest default + the control's name and
options; switching writes ?sort= by replace and the memory; URL > memory >
default; another viewer does not inherit; unknown and off-Action ?sort=
stripped; a sort change is a new visit; priority + expiry marks in both
orders; one head-line box across states. Mutation: dropping sort from
visitKey turns two red.
Refs abilityai/trinity-enterprise#856
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
workspace-inbox.md: Action's order line reads Newest by default with Urgency on the switch; a new section maps sort -> actionItems -> visitKey -> onSort (URL replace + identity-keyed memory), the stray-key watcher, the 44px head-line box, the rail's explicit default (T3) and the T2/T5 records; the test table gains the ent#856 specs. Full npm run test:unit (--maxWorkers=2; the default worker count was OOM-killed): 357 files, 5801 tests passed, ratchets included. Refs abilityai/trinity-enterprise#856 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ent#856) The shell skeleton (drawn by Portal.vue until the stage has a verdict) still drew the old ~36px head line while PortalInbox's reserve and PortalInboxList's ready head are the 44px h-11 box, so the first-paint -> Inbox hand-off moved every row down 8px. The drift spec compared only the header and tab strip; it now compares the head line against the reserve's classes and the ready head's box too. Red before the fix: 'the list head line is the Inbox's one 44px box' (received 'shrink-0 px-4 pt-3 pb-2 ...'). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… (ent#856) Two immediate watchers each called replaceQuery in the same tick over the same stale route.query, so the second replace put the first's key back (?tab=unread&from=x&sort=newest ended as ?tab=unread&from=x). One watcher now returns both stray keys and clears them in a single replace. Red before the fix: 'a stray ?from= AND a stray ?sort= together both leave the URL in one replace (review I2)' (received from: relay). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Full run-full.sh on a local stack (pytest-randomly installed, so test order is shuffled) failed several tests that pass in isolation. None of these are product bugs; each fix is test-side. Integration/API tier: - Default-state tests (Nevermined no-config x2, shared-folders defaults, read-only default) asserted "fresh agent" state on the module-scoped created_agent that sibling tests mutate. They now use isolated_agent. - test_rename_agent_name_sanitization renamed to a fixed name. A deleted agent's soft-deleted ownership row keeps that name and its volume base reserved by design (#1664/#1671), so every run after the first got 409. The target is now derived from the fixture agent's unique suffix. - TestSubscriptionAutoAssign round-robin/tiebreak: since #2409 auto-assign ranks by cached headroom, so a real operator subscription always beats the suite's fake tokens. Skip on foreign subscriptions like the existing no-subscription test (#2420); the tiebreak subs now carry the suite prefix. - test_respond_to_pending_item: answering a platform-raised item lands `acknowledged` (#2372); pin both shapes. Unit tier (fixed at the polluter, not the victim): - test_brain_orb_postprocess: a lazily imported task_execution_service bound the patched database.db stub; pre-import it before each patch. - test_monitoring_dormant_skip: bare sys.modules assignment of a stub services.monitoring_alerts; use monkeypatch.setitem. - test_subscription_auto_switch_no_cred_import: restore the parent-package attribute as well as the sys.modules slot. - test_ec_loop_service_edges m16: the [postgres] variant fails on the known float4 REAL-column bug, not pollution; strict xfail citing #3441. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…l tabs, Asks always present (#3672, ent#892) #3672 — the Workspace sidebar's agent and chat lists jumped on a cold load. The roster, the chat list and the asks feed land at different times, and each re-laid the list: agent rows re-sorted and grew a subtitle when chats arrived, an asked agent was appended past the collapse limit when asks arrived, chat groups popped in, and the Projects link (its capability rides the roster) pushed the search box and both lists down. - `sidebarSettled` (portalUtils, pure): bootstrap placed the caller AND the asks feed has a verdict (loaded / 404-403 / failed) — state, never a timer. Portal.vue latches it (`sidebarReady`), resetting only when the session ends. - PortalSidebar: both sections hold skeleton rows (new PortalSkeleton `sidebar-agents` / `sidebar-chats` variants) until settled, then render in one pass inside one `.ws-fade-in` (180ms opacity, off under reduced motion). A platform session holds the Projects link's slot (`sidebar-link`). An empty chat list keeps a stable "No chats yet" state. - Measured (Playwright, cold /workspace, 3 runs, fast and with staggered latency): CLS 0.008 -> 0.0003; the residual is in the Inbox pane header. Mutation check: with PortalSidebar.vue reverted, 5 of the 11 tests in portalSidebarCalmLoad.mount.spec.js go red; restored byte-identical. trinity-enterprise#892 — the agent rail's tabs read Info, Canvas, Files, Projects, Asks, Work, Loops. Asks is always present (operator ruling 2026-10-13): ent#836's presence rule and `asksTabPresent` are removed, the registry entry declares an empty state ("No asks" / "Open the Inbox"), and PortalRailAsks owns all four states via `railAsksView` (skeleton before a verdict, LoadFailed, empty, rows). Default tab stays Work; a remembered tab still wins; ⌥. walks the new order. With portalRail.js reverted, 5 of 14 portalRailAsks.spec.js tests go red. Order-pin specs updated per the issue; two source-text pins moved to the new sidebar gate; PortalSidebar's raw-gray ratchet entry lowered 35 -> 29. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The `sidebar-link` skeleton held a 36px slot for a platform session until the sidebar settled, then vanished on every install not entitled to Projects (all OSS installs) - a new layout shift to spare a late pop-in on the few entitled ones. Operator ruling: no hold. The link renders once the roster says `projectsAvailable`; the unused `sidebar-link` variant is removed, the spec asserts nothing sits between New chat and the search box before the verdict, and the feature flow no longer promises a held slot. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
When the first `fetchAllSessions` fails, the store returns its last good list - `[]` - and the sidebar told the person they had no chats. Portal.vue now passes `:threads-failed="store.sessionsFailed"` (the same signal the Inbox reads) to both sidebars. The empty copy renders only for a trusted list; a failed read shows an InlineError instead (the stale banner over rows, "Couldn't load your chats" over none) with a retry wired to `refreshThreads`, dismissible and back on the next failure. Mount tests cover all four cases. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PortalRailAsks read `view.state` but never `view.stale`, so a failed asks
refresh left the last list (or the empty state) on screen with no word of
it. It now raises the stale InlineError the way PortalRailCanvas,
PortalRailFiles and PortalLoops do ("Couldn't refresh your asks - showing
data from ..."), with a retry and a dismiss that re-arms on the next
failure. The two `./portalUtils` imports are merged into one. Mount tests
cover the banner over rows, over the empty state, and its absence.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…3672) Binding `:settled="true"` in Portal.vue left the suite green: the calm-load spec mounts PortalSidebar alone. portalSidebarLatch.mount.spec.js mounts the shell (shallow, the portalAgentLandingStage harness) with the asks read held open and checks both sidebars hold `settled=false` until the verdict, latch once, stay latched when the verdict's inputs drop on a later refresh, and reset on sign-out so the next session's first load holds again. Mutation check: `:settled="true"` -> 3 of 4 red; dropping the sign-out reset -> 1 red; replacing the latch with a plain mirror -> 1 red; restored byte-identical. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This was referenced Oct 13, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Integration surface for #3674, #3665, #3681, #3682, #3683. Never merged; members merge individually once green.