Commit graph

339 commits

Author SHA1 Message Date
Danny Avila
038903abc0
📱 fix: Open Sign-In and External Links From iOS Home-Screen Apps (#16077)
* 📱 fix: Open Sign-In and External Links From iOS Home-Screen Apps

Sign-in prompts, the MCP OAuth Continue buttons and the account menu links opened with window.open and a noopener,noreferrer features string. WebKit treats a features string as a popup-window request, which an iOS home-screen app cannot open, and the chat sign-in also awaited its CSRF bind before opening, outside the tap. Links now open through a target=_blank link with rel=noopener noreferrer, and the chat prompt binds when it appears so the tap opens the provider synchronously.

* fix: Keep sign-in closed until its CSRF bind lands, and surface builder OAuth in its dialog

A tap while the chat prompt's bind was still in flight could open the provider before its CSRF cookie existed, and an Action callback has no fallback without that cookie; the button now stays disabled until the bind lands. The MCP builder card opened the authorization URL only after its initialize request, which iOS home-screen apps refuse and which left the card with no way to continue; it now shows the MCP OAuth dialog instead.

* fix: Let the tapped prompt own the CSRF binding, and show only the card's live OAuth flow

Live sign-in prompts share one CSRF cookie per callback path, so the last prompt to bind owned it; a tap now re-binds its own prompt right after opening the provider. The builder card kept a local copy of the authorization URL that outlived its flow and could reopen it on a later initialization; it now shows the manager's shared flow URL, cleared when the flow ends, and only for a flow the card itself started.
2026-09-19 08:09:25 -04:00
Marco Beretta
5c72d4146e
📱 fix: Fit the Tool Library and Skills Pickers to Small Screens (#16067)
* 📱 fix: Fit the Tool Library and Skills Pickers to Small Screens

The Tool Library dialog assumed a viewport wide enough for its 224px
navigation rail: below `md` the rail took more than half of a 390px
screen and the card grid was squeezed into the remainder. The rail is
now `md`-only, and the same kind and view navigation renders as a
horizontally scrollable chip row under the search field.

Card and row actions were hover-only, so Configure, Favorite and Remove
were unreachable on a touch screen. They stay hidden until hover only
where hover exists (`@media (hover: hover)`), and are always visible
otherwise.

The Skills picker header put the create button, the filter field and the
view radio on one row, where the field collapsed to 50px beside a
`shrink-0` radio. The view radio and the create button now sit on the
title row, the create button flush right, and the filter field owns the
row below them.

The dialogs fill the viewport below `md` and switch `vh` to `dvh`, so
mobile browser chrome no longer cuts off the bottom of the panel.

* ♿ fix: Gate Hidden Tool Actions on Touch, Not on Hover

Hiding an action until hover has to ask whether a finger can reach it, and
`(hover: hover)` does not answer that: on a 2-in-1 it is true because it
describes the trackpad, while the touchscreen sits right next to it. Tool cards
and tool rows gated their Configure, Favorite and Remove controls on that query,
so a finger user of such a device saw no actions at all.

`packages/client/tailwind.preset.cjs` already carries `touch` for this exact
distinction, `(any-pointer: coarse)` rather than `(pointer: coarse)`. This adds
its inverse, `no-touch`, and gates the hidden state on it, so the hide applies
only where no coarse pointer exists at all and hover no longer takes part in the
decision.

* 📐 fix: Split the Skills Header Onto Two Rows Below md

The Skills header put the create button, the filter field and the view radio on
one row. The radio's segments are `whitespace-nowrap`, so the group never
narrows below 261px, and the field — `flex-1`, basis 0 — absorbed the shortfall
and measured 50px on a 390px screen.

Below md the radio and the create button now take the first line, the create
button flush right, and the field takes the second. From md the row is exactly
what it was: create button, field, category, radio. One DOM order serves both
through `order-*`, so the radio and the create button each exist once.

* 🧷 fix: Let the Skills View Radio Wrap Instead of Clipping

The three view segments carry `whitespace-nowrap` and `px-4`, so the group has a
hard 261px minimum in English and more in longer locales. On a 360px phone that
group and the 42px create button exceed the dialog, which is `overflow-hidden`,
and the trailing option becomes unreachable. The group now uses the primitive's
`wrap` variant and may shrink, so the segments flow onto a second row; at 390px
they still occupy one.

`wrap` reproduced `inset-y-1` by assuming the group had no vertical padding, so
turning it on for a `p-1` group shrank the moving indicator by 8px and left it
floating inside its segment. It now measures that padding from the first row's
offset, which leaves an unpadded group's geometry unchanged.

* 🧪 test: Cover the Small-Screen Tool Library and Skills Header

Ten mock-harness scenarios for the behaviour this branch changes, one per
acceptance scenario id, asserting rendered geometry and computed styles rather
than class strings.

The tool library file covers a phone viewport (the dialog fills it, the rail is
gone, the chip row filters, and no card crosses the viewport edge — the 744px
grid-column overflow), the action gate on all three pointer profiles (visible
with a coarse pointer, hover-gated with a mouse, revealed by keyboard focus),
the selected tool row, and the unchanged desktop rail.

The skills file covers the two-row header below md, the wrapped view radio at
320 and 280 CSS px with every option inside the dialog and the indicator on its
checked segment, and the single desktop row.

* 🧪 test: Pin the Scenario Contexts and Settle Their Geometry

The verifier runs every tagged test in three projects, one of which is a touch
phone, so the mouse scenarios asserted a mouse on a device that reports a coarse
pointer, and the desktop rail on a 412px screen. They now pin their own viewport
and pointer instead of inheriting the project.

Geometry and opacity were also read mid-animation: the dialog scales as it
opens, which put its left edge at 1.5px rather than 0, and the actions fade over
the shared motion duration, which caught keyboard focus at 0.84. Both now poll
for the resting value.

With nothing selected the skills section offers only the dashed empty-state
card, whose accessible name carries the hint line, so the exact-name locator for
Add skill never matched.

* 📏 fix: Keep the Tool Library Shell and Its Body the Same Height

Giving the shell an explicit `md:h-[88vh]` while the body kept its 840px ceiling
let the two disagree: on a viewport taller than about 955px the difference became
dead space below the catalog, 216px at 1280x1200, and below `md` the full-height
sheet left 293px on a 744x1133 tablet.

The body carries the height from `md` again, the way it did before this change,
so the shell wraps it. Below `md` the shell stays the full-bleed sheet and the
body is capped to the same `100dvh`, which it needs because its `h-full`
resolves against a grid area sized to the whole catalog.

Measured shell and body heights now agree exactly: 840 at 1280x1200, 757 at
1280x860, 1133 at 744x1133 and 844 at 390x844, with the catalog still scrolling
inside its own container.

* 🎯 fix: Keep the Search Field Clear of the Dialog Close Button

`md:px-6` on the search row is a responsive variant, so it is emitted after the
base utilities and reset the `pr-12` that reserves room for the close button:
on desktop the row's right padding dropped from 48px to 24px and the field ran
16px underneath the button. It now sets only the left side at `md`.

Measured gap between the field's right edge and the button: 8px at 1280x860 and
at 390x844, with no vertical overlap. Asserted inside the phone-viewport and
desktop-rail scenarios rather than as a new id, since it is the same surface.

* ⌨️ fix: Read the Skills Header in Tab Order and Keep the Chip Row Draggable

Two defects with the same shape: a control that the eye can reach and the
keyboard or the mouse cannot.

The Skills header used `order-*` to serve both layouts from one DOM order, which
left tab order disagreeing with the visual order from `md`: focus went from the
rightmost radio group back to the create button and then into the middle. The
breakpoint now selects the DOM order instead, the way the marketplace dialog
already selects its rail, so each layout reads in the order it renders. Every
control is still built once, so there is one radio group and one create button at
any width.

The mobile chip row hid its scrollbar unconditionally. A narrow desktop window is
below `md` as well, and there a wheel scrolls the page rather than the row, so
Made by you and Favorites could not be reached with a mouse at all. The hiding is
now gated on `touch`, the same `(any-pointer: coarse)` query the rest of this
branch uses.

Measured: desktop DOM order create, field, radio and mobile radio, create, field,
each matching its layout; `scrollbar-width` resolves to `auto` in a 520px mouse
window whose row overflows, and to `none` on a touch phone.

* 👆 fix: Give the Touch Layout Touch Geometry

Two findings with one cause: the coarse-pointer layout inherited the mouse
layout's geometry after this branch changed what it shows and in what order.

Card actions no longer fade out where a finger can reach them, but they are
absolutely positioned over the card's content, so on touch they covered the last
line of a three-line description for as long as the card was on screen - an
overlap that used to be transient. The card now reserves a strip for them there,
`touch:h-36` with `touch:pb-9` on the content, which keeps all three lines.

The chip row is that layout's primary navigation and exists only where a finger
is what reaches it, yet its chips were about 30px tall and its create control
32px, while the repository already exposes the 44px `theme-control-touch` size
for this. Both now carry that floor through `touch:`, and the mouse layout keeps
its compact sizing.

Measured on a 390x844 touch profile: all eight chips 44px, the card 144px, and
the description's box clear of the action cluster. Both are asserted in the
phone-viewport and touch-action scenarios.
2026-09-18 17:29:39 +02:00
Danny Avila
299dbc6075
📖 Load Repository Instructions for Attached Workspaces (#16008)
* Load authorized workspace repository instructions into agent context

* Unify repository instruction loading across agent ingresses

* Await cancellation regression assertion directly

* Annotate the shared instruction loader for declaration builds

* Expose repository read budget and inject the host-owned cache
2026-09-16 09:00:35 -04:00
Danny Avila
5e2a2c65a3
📖 fix: Preview Extracted Upload Text (#15946)
* fix: preview extracted upload text

* fix: preserve extracted-text preview metadata

* fix: complete extracted preview recovery paths

* fix: retain preview metadata across attachment recovery

* fix: preserve preview metadata across legacy steering

* fix: hydrate legacy steer preview metadata

* fix: hydrate reconnect steer metadata

* fix: preserve text preview metadata in resumed shares

* fix: hydrate legacy preview metadata safely

* fix: close preview lifecycle and recovery gaps

* fix: preserve preview ownership and wire metadata
2026-09-14 20:58:04 -04:00
Danny Avila
74fa60cee1
🔎 fix: Enforce WEB_SEARCH Role Permission on Agents (#15875)
* 🔎 fix: Enforce WEB_SEARCH Role Permission on Agents

Squash of #15822 onto dev, preserving authorship:

- 🔎 fix: Enforce WEB_SEARCH Role Permission on Agents
- 🧪 test: Cover the Runtime Tool Loader's Web Search Gate
- 🔒 fix: Gate Web Search on the OpenAI-Compatible Embedder Surface
- 🩹 fix: Authorize Web Search by the Execution Context's User
- 📝 docs: Note the Grant Memo's One-Subject-Per-Request Assumption
- 🧹 fix: Deny Web Search Before an Endpoint Default Can Grant It
- 🙈 fix: Withhold the Web Search Model Parameter From a Denied Role
- 🧱 fix: Strip Provider-Native Search an Endpoint Re-Enables After Denial
- 🔌 fix: Close the OpenRouter Plugin Path and Stop Pruning Hidden Parameters

* 🧭 fix: Gate Provider-Native Web Search on the Built Config

The web search role gate ran before the provider builder and only when the
agent's stored `web_search` was not `false`, so an endpoint's `addParams`
re-enabled native search for a denied role, and every OpenAI-compatible and
Responses request read the role whether or not search was in play.

The gate now reads the builder's output: when it carries a native search tool
or OpenRouter's `web` plugin, the `WEB_SEARCH` grant is resolved and the search
stripped on denial. The routes that reach `initializeAgent` with `runtime` and
no `req` pass a lazy `resolveWebSearchGrant` that joins their request-memoized
grants, forwarded through agent discovery to handoffs and subagents. The
embedder surface wires it whenever `getRoleByName` is supplied, independent of
`appConfig`.

---------

Co-authored-by: TomasPalsson <tomas2212042710@gmail.com>
2026-09-12 23:00:10 -04:00
Joseph Licata
f25242df98
🦘 feat: Share Run-Scoped Files With Subagents (#15848)
* feat: add run-scoped file sharing for subagents

Delegate current-message files through authorized execution catalogs and publish selected immutable code-output versions through the existing file pipeline.

Requires the companion agents SDK context adapter before the dependency can be released and pinned.

* fix: preserve child tool artifacts and verify file-sharing lifecycle

Keep citations and other non-sandbox artifacts on their existing delivery path while retaining private code outputs and inspected images until publication.

Cover native and text delivery, nested and sibling authorization, concurrent runs, cancellation cleanup, and checkpoint recovery through the browser harness.

* test: align file-sharing fixtures and integration cleanup

* test: repair LibreOffice conversion fixtures

* test: verify deployment skill bytes across checkout formats

* fix: align subagent file budgets and refresh boundaries

* chore: update agents SDK to 3.8.6

* refactor: move code output processing into API workspace

* fix: enforce shared file budgets for parents and teams

* fix: declare the shared file encoder interface
2026-09-12 20:14:22 -04:00
Marco Beretta
246e7e5de7
💠 fix: Paint Native Tool Verification Badges (#15862)
* 🪪 style: Give Native Tools a Painted Verified Mark

The native badge on a tool card was lucide's BadgeCheck filled with status-success-strong: green, the same hue the card already uses for selected, and stroked in currentColor, so the white check color also outlined the badge and left a white halo on any surface that is not the card's own white — hover, selection, dark mode.\n\nVerifiedIcon paints the badge instead of stroking it and carries the check as the only stroked path, and status-verified gives provenance its own blue across all four themes.

* 🩹 fix: Hold the Verified Mark at 3:1 on Every Card State

The dark fill was tuned against surface-secondary, but ToolCard repaints to surface-tertiary (#2f2f2f) on hover, where #0b74d4 fell to 2.85:1 — below the non-text floor, and invisible to the spec, which asserted the panel only. The fill moves to #1a7fd8 and the contract now covers every background the card takes. Both of the mark's relationships are graphical objects under WCAG 1.4.11, so the check is held at 3:1 rather than AA: no hue clears 3:1 on the hover surface and 4.5:1 under a white check at once.\n\nA theme that predates the token names status-success-strong and not status-verified, and would otherwise wear LibreChat's stock blue beside its own text-on-status and card surfaces. resolveTheme and the legacy mapColors adapter now carry the mark onto the fill it had before the token existed, following the six derived-token fallbacks already there.

* ✅ test: Exercise the Verified Mark in the Tool Library

Four scenarios in the mock harness, each named by the behavior it observes: the mark a native card wears, its silhouette under the dark hover repaint that the retune fixes, a pre-token deployment palette carrying it on the theme's own success fill, and a third-party MCP server card carrying no mark at all. The contrast assertions read the rendered colors rather than the tokens, so a regression in the component or in theme application fails them.

* 🧪 fix: Teach the Marketplace Dialog Spec About VerifiedIcon

The suite hand-lists the @librechat/client surface it mocks, so ToolCard's new icon resolved to undefined and every render in the file threw.

* 🧪 fix: Reach the Agent Builder on a Narrow Viewport

The scenario run covers mobile emulation, where the side rail is collapsed and the panel opens from the header's Control Panel toggle; the helper only knew the rail. The MCP scenario also minted its access token before the app had loaded, so the refresh request had no origin to resolve against.

* 🧪 fix: Run the Verified-Mark Scenarios on a Touch Viewport

Mobile emulation has no side rail: the panel switcher lives inside a sidebar that starts off-canvas, so the helper opens the sidebar and picks the Agent Builder entry from the switcher's menu. A touch viewport also has no hover state, so the hover scenario measures whatever background the card actually takes there and still holds the mark to 3:1 against it.

* 🩹 fix: Keep Pre-Token Themes on the Fill the Mark Inherited

The first fallback only fired when a theme restated status-success-strong, so a partial palette that customizes surfaces or text-on-status and inherits the green still landed on the stock blue. resolveTheme now derives the omission for any custom theme from the resolved success fill, and the legacy adapter takes the mode's bundled palette from ThemeProvider so it can do the same for a theme that names neither.\n\nThe scenario helper picked its path from a one-second probe for the desktop rail, which turns a slow first paint into the mobile branch; it now decides on viewport width and keeps the suite's full assertion timeout in each branch. The mock project is Desktop Chrome, so a touch scenario asks for a touch context explicitly rather than leaving that layout unexercised in CI.

* 🧪 fix: Match the App's Inclusive Layout Breakpoint

useSidebarState switches on (max-width: 768px), so a 768px-wide context renders the sidebar switcher while the helper's strict comparison went looking for the desktop rail.

* 🩹 fix: Scope the Verified Fallback to Themes That Painted the Mark

Two rounds pulled the same predicate in opposite directions: keying it on a theme restating status-success-strong missed palettes that only repaint the check or the card, and keying it on any custom color at all gave a new partial theme a green mark identical to the selected cue. Both are answered by asking what the theme actually painted — the fill the mark wore, the text-on-status check it carries, or the surfaces ToolCard takes — which is the same shape as the series-scale fallback above it. A theme that repaints none of those keeps the bundled blue.

* 🧪 fix: Wait for the Shell Before Reading the Sidebar State

isVisible() answers immediately, so a slow first paint read as an already-open drawer and the helper then waited on a switcher sealed inside it. It now waits for whichever of the opener and the switcher the shell paints first, which is the same assertion timeout the desktop branch gets.

* 📝 fix: Keep the Token Catalog Entry Catalog-Sized

The entry restated the icon's construction, the contrast rule and the pre-token fallback, all of which live on the token's own doc comment and in semanticTokens.spec.ts; the catalog around it is one line per role.

* 🔦 fix: Let a Crashed Lighthouse Run Be Retried

The audit shells out to Lighthouse three times and fails the lane if any single process exits nonzero, so a NO_NAVSTART — Lighthouse losing the trace of a page load, whose own message is "run Lighthouse again" — sinks a pull request that has nothing wrong with it. That is what happened on this branch: runs 1 and 2 wrote reports, run 3 crashed, and the identical head passed on rerun.\n\nEach run now gets a second attempt before the failure propagates. The repeat is per run, so the median is still taken over three completed reports, and a page that genuinely cannot be traced fails both attempts and still fails the lane.

* 🧪 fix: Read the Drawer State From Its Inert Marker

The readiness wait accepted the panel switcher, which is mounted whether or not the drawer is open — translated off-canvas and inert — so a shell that painted the sidebar before the chat header could satisfy it with the drawer closed and leave the helper clicking an off-viewport control. The drawer's own inert attribute is the state, so the helper waits for the drawer to attach and reads it there, with no timing probe on either side.

* 🧪 fix: Open the Drawer Before Trusting the Restored Form

A narrow context that had already selected the builder restores the form inside the closed drawer, and a translated element passes a visibility check, so the early probe accepted a form sitting under an inert drawer and handed it to the caller. The layout decision now comes first and the drawer is opened before anything inside it is read.
2026-09-12 16:03:03 -04:00
Marco Beretta
5b4ec6bfe2
🔎 feat: Filter and Sort the Sidebar Chats List (#15842)
* 🔎 feat: Filter and Sort the Sidebar Chats List

The sidebar could only list every unarchived chat by last activity. Archived chats
lived behind a settings modal, sorting was not exposed at all, and the only filter
was a bookmark dropdown squeezed next to the search field.

A single control beside the Chats heading now carries the status (active or
archived), the sort field and direction, and the bookmark filter, each scoped to
that list alone: Projects, Pinned and Favorites are untouched by it.

* 🧹 fix: Answer Review Round on Sidebar Chat Filters

Rows derive their archive action from `conversation.isArchived`, which the list
projection now carries, instead of from the sidebar's filter mode: the archive view
keeps the unarchived pins beside it, and its previous page's rows outlive a status
switch, so the list mode offered to restore chats that were never archived.

Mutations reconcile the archived list caches they can now reach — rename, pin,
tags, project assignment, shared links — and archiving refetches the archived
query instead of inserting into page one, which dropped the row whenever the
active sort placed it later.

Title grouping keeps the order the server paged, heads each group with the title's
first code point, and the archive's legacy group falls back to `createdAt` the way
the cursor orders it. A failed list request renders an error with Retry rather
than "No chats yet", the status setter sanitizes a malformed stored sort, and
signing into another account no longer inherits the previous one's filters.

* 🧵 fix: Keep Sidebar Cache Writes Inside the Chosen Sort

Title groups are emitted as contiguous runs, so a `#` heading that the server's
order places either side of a letter no longer merges and drags a row across the
cursor boundary the next page continues from.

The live cache writers only reproduce newest-first, which is the one order they can
place a row in without the server's keyset. A title, created-at or ascending variant
now takes the field update in place and is refetched instead of having rows moved to
or seeded at the front of page one.

Archive and restore refetch every loaded page of the lists they touch — the archive
preserves `updatedAt`, so a restored chat can belong on a later active page — while
inactive variants are left stale rather than eagerly refetched. Archiving everything
still refetches them all, because every cached list is wrong afterwards. Renaming or
deleting a bookmark now rewrites the sidebar's selected tags, which otherwise kept
filtering by a name no menu could uncheck.

* 🗂️ fix: Treat the Archived List as a Live Conversation Cache

* 🔍 fix: Refetch List Variants Whose Membership the Client Cannot Decide

A cached search result is evaluated by the server, so a newly created row may or may
not belong in it. Skipping those variants left a mounted search list missing a row
it should show — forking from an archived search result being the case that exposed
it. The insert decision is now three-valued: insert into a variant whose membership
and order the client can reproduce, skip one the row provably does not belong to,
and refetch anything undecidable.

Forks, duplicates and imports inherit the source's archive state, so their
invalidations cover the archived list too, and the delete path reads `chatProjectId`
from archived caches as well.

* 🔎 fix: Reconcile Searched List Variants on Every Write

* 📣 fix: Announce Archive and Restore From the App Live Region

* 🧹 fix: Drop Unmounted Lists on Archive-All and Fill a Missing Archive State

Archive-all refetched every loaded page of every cached list, mounted or not, so a
session that had visited several sort, bookmark or search combinations paid a burst of
requests for lists nobody was looking at. Mounted variants still refetch all pages;
the rest are dropped, which reconciles them more thoroughly than a refetch and costs
nothing until something mounts them.

A row's own `isArchived` decides whether its menu offers Archive or Restore, and this
branch is what adds that field to the list projection — so mid-rollout, an archived
row from an older backend would offer Archive and submit a no-op. The list query now
fills a missing value from what the variant asked for, leaving an explicit one alone.

* 🧭 fix: Give Callers One Way to Reconcile Both Conversation Lists

Callers that cannot say what changed — a stream recovered after a disconnect, a
schedule whose run moved a chat, a project that took its chats' fields with it — were
invalidating the active list by hand, so a mounted archive kept a row the server had
already changed. `invalidateConversationLists` reaches both prefixes and is now the
only way those paths reconcile; `findConversationListQueries` shares its key list, so
adding a prefix is one edit rather than a sweep.

* 🔤 fix: Follow the Server Order, Index the Created Sort, and Tell the Truth When Empty

* 🎯 fix: Identify the Open Chat When a Restore Resolves

A restore leaves the chat open, so its success callback writes the new archive state
into the active conversation — but the callback captured that conversation when the
menu item was clicked. Open another chat before the request lands and the identity
check still matched the captured row, replacing the conversation the user had just
opened while the URL kept pointing at it.

The write is now a functional update that compares `conversationId` against current
state, so a stale callback leaves the open chat alone. Reading the conversation value
is no longer needed either, which also stops every row's menu from re-rendering on
conversation updates.

* 🧷 fix: Recheck the Open Route When an Archive Resolves
2026-09-12 15:56:10 -04:00
Danny Avila
1d45f5bd40
🏡 feat: Remember Agent Workspace Defaults (#15843)
* 🧭 feat: Remember Attached Workspace Choices

* Guard Workspace Defaults Across Navigation and Version Restore

* Order Workspace Preference Imports

* fix: Harden agent workspace defaults

* fix: Type workspace validation schema

* test: Cover safe execution projections

* fix: Bind workspace defaults to attached environments

* fix: Preserve unchanged workspace bindings

* fix: Scope workspace preferences to active agents

* fix: Preserve workspace selection state

* test: Target workspace discovery status

* fix: Keep workspace reset available

* fix: Resolve workspace defaults per environment

* fix: Collect per-environment workspace defaults

* fix: Reconcile Deleted Agent Graph Caches

* fix: Reject Stale Workspace Bindings
2026-09-12 05:45:16 -04:00
Danny Avila
6965f8e7d3
🛂 fix: Back Off MCP OAuth Discovery (#15783)
* fix: Back Off MCP OAuth Discovery

* fix: harden MCP discovery recovery state

* fix: synchronize MCP authorization recovery

* fix: fence MCP discovery authorization state

* fix: retain recovery generation through cache outages

* fix: close MCP authorization fence gaps

* fix: complete MCP authorization lifecycle

* fix: close MCP authorization recovery races

* fix: centralize MCP recovery lifecycle

* fix: fence ephemeral MCP catalog observers

* fix: close MCP authorization recovery races

* test: cover Redis token settlement race

* fix: prefer authoritative token settlement

* fix: harden MCP authorization retry recovery

* fix: back off stalled OAuth refreshes

* fix: bound durable authorization retries

* test: isolate authorization retry adapter

* fix: close MCP authorization recovery boundaries

* fix: defer agent authorization observers

* fix: close OAuth fence race windows

* fix: keep retry deferrals monotonic

* chore: sort startup hook imports
2026-09-09 22:21:34 -04:00
Danny Avila
442d477acf
🛎️ fix: Preflight MCP Readiness for Scheduled Chats (#15782)
* fix: Validate Scheduled MCP Readiness 🕰️

* fix: Keep Schedule Permission Store Failures Retryable

* perf: Parallelize Scheduled MCP Preflight

* fix: Preserve Scheduled MCP Invariants

* fix: Close scheduled MCP recovery gaps

* fix: harden scheduled MCP recovery

* fix: bound scheduled MCP validation

* fix: harden scheduled MCP admission bounds

* fix: align scheduled MCP graph admission

* fix: bound scheduled MCP admission

* fix: align scheduled MCP preflight with runtime

* fix: persist scheduled MCP failure outcomes

* fix: match scheduled MCP runtime semantics

* fix: parallelize scheduled MCP admissions

* fix: mirror scheduled subagent traversal

* fix: enforce scheduled subagent config limits

* fix: bound scheduled MCP initialization

* fix: mirror lazy subagent metadata budgets

* fix: complete scheduled MCP recovery

* fix: attribute missing MCP tools

* fix: harden scheduled MCP admission

* fix: preserve scheduled admission invariants

* fix: retain inaccessible MCP requirements

* fix: make MCP admission recovery durable
2026-09-09 22:18:45 -04:00
Danny Avila
af6e97f0ae
🏠 fix: Keep MCP Caches Fresh on Demand (#15765)
* fix: stop refreshing MCP caches on every chat message

* fix: reconcile MCP caches independently of chat turns

* fix: scope MCP polling to visible controls

* fix: refresh status in the unpinned MCP submenu
2026-09-09 00:36:04 -04:00
Danny Avila
1dfefc3037
🪹 fix: Attach Ready OAuth MCP Servers With Empty Catalogs (#15727) 2026-09-08 07:14:13 -04:00
Danny Avila
f50c40e2f5
🧰 feat: Move Code Settings Into Run Code (#15722)
Some checks are pending
Backend Unit Tests / Build packages (push) Waiting to run
Backend Unit Tests / Codegraph select (push) Waiting to run
Backend Unit Tests / TypeScript type checks (push) Blocked by required conditions
Backend Unit Tests / Circular dependency checks (push) Waiting to run
Backend Unit Tests / Tests: api (shard 1/3) (push) Blocked by required conditions
Backend Unit Tests / Tests: api (shard 2/3) (push) Blocked by required conditions
Backend Unit Tests / Tests: api (shard 3/3) (push) Blocked by required conditions
Backend Unit Tests / Tests: data-provider (push) Blocked by required conditions
Backend Unit Tests / Tests: data-schemas (push) Blocked by required conditions
Backend Unit Tests / Tests: @librechat/api (shard 1/4) (push) Blocked by required conditions
Backend Unit Tests / Tests: @librechat/api (shard 2/4) (push) Blocked by required conditions
Backend Unit Tests / Tests: @librechat/api (shard 3/4) (push) Blocked by required conditions
Backend Unit Tests / Tests: @librechat/api (shard 4/4) (push) Blocked by required conditions
Codegraph E2E Votes / vote (full suite) (push) Waiting to run
Docker Dev Branch Images Build / publish (push) Waiting to run
Frontend Unit Tests / Client build recovery regression (push) Waiting to run
Frontend Unit Tests / Codegraph select (push) Waiting to run
Frontend Unit Tests / Build packages (push) Waiting to run
Frontend Unit Tests / TypeScript type checks (client) (push) Blocked by required conditions
Frontend Unit Tests / Tests: @librechat/client (push) Blocked by required conditions
Frontend Unit Tests / Tests: Ubuntu (shard 1/2) (push) Blocked by required conditions
Frontend Unit Tests / Tests: Ubuntu (shard 2/2) (push) Blocked by required conditions
Frontend Unit Tests / Vite build verification (push) Blocked by required conditions
Sync Locize Translations & Create Translation PR / Sync Translation Keys with Locize (push) Waiting to run
Sync Locize Translations & Create Translation PR / Create Translation PR on Version Published (push) Blocked by required conditions
Docker Dev Images Build / publish (push) Waiting to run
Sync Helm Chart Tags / Ignore non-main push (push) Waiting to run
Sync Helm Chart Tags / Sync chart tags (push) Waiting to run
2026-09-07 23:13:39 -04:00
Danny Avila
deda142d58
🗓️ fix: Refresh the Sidebar When a Scheduled Run Fires or Settles (#15631)
* 🗓️ fix: Refresh the Sidebar When a Scheduled Run Fires or Settles

An automatic occurrence is the one generation nothing in the client starts,
so no submission, stream or event handler tells the sidebar its chat exists,
and the conversation list's five-minute staleness leaves it out of a tab
that stays focused.

The schedules panel is already polling for its cards, and `lastRun` moves at
exactly the two moments the list should be re-read — a fire adds a chat, a
settlement retitles and reorders it — so read the refresh out of that poll
rather than adding one. Only transitions refresh; a schedule first seen is
recorded in silence, so a panel opened long after a run does not refetch a
list that has held its chat all along.

This also settles the row Run Now seeds: the schedule's own completion is
what replaces the schedule name with the generated title.

* 🗓️ fix: Name the Running Occurrence's Chat So the Sidebar Can Fetch It

Four review rounds each found a new way the schedule projection diverges
from the run it stands in for: `lastRun` is not written at fire, `nextRunAt`
moves before the conversation exists, and an edit mid-flight fences the
terminal write so `lastRun` never changes at settlement. Nothing on a
schedule document means "this run's chat is listable now", so every signal
read from one was too early or too late.

Stop inferring. The list handler already reads the user's schedules and
limits in parallel; read their active run rows alongside — one user-scoped
query, joined by schedule id — and project the running occurrence's
conversation id as `activeRun`. It comes from the run's own row, so the
configRevision fence on the schedule projection cannot hide it, and it is
absent until dispatch has actually reserved a conversation.

The panel hook then hands every id the list names to the same admission
watch Run Now uses, which probes until the server has the chat and adds the
server's own row: one mechanism for both paths, idempotent per id. A run
leaving the list has settled, and the list is re-read once for the order
its settlement changed.

* 🗓️ fix: Fetch Every Chat the Schedule List Names, and Scope the Run Read

Three from codex on the rebuilt design.

A run short enough to start and settle between two polls is never seen
running, and most scheduled chats are short. The list also names the chat
the last occurrence produced, so the panel now fetches every chat the list
names that it has not seen — active or settled — through the same admission
watch. The first observation still fetches only runs in flight: a panel
opened long after a run must not go and fetch history.

Project every running occurrence rather than the latest. A `requires_action`
pause does not block the next occurrence, so a schedule can hold two, and
reducing to one hid the pause's settlement. This also deletes the picker.

The single-schedule route's run read started before ownership was known and
was scoped only by schedule id. The response was gated, but a speculative
read must stay scoped to the caller: use the user-bound query the list uses
and narrow it, rather than touch rows the caller may not own.

* 🗓️ fix: Let One Layer Decide What to Watch, and Say Which Runs Are Generating

Two from codex on the previous round.

The admission watch forgets a run whose delivery outlasted its budget so a
later announcement can retry, but the panel kept its own permanent record of
every id it had handed over and never announced one twice — so that retry
could never happen. Drop the panel's record: every chat the list names is
handed to the watch on every observation, and the watch is the one place
that decides whether an id is landed, in flight, or worth trying again.

The one thing the panel still must not do is fetch history. Opening it used
to start a watch per active occurrence, and a run parked on an approval can
sit there indefinitely and accumulate, its chat listed long ago. Carry each
occurrence's status on the wire, and on the first observation treat only a
generating occurrence as in flight; a settled one and a paused one are
history, and a pause resolving later still re-reads the list.

* 🗓️ fix: Read Only Generating Runs for the List, and Hand Over Only Those

Three from codex, and this round removes more than it adds.

The list's run read returned every paused row, unbounded, on each poll.
`ScheduleRun` is indexed by status rather than by user, `requires_action`
rows accumulate for as long as their approvals wait, and `started` rows are
bounded globally by the capacity slots — so the read now asks for `started`
alone, through an optional status narrowing on the existing method. A pause
resolving is still seen through `lastRun`; what this gives up is only the
corner where that projection is fenced by a mid-flight edit.

A run that failed during initialization settles with its reserved id but
never writes a conversation, so its route answers 404 for good — and handing
settled ids to the watch meant re-probing it every poll, forever. The watch
only ever needed generating runs: a settled one is already covered by the
list re-read its settlement triggers, which brings the server's own row.

A deleted schedule leaves the list, and comparing only the schedules still
present missed the abort-and-settle that deletion performs on its run. A
schedule vanishing now counts as its state moving.

* 🗓️ fix: Compare a New Schedule Against Idle, and Bound the Landed-Run Record

Two from codex on the head GitHub replayed onto dev.

A schedule created in another tab can arrive in this client's list with its
first run already settled, and a comparison that walks only the schedules
seen before never looked at it. Compare a newcomer against the idle state:
one that arrives with a run behind it is news the list should hear, one
that arrives with no run yet is not.

The watch remembers every run it landed so a later announcement is a no-op,
and a panel that stays mounted sees a new id per occurrence for as long as
schedules keep firing. Keep the most recent 256: a run is announced only
while in flight, and none is still in flight by the time that many have
started after it, so the oldest can go without a watch ever running twice.

* 🗓️ fix: Re-read on First Sight Only When a Run Outran the Sidebar, and Never Evict a Live Watch

Two from codex.

"The first observation is silent" assumed the sidebar's lists had been read
after any run they should hold, and that is not guaranteed: open the panel
after a run fired and finished with it closed, and the card says the run
succeeded while the sidebar was read before it ever fired. The first
observation has no earlier one to compare against, so compare against the
sidebar instead — re-read only if some run fired after the lists were last
read, which is precisely the case where they lack its chat, and stay silent
otherwise.

The landed-run cap assumed fewer than 256 generations could be in flight at
once, and the config schema bounds neither maxPerUser nor fireConcurrency.
Keep in-flight watches in their own set, bounded by real concurrency and
never evicted — a second watch for a run still being watched is the one
thing that set exists to prevent — and cap only the landed record, where a
forgotten id costs at most one extra probe.

* 🗓️ fix: Re-read on First Sight When Anything Has Run, Release Runs the List Drops, Observe Every Poll

Three from codex, each a defect in an earlier fix, and each answered by
removing something.

Comparing a run's fire time against the list cache's `dataUpdatedAt` was
measuring cache writes, not server reads: a foreground chat inserted after
the run advanced the stamp without adding the scheduled chat. There is no
timestamp for "last read from the server", so stop guessing — the first
observation re-reads if any schedule has run at all, and one refetch on
opening the panel is the honest price.

A run can be landed and still generating for minutes, announced on every
poll; a count cap evicted it and restarted its probes. The hook is the one
place that sees an occurrence leave the list, so it releases the id then,
and the watch remembers a run exactly as long as something can announce it.
No cap.

React Query hands back the same array for a deep-equal poll, so an effect
keyed on the data never ran again while a run stayed in flight — and a watch
that gave up was never asked to try again. Key on the poll's timestamp.

* fix: Release Scheduled Run Retention Across Admission and Unmount 🧹
2026-09-06 15:28:11 -04:00
Danny Avila
aeac7a2065
🪪 feat: Add Per-Agent Git Identity (#15628)
* feat: Add per-agent Git identity

* fix: annotate git identity schema

* fix: complete git identity controls

* fix: clear git identity on version restore

* fix: Require a complete agent Git identity

* fix: Complete Git Identity Across Bash And Form Validation

* fix: Discard Incomplete Hidden Git Identity Edits

* fix: validate Git identity across panel lifecycle

* fix: preserve saved Git identity when discarding drafts
2026-09-05 23:56:23 -04:00
Marco Beretta
971e076015
🎛 refactor: Replace Agent Skills Toggles With a Three-Way Mode Control (#15520)
* refactor: replace agent skills toggles with a three-way mode control

The Skills section in the agent builder exposed one dependent choice as two
peer switches. "Use all skills" only meant anything while "Enable skills" was
on, and turning it on hid the skill list, so the user could not see what "all"
covered. Two Add affordances competed for the same action and both labels
leaned on a tooltip to explain themselves.

Replace both switches with a single segmented control over the SkillsScope
enum that already existed. Off, All, and Selected now write skills_enabled and
skills_scope outright, so the mode is the only source of truth for catalog
exposure and the allowlist survives a round trip through All.

Extract the section into SkillsSection. That drops the badgeText, showAdd,
showBody, and children props SelectedSection only carried to bend the tools
layout into a skills layout, and it retires skillsSelectionTransition, which
inferred the capability flags from the allowlist length and flipped skills off
whenever the last skill was removed.

All mode reports how many skills the deployment exposes and expands a
read-only list, so the catalog is never hidden. Selected mode keeps one Add
affordance: the dashed card while nothing is chosen, a header button once rows
exist.

Legacy agents keep loading through resolveAgentSkillsScope. The one ambiguous
persisted shape, skills_enabled with no explicit scope, is normalized on load
without dirtying the form so the next save records the mode explicitly.

Also give the shared Radio the roving tabindex and arrow key handling a
radiogroup owes its users; every segment was previously its own tab stop with
no way to change the selection from the keyboard.

* chore: add temporary PR comparison screenshots

* chore: drop temporary PR comparison screenshots

* fix: address PR review bot findings on the skills mode control

Codex (P1):
- Compare the resolved mode, not the raw persisted scope, before treating a
  segment click as a no-op. An agent left with `skills_enabled: false` and a
  retained `skills_scope` renders as Off, so matching on the raw field
  swallowed the click that re-enables it while the shared Radio moved its own
  selection, leaving the form clean and the disabled state persisted.
- Exempt explicit scopes from the legacy allowlist normalization in
  AgentSelect. Off deliberately keeps the allowlist so returning to Selected
  restores it, and flipping `skills_enabled` there persisted skills-enabled on
  an agent shown as Off, which skillDeps reads as permission to inject the
  skill-authoring tools.

Codex (P2):
- Report a failed catalog request instead of rendering it as an empty catalog.
  All mode showed "0 skills available" over an empty list when /api/skills
  failed; it now renders the picker's alert with a retry.
- Stop paging the whole catalog before the list is opened. Merely viewing an
  All-scoped agent walked every cursor at 100 records a page and retained each
  summary. Pagination now waits for the expansion, and the header reports the
  loaded count as a floor while more pages remain.
- Honor an explicit `skills_scope` in both skill cleanup paths. Deleting the
  last allowlisted skill, or pruning a dangling id on save, forced
  `skills_enabled` to false without consulting the scope, so an All-scoped
  agent that had once picked a single skill was silently switched off. The
  fail-closed inference now applies only to legacy agents that persist no
  scope at all.

Copilot:
- Skip the onChange when Home or End lands on the already-checked segment.
  Focus still follows the key; only a selection that does not move stops
  firing a form write.

* fix: address second round of PR review bot findings

Codex (P1):
- Normalize authoring left on under a resolved Off. An agent carrying
  `skills_enabled: false` with `skill_authoring_enabled: true` renders as Off,
  yet skillDeps enables the skill-authoring tools for either flag, so it ran
  with skills the section reported as disabled. Clicking the already-selected
  Off segment is a no-op, so nothing ever cleared it. The section now clears
  the flag on load without dirtying the form, matching what selecting Off
  already writes.

Codex (P2):
- Apply the scope-aware pruning in createAgent and revertAgentVersion too, not
  only updateAgent. Restoring an All-scoped version whose skills have since
  been deleted silently came back as Off, and createAgent had the same
  unconditional branch for ids that go stale before the write lands.
- Collapse the duplicate first-page skills request. ToolsSection listed skills
  through useListSkillsQuery while the section and the picker used
  useSkillsInfiniteQuery, so the same endpoint was fetched under two React
  Query keys and could not be deduplicated. All three consumers now share the
  infinite query, which leaves one cache entry instead of two.

Also sorts the imports the repo's checker flagged in SkillsSection,
ToolsSection and the section's spec.

* fix: clear the skills master flag under a resolved Off

`skills_enabled: true` with `skills_scope: none` is a shape the API accepts,
and the resolver reports it as Off. `skillDeps` reads the master flag on its
own as permission to expose the skill-authoring tools, so the agent ran with a
capability the section said was disabled, and because clicking the
already-selected Off segment is a no-op there was no way to clear it without
cycling through another mode.

The normalization that already cleared `skill_authoring_enabled` now clears
both flags whenever the resolved mode is Off, still without dirtying the form.
Neither write fires while a catalog mode is active.

* fix: disable skills for an explicit none scope in allowlist cleanup

The scope-aware fail-closed rule exempted every explicit scope, which spared
one shape it should not have: `skills_enabled: true` with `skills_scope: none`.
The API accepts that combination, the builder renders it as Off, and skillDeps
reads the master flag on its own as permission to expose the skill-authoring
tools. Cleanup runs server side without the builder mounted, so the frontend
normalization cannot repair it.

Only `all` and `selected` opt out now, since each already defines what an
empty allowlist means. The three self-heal paths in agent.ts share a
`requiresSkillsDisable` predicate, and the Mongo filter in skill.ts uses
`$nin: [all, selected]`, which still matches an absent field for legacy
agents.

* test: align the skills e2e spec with the explicit-mode contract

Three assertions still encoded the removed toggle behavior:

- The remove button is labelled `Remove {name}` in the new section, not
  `Remove from agent`, so both removal steps waited 180s for a button that no
  longer exists. That was the timeout, not a failed expectation.
- Emptying the allowlist no longer infers Off. The save writes `skills: []`
  and leaves the mode at Selected, which resolves to no skills at runtime, so
  the spec now asserts `skills_enabled: true` with `skills_scope: 'selected'`
  and keeps the existing proof that the picker and the run expose nothing.
- All leaves `skills` untouched, so a fresh agent serializes no allowlist at
  all rather than an empty one: `composeAgentUpdatePayload` always carries the
  four skill fields and `JSON.stringify` drops the undefined ones. Create and
  Off now assert `skills_scope` instead of an empty array.

Adds `skills_scope` to the spec's payload and detail types.

* fix: address third round of PR review bot findings

All three are P2 accuracy and cost issues in All mode:

- Count only the skills the runtime would inject. The list endpoint returns
  every skill the user can VIEW, while catalog resolution filters each through
  `resolveSkillActive`, so a shared skill left inactive was counted and listed
  as available even though the agent never receives it. The section now shares
  the `isActive` predicate the chat-side picker already filters on, so the
  count cannot promise more than the runtime delivers.
- Report loading instead of an empty catalog. With a cold query `data` is
  undefined and the count derived to "0 skills available", presenting an empty
  deployment while it was merely loading.
- Skip the allowlist lookups outside Selected. `useResolvedSkills` issues one
  `getSkill` per allowlisted id missing from the first catalog page, and since
  All now retains previous picks, merely viewing such an agent could fire
  hundreds of requests for rows that mode never renders. The fallback is gated
  on Selected, where the rows exist.

* fix: gate the All catalog on resolved skill states

`useSkillActiveState` substitutes an empty override map while its own request
is in flight and after it fails, so `isActive` reports every skill active in
both states. The section consumed the predicate but discarded the status, so a
deliberately deactivated owned or deployment skill was counted and listed as
available: briefly during load, and permanently after an error.

The count now waits for both requests, and a failure in either renders the
alert instead of a count the runtime would not honor. Retry refetches both,
which needed the hook to expose its `refetch`.

* fix: resolve the config behind defaultActiveOnShare before filtering

`useSkillActiveState` reads `defaultActiveOnShare` from the interface config
and falls back to `false` while that request is pending or after it fails, so
a deployment configured with `true` had its shared skills filtered out of the
All count until the config landed, and permanently on error. The section only
gated on the skills and skill-states queries.

Both dependencies feed `isActive`, so the hook now reports them as one
signal: `isLoading` and `isError` cover the override map and the config
alike, and `refetch` retries both. A consumer gating on the pair can no
longer miss a dependency, which is how the previous gap arose. The section is
the only consumer of those flags.

Also composes the retry from the shared `Button` (`outline`, `sm`), matching
the picker's own retry instead of a feature-local recipe.

* fix: report what the All catalog loaded, not a lower bound

`hasNextPage` tracks raw pagination while the rows are filtered to the
runtime-active skills, so the two disagree: 30 active skills followed only by
deactivated ones made the header read "30+ skills available" when 30 was the
true total, and it stayed wrong until the user expanded the list and finished
paging.

While pages remain the header now describes what has been loaded rather than
claiming a floor on what is available, so it cannot overstate the catalog. The
exact "N skills available" phrasing still lands once the catalog is complete.
`com_ui_skills_available_count_more` is replaced by singular and plural loaded
keys.
2026-09-06 03:32:56 +02:00
Marco Beretta
8144574b43
♿ feat: High Contrast Light and Dark Appearance Modes (#15178)
Some checks are pending
Backend Unit Tests / Build packages (push) Waiting to run
Backend Unit Tests / Codegraph select (push) Waiting to run
Backend Unit Tests / TypeScript type checks (push) Blocked by required conditions
Backend Unit Tests / Circular dependency checks (push) Waiting to run
Backend Unit Tests / Tests: api (shard 1/3) (push) Blocked by required conditions
Backend Unit Tests / Tests: api (shard 2/3) (push) Blocked by required conditions
Backend Unit Tests / Tests: api (shard 3/3) (push) Blocked by required conditions
Backend Unit Tests / Tests: data-provider (push) Blocked by required conditions
Backend Unit Tests / Tests: data-schemas (push) Blocked by required conditions
Backend Unit Tests / Tests: @librechat/api (shard 1/4) (push) Blocked by required conditions
Backend Unit Tests / Tests: @librechat/api (shard 2/4) (push) Blocked by required conditions
Backend Unit Tests / Tests: @librechat/api (shard 3/4) (push) Blocked by required conditions
Backend Unit Tests / Tests: @librechat/api (shard 4/4) (push) Blocked by required conditions
Codegraph E2E Votes / vote (full suite) (push) Waiting to run
Docker Dev Branch Images Build / publish (push) Waiting to run
Frontend Unit Tests / Codegraph select (push) Waiting to run
Frontend Unit Tests / Build packages (push) Waiting to run
Frontend Unit Tests / TypeScript type checks (client) (push) Blocked by required conditions
Frontend Unit Tests / Tests: @librechat/client (push) Blocked by required conditions
Frontend Unit Tests / Tests: Ubuntu (shard 1/2) (push) Blocked by required conditions
Frontend Unit Tests / Tests: Ubuntu (shard 2/2) (push) Blocked by required conditions
Frontend Unit Tests / Vite build verification (push) Blocked by required conditions
Sync Locize Translations & Create Translation PR / Sync Translation Keys with Locize (push) Waiting to run
Sync Locize Translations & Create Translation PR / Create Translation PR on Version Published (push) Blocked by required conditions
* refactor: read status fill labels from the theme token

Feature call sites paint the label on bg-surface-destructive,
bg-surface-submit and the bg-status-*-strong fills with the raw
text-white utility, while the shared Button, IconButton and Toast
primitives already use text-text-on-status. The token exists so the
label colour can differ per theme mode, and hard-coding white pins it.

Point every call site at text-text-on-status, and compose
DeleteIconButton onto the Button destructive variant rather than
re-declaring the fill classes. Five call sites paired a status fill with
the wrong token entirely (text-text-secondary, text-text-inverted) or
used the non-strong hue as a solid fill; those move to the -strong fill
with the on-status label. Tooltip.css and the app stylesheet copy of the
same rule read colour from --text-primary instead of literal black and
white, and the dead outline: var(--bg-surface-hover) in Dropdown.css
becomes a real 2px token outline, since that variable is not declared
anywhere and the focus ring never rendered.

--text-on-status is 255 255 255 in both shipped themes, so this is a
no-op for the light and dark palettes.

* feat: add high contrast light and dark appearance modes

Adds two appearance modes to the theme setting for users who need more
than the standard palettes, plus contrast resolution for the system mode.

The palettes are complete IThemeRGB maps rather than partial overrides,
because a token left to fall back to defaultTheme or darkTheme would
reintroduce a mid-grey the mode exists to eliminate. They resolve as a
built-in ThemeDefinition that outranks a deployment's custom theme, since
a contrast choice is an accessibility need rather than a branding
preference.

Text clears WCAG AAA on every surface it can render on, including the
hover and active fills. Borders, rings and series marks clear the 3:1
non-text floor. Solid fills clear AAA against both their label and the
page, which is what the per-mode text-on-status token buys: each mode
paints its fills on the far side of its own canvas, dark fills under a
white label on white and bright fills under a black label on black, so
the two ratios coincide instead of pulling apart.

The system mode now follows prefers-contrast the same way it already
follows prefers-color-scheme, so a user who has switched on Increase
contrast or Contrast themes gets the accessible palette without first
finding this setting. isDark reports high-contrast-dark as dark;
isHighContrast answers what the user picked, which is what the theme
toggle preserves when it flips the scheme; resolvesToHighContrast answers
whether the palette applies.

The icon toggle keeps a contrast choice when it flips the scheme, and
announces which of the four states it landed in.

* feat: cover the non-tokenized chrome in high contrast

The token layer reaches every IThemeRGB surface, but a good deal of the
stylesheet predates it and holds literals the palette cannot move. Under
html.high-contrast:

The typography variable block is 32 hard-coded hex values, and message
content carries prose dark:prose-invert, so the invert half is what dark
mode reads. Both halves now resolve from tokens: prose links were 3.77:1
on a black canvas and 5.40:1 on a white one, and every rule and border in
the set sat below the 3:1 non-text floor in one mode or the other.

The highlight.js palette is raw hex and tops out at AA, worst cases
4.84:1 for dark attributes and 5.32:1 for the alpha-blended dark comment;
these reuse the palette's own hues at AAA on each canvas.

Focus perimeters go to 3px and read the ink token rather than literal
black and white. Selection was left to UA system colours and is now an
explicit inversion. Ring opacity is pinned at full strength for the
controls that halve their own. Popover and tooltip elevation trades a
black shadow, which separates against nothing on a black surface, for a
token border. Prose links carry an underline so they are not distinguished
by colour alone. The switch track, the scrollbar thumb and track, the
markdown structure borders and the body background that .dark pins to
#212121 all move onto tokens.

* feat: draw panel and dialog edges in high contrast

Adds a high-contrast: Tailwind variant to the shared preset, so styling
can be scoped to the contrast modes the way dark: is scoped to the dark
class. Deliberately not named contrast-more, which Tailwind already ships
as the raw prefers-contrast media query: this follows the resolved app
mode, so it covers an explicit choice as well as the OS preference.

Two surfaces rely on a colour step that high contrast removes. The
sidebar's right edge is a resize separator that is transparent until
hovered, which is fine where surface-primary-alt carries the boundary but
leaves no edge at all when both sides are the same pure white or black.
Dialogs are borderless and lean on a black shadow, which separates
against nothing on a black canvas.

Both now draw a real edge in the contrast modes only. The dialog change
sits in the Radix content primitives and the two shared templates so the
feature dialogs inherit it, plus the settings panel, which is Headless UI
and bypasses them. OGDialogTemplate carries border-none, so the variant
restores the border style as well as the width.

* fix: address PR review bot findings

chatgpt-codex-connector:

Move the syntax palette into semantic theme tokens. The highlight.js
colours were raw hex in style.css, so a palette change had to be made
twice and neither copy was covered by the registry's completeness check
or any contrast test. They are eight rgb-syntax-* tokens now, declared
across all four theme maps and consumed as CSS variables, which lets the
whole high-contrast block for highlight.js go away. Two guardrails come
with them: the contrast modes assert AAA for every syntax role on the
code surface, and the stylesheet is pinned to the runtime themes with no
hard-coded hex allowed in an .hljs rule.

Two values could not be carried over literally: the dark comment and meta
colours were alpha-blended whites, so they are stored as the flattened
equivalents over the #212121 code surface. Everything else, including the
dark code text, renders byte-identical to before.

Declare Tailwind as a preset dependency. The preset ships as a raw file
through the ./tailwind-preset export while tailwindcss is only a
devDependency, so require('tailwindcss/plugin') could fail to resolve for
a consumer under pnpm or Yarn PnP. Dropped the require instead of adding
a peer dependency, since Tailwind accepts a plain function in plugins and
the wrapper only adds option handling this variant does not use.

Route remaining solid controls through semantic status roles. The
assistant builder kept bg-red-600/bg-red-700 and bg-green-500 with
text-white, which the built-in palette cannot reach, leaving them below
the mode's contract at roughly 2.28:1 for the green controls. All five
now use the destructive and submit surfaces with the on-status label, and
the two that also suppressed their focus ring get a semantic one.

* fix: address second round of PR review bot findings

chatgpt-codex-connector:

Migrate the builder submit to semantic status colors. AssistantPanel's
submit button kept btn-primary, which hardcodes #10a37f and white in
style.css, alongside a raw green hover and focus border. It uses the
submit surface with the on-status label now, and drops btn-primary rather
than fighting its specificity. Left .btn-primary itself alone: restyling
every primary button in the app is not this PR's business.

Give provider avatars contrast-safe brand colors. The white glyph sat at
2.30:1 on OpenAI green, 2.45:1 on Anthropic tan and 3.41:1 on GPT-4
purple. brands was a single theme-wide set, and a brand fill has to carry
a glyph and stand out from the canvas, both of which flip between modes,
so one set cannot serve both at enhanced contrast. ThemeModeDefinition
takes an optional brands block now, resolved after the theme-wide set, and
high contrast declares dark tints under a white glyph for light and bright
tints under a black glyph for dark. Worst pair is 8.76:1 for both the
glyph and the silhouette, with provider hues still recognisable. Brand
validation is shared between the two levels rather than duplicated.

Raise active-surface contrast against the canvas. The active fills sat at
1.48:1 and 1.98:1 on white, 1.93:1 and 2.91:1 on black, and since every
ink token collapses to one colour here, dozens of call sites were left
conveying the selected state with a sub-3:1 fill alone. The fills now
clear 1.4.11 at 3.14:1 to 4.29:1.

That costs the AAA label on those two fills, and the trade is unavoidable:
above the luminance where a fill clears the 3:1 state floor, its own label
cannot reach 7:1, and below it the state disappears. No integer grey
satisfies both. 1.4.11 is Level AA and 1.4.6 is Level AAA, so the state
wins and the labels land at 4.89:1 to 6.69:1. The spec now separates
canvas and hover surfaces, which keep AAA, from the active fills, which
assert the state floor and an AA label.

* fix: address third round of PR review bot findings

chatgpt-codex-connector:

Drop the overriding btn-primary class from Select. My previous round
assumed Tailwind utilities would win here by source order; they do not.
In the built stylesheet .btn-primary sits at byte 136886 and
.bg-surface-submit at 69233, equal specificity, so the hardcoded green won
and that button had never followed the palette. Removed btn-primary, the
same way as the submit button beside it.

Preserve contrast mode when theming Mermaid output. isDark collapses a
contrast mode onto the same boolean, so the cache key was unchanged across
light to high-contrast light and an already-rendered diagram survived the
switch untouched. Both diagram paths take the resolved contrast now: the
cache key gains a suffix, and the config carries themeVariables built from
the contrast palette, since Mermaid's own neutral and dark themes hold a
palette no theme token can reach. Diagrams render on the canvas with ink
nodes, ink edges and ink labels, which is 21:1 for every mark and label.

Artifact diagrams go through the same helper: getMermaidFiles takes the
flag, switches the generated document to Mermaid's base theme, which is
the only built-in that honours themeVariables, and paints the page with
the contrast canvas instead of the fixed #212121 or #FFFFFF. The helper
returns undefined outside the contrast modes, so standard themes keep
their current rendering exactly.

* fix: publish resolved contrast through theme context

chatgpt-codex-connector: notify consumers when system contrast changes.

Under `system` the contrast comes from a media query, so `theme` stays
`system` when the OS preference flips and React has nothing to rerender
on. The listener I added last round only touched the root element, which
is enough for CSS but not for anything computing from the resolved value:
the Mermaid cache key and the artifact files would have kept the old
palette while the rest of the app switched.

ThemeContext exposes `highContrast` now, held as state and set wherever
the mode is applied, which covers both an explicit choice and a media
change. `useMermaid` and `useArtifactProps` read it from context instead
of deriving it locally. `resolvesToHighContrast` stays exported as the
predicate the provider itself resolves with.

The existing media-change test now also asserts the context value moves,
not just the class on the root, since the DOM assertion alone passed
before this fix.

* fix: publish the resolved scheme through theme context too

chatgpt-codex-connector: publish resolved scheme changes through context.

Same staleness as the contrast case, one field over. With the mode at
`system` and contrast already on, an OS colour-scheme change reapplied the
DOM palette but `setHighContrast(true)` was a no-op and `theme` stayed
`system`, so no consumer rerendered and the diagram hooks kept the
previous `isDark` result. This half predates the PR: those hooks derived
the scheme from `theme` before it too, so a scheme flip under `system` has
never reached them.

ThemeContext exposes `resolvedMode` alongside `highContrast`, both set
wherever the mode is applied. The two diagram hooks read the scheme from
context instead of calling isDark themselves.

The other isDark call sites are left alone: none of them cache, so they
recompute whenever something else rerenders them, and sweeping every
consumer onto the context value is a wider change than this PR should
carry.

* fix: address sixth round of PR review bot findings

chatgpt-codex-connector:

Restore a visible focus cue for dropdown options. `.select-item` blanks
its own outline with `outline: none !important`, so the contrast focus
rule never landed and the navigated option was marked only by a
surface-hover fill at 1.48:1 on white and 1.93:1 on black. Hover stays
deliberately soft in these palettes, since it is transient and can afford
an AAA label, so the cue has to be a perimeter: a 3px inset outline on
the active and focused option, matching the !important it has to beat.
Verified with real keyboard navigation that the outline resolves to the
ink colour in both modes.

Apply contrast tokens to artifact controls. getMermaidFiles took the
contrast flag last round but still built its controls from the standard
palettes, whose fixed greys leave the zoom label below AAA and whose
10%-alpha border and divider sit below the 3:1 non-text floor against a
pure canvas. The controls now come from the same tokens as the diagram:
canvas fill, ink border and label, the active surface for hover, and no
shadow, since a black shadow separates against nothing there.

* fix: ignore undefined brand overrides when resolving a theme

chatgpt-codex-connector: ignore undefined mode brand overrides.

Partial<IThemeBrands> lets a key be present with undefined, and the brand
validator accepts it, so spreading a mode block over the inherited set
could overwrite a brand with nothing. Colors survive this because
mapColors skips undefined, but every brand token is written to the DOM
unconditionally, so the avatar would have lost its fill instead of falling
back the way a partial theme promises.

Both spreads now drop undefined entries, which covers the theme-wide set
as well as the mode block, and keeps ResolvedThemeDefinition.brands honest
about being fully populated.

* fix: address eighth round of PR review bot findings

chatgpt-codex-connector:

Guard media queries during server rendering. The state initializers I
added read prefers-color-scheme and prefers-contrast synchronously during
render, so a server-rendered provider threw before producing markup. That
was a regression: nothing in the provider read a media query during render
before. Both reads go through a helper that returns false without a
window, the way the storage helpers already tolerate its absence, so the
server renders the light palette with no contrast override and hydration
applies the real values. The subscription effect is guarded too, which
also covers a host that has no matchMedia at all.

Keep Mermaid pie slices distinct in contrast mode. My own regression: the
contrast variables set primaryColor, secondaryColor and tertiaryColor to
the canvas, and mermaid derives pie1 through pie3 from those, so every
slice collapsed into the background and into its neighbours. The pie slots
come from the palette's series ramp now, which is the categorical scale
already built to clear the mark floor on that canvas and stay separable
under simulated deuteranopia. Twelve slots wrap over seven, as mermaid's
own palettes do.

Remove the raw white override from bookmark submit. The save button passed
text-white into the shared submit variant, which tailwind-merge resolved
in favour of the local class, leaving white on the bright submit fill at
1.40:1. Dropped the override so the variant's token applies.

Preserve the high-contrast canvas in screenshot exports. The export chose
between two fixed colours, so a contrast export was filled with #171717
instead of the mode's canvas. It reads the resolved canvas token now, with
the old pair as the fallback.

Theme markdown artifact previews in contrast modes. useArtifactProps
forwarded the appearance only to the mermaid branch. getMarkdownFiles
takes it now and appends a contrast block derived from the palette; being
unconditional it also outranks the sheet's own prefers-color-scheme query,
so an explicit contrast choice is honoured whatever the OS reports.

* fix: migrate agent version controls to semantic status roles

chatgpt-codex-connector: migrate agent-version controls to semantic status
roles.

The active-version marker and the Restore action used raw green fills with
text-white, about 2.28:1 and 3.30:1, and the built-in palette cannot reach
either. The marker takes the strong success fill with the on-status label,
and Restore takes the submit surface, matching the assistant builder
migration.

Also finished the rest of that component's active-version treatment rather
than leaving it half migrated: the card border and tint, the title, and
the Current chip now use the success subtle, border and strong roles. The
same indicator was expressing one state through both raw and semantic
colour otherwise.

ToolCard's native badge and selected tick were the same shape, raw emerald
under white, so they move to the strong success fill too.

Left alone: the HTTP method badges in ActionsTable/Columns.tsx, the
tool-kind colours in items/icons.ts, and CategoryFilter's active border.
Those are categorical identity rather than status, so mapping them onto
status roles would be wrong; they need the treatment provider brands got,
which is its own change.

* fix: address tenth round of PR review bot findings

chatgpt-codex-connector:

Use opposing ink for pie-section labels. My own regression from the slice
fix: the series ramp deliberately sits on the far side of the canvas so
slices are visible, which makes the canvas ink the wrong colour for a
label drawn inside one. Section labels take the on-status ink now, the
same colour the palette uses for a label on any solid fill, while the
title and legend keep the canvas ink because they are drawn on the canvas.

Preserve the blue initializing status in standard themes. Also mine, from
the first round: I moved this dot to bg-status-info-strong, and the strong
info slot is a neutral grey in both standard palettes, so a blue
indicator turned grey. Back on bg-status-info, with the pulse taking the
on-status ink so it inverts with the fill rather than being a fixed white.

Route action method badges through contrast tokens. I declined this last
round as categorical identity rather than status; the reviewer's framing
is better. The series ramp is precisely the home for categorical identity,
it is contrast-checked per mode, and its slots are separable under
simulated deuteranopia, so the verbs map onto it while staying close to
their conventional colours. `delete` keeps a red by taking the error role,
since destructive genuinely is a status.

* fix: seed published theme values without reading a media query

chatgpt-codex-connector: defer media-query resolution until after
hydration.

Guarding window stopped the crash but left the server and the first client
render disagreeing: the server produced light and false while the client
resolved the OS preferences during that same first render, so any child
reading the published values hydrated against different context.

Both are seeded from the mode string alone now. An explicit mode needs no
query, so it publishes immediately and stays consistent with isDark; only
system starts at the light palette with no contrast override, on both
sides, and applyThemeMode publishes the resolved values from its effect
after hydration.

The pre-existing localStorage read in getInitialTheme has the same shape
of problem for `theme` itself, but changing where the stored mode is read
is a wider change than this and belongs on its own.

* fix: address eleventh round of PR review bot findings

chatgpt-codex-connector:
- keep the active version timeline rail opaque so it clears the 3:1 mark
  floor in both high contrast modes instead of compositing to ~2:1
- balance the dark success fill (#088759) so the selected-tool marker,
  the timeline rail and the prompt-version chip keep a visible silhouette
  on the #212121 panel while the white label stays at WCAG AA
- include the resolved scheme and contrast in the mermaid cache key even
  when a custom theme is supplied, so a contrast switch re-renders

* fix: address twelfth round of PR review bot findings

chatgpt-codex-connector:
- select mermaid's base theme ahead of a caller-supplied theme in the
  contrast modes, after the config spread, so the contrast variables
  actually reach the diagram
- ship the tooltip and popover contrast edges in the package stylesheets
  they belong to, so a consumer importing @librechat/client gets them

* chore: sort imports in the theme registry, provider and specs

* fix: override the markdown preview table header in contrast modes

The appended contrast block answered the ink, the canvas and the cell
borders but not the thead tint, so an explicit contrast choice on the
opposite OS scheme left white text on #f6f8fa or black text on #161b22.
The header now takes the palette canvas and reads through its bold cells
and border, and a new guardrail asserts every colour the base sheet sets
inside prefers-color-scheme is answered by the override.

* fix: address thirteenth round of PR review bot findings

chatgpt-codex-connector:
- keep an OS-requested contrast when the scheme toggle runs under the
  system mode, by reading the resolved contrast rather than the stored
  appearance name
- move the unchecked switch track into the theme token registry so the
  shared Switch keeps its track for consumers of @librechat/client, and
  drop the SPA-only high-contrast overrides it no longer needs

* fix: address fourteenth round of PR review bot findings

chatgpt-codex-connector:
- migrate agent and assistant validation states off raw red utilities and
  onto the destructive text and border roles, so the messages, the field
  outlines and the required markers follow the contrast palettes
- theme the file-source badges with series slots and the status label ink
  instead of fixed blue and yellow fills behind an alpha

* fix: paint the loading canvas from the resolved appearance

The bootstrap in index.html read the stored mode's scheme but not the OS
contrast, so system plus prefers-contrast: more flashed the standard-dark
#0d0d0d behind the pure-black palette for the whole application load. An
unset mode also painted white on a dark OS, though getInitialTheme
resolves it as system. Both branches now resolve the same way the theme
provider does, and a new spec runs the shipped script to prove it.

* fix: theme the markdown renderer failure message

The notice painted when the marked CDN does not load carried an inline
color:#e53e3e, the one colour the appended contrast block could not
reach, leaving it at 4.13:1 on the high-contrast light canvas. It is a
.markdown-error rule now, which the base sheet declares with the current
red and the contrast block overrides with the palette's destructive ink.

* fix: address PR review bot findings

chatgpt-codex-connector:

- Route shortcut, steer, warning, and help-link states through semantic theme roles.
- Add contrast-safe Mermaid Gantt variables for task, status, section, grid, and label states.
- Theme Mermaid artifact render failures with destructive text colors.
- Apply native color-scheme behavior through ThemeProvider and restore host values.

* fix: address follow-up PR review findings

chatgpt-codex-connector:

- Route conflicting shortcuts through the semantic warning border.
- Give checked composer capabilities opaque semantic series borders.
- Move search results and their fade onto the presentation token.
- Restore a visible semantic focus ring on the virtualized results list.

* fix: address third-round review findings

* fix: address fourth-round review findings

* fix: align accessibility theme after dev rebase

* fix: address PR review bot findings

chatgpt-codex-connector:
- Route the steer and queue identities through semantic `status-warning` /
  `status-info` roles in `Chat/Steering/identity.ts`, so the receipt label,
  its pulse dot and every Zap/Clock mark move with the theme instead of
  staying pinned to amber-500/600 and cyan-500. Fixes the high-contrast
  light failures (3.19:1 label text, 1.97:1 and 2.43:1 marks) and the
  default light label, which was also below the 4.5:1 text floor.
- Pair each Monaco theme with the canvas it paints in `ArtifactCodeEditor`,
  so the transparent `@monaco-editor/react` loading view no longer flashes
  the light `surface-code` fill before the `vs-dark` editor mounts outside
  high contrast.

* fix: declare the new dev theme tokens in the high contrast palettes

dev added rgb-text-muted and the two chart widget tokens after this branch
last rebased. The high contrast palettes did not name them, so the widget
canvas and muted text fell back to the standard themes' greys, which the
palette coverage spec caught. Both modes now take the plain canvas, pure
ink and a pure-ink edge, and the contrast tables cover the new tokens.

* style: sort imports in the specs this branch touched

The added imports put ArtifactCodeEditor, ShortcutRecorder and OptionToggle
out of the repo's import order, which the static-checks import-sort gate
rejected.

* fix: address fifth-round review findings

chatgpt-codex-connector:
- Preserve the pressed treatment on a disabled OptionToggle, so a tool made
  programmatic-only keeps showing the Intent state aria-pressed still reports.
- Expose high contrast from the auth appearance control: the scheme toggle
  preserves contrast but never introduces it, so a logged-out user had no way
  to reach the palette. Adds a contrast toggle beside it, and the scheme toggle
  now shows the scheme rather than duplicating the contrast glyph.
- Resolve system contrast from prefers-contrast: custom and forced-colors:
  active as well as more, since a Windows Contrast Theme never reports more.
  Mirrored in the index.html loading bootstrap and the theme README.
- Route the remaining raw palette utilities through theme roles: agent success
  checks (Action, AdvancedPanel, ToolSection) to status-success, the favorited
  tool star to series-4, the orchestration Beta pill to brand-purple, the
  temporary-composer accent to series-6, and the pending quote marks to
  status-info.
- Override the legacy btn-neutral chrome in the contrast block, whose
  hard-coded edge and hover fill sat at 1.1:1 and 2.09:1 on the new canvases.

Rebased onto dev.

* fix: mark the selected tab and theme placeholder ink in contrast modes

Two states rendered invisibly under the high contrast palettes, both verified
in a browser before and after:

- Radix tab triggers mark selection with `surface-tertiary`, which is the plain
  canvas here, so the selected Settings section computed to the same
  rgb(0 0 0) fill as every unselected sibling. Give it the same inset perimeter
  `.select-item` already uses for its navigated option.
- Tailwind preflight paints `::placeholder` with the raw gray-400 #999696, and
  a field that omits a `placeholder:` utility inherits it: the Custom
  Instructions textarea measured 4.06:1 on the black canvas and 3.08:1 on the
  white one, both under the 1.4.3 floor for what is ordinary text. Route it
  through `text-secondary`, which is pure ink in both palettes.

* fix: mark selection with a perimeter and lighten the contrast selected fills

The selected fill was carrying WCAG 1.4.11 on its own, which forced it dark
enough (#8c8c8c light, #5c5c5c dark) that its own label capped at 6.25:1 and
6.69:1 — under the AAA these modes exist for, and visibly heavy on the light
canvas. The `.select-item` rule already showed the better answer: mark the
state with a `text-primary` perimeter at 21:1 and let the fill stay light.

Generalise that rule to the two families it missed — every other Ariakit menu,
whose navigated row carried only a `surface-hover` fill at 1.48:1 on white, and
any row painted with the selected-surface role — then lighten the fills behind
it: #e3e3e3 / #d4d4d4 light and #2b2b2b / #3d3d3d dark, all four AAA under
their label. `-alt` meets `surface-hover`, as it does in the standard palettes.
Inline `span` is excluded so a highlighted citation passage is not boxed per
wrapped fragment.

The palette contract moves with it: `activeFills` is now held to an AAA label
rather than to 3:1 against the canvas, which the perimeter discharges.

* fix: address sixth-round review findings

chatgpt-codex-connector:
- Ship the selected-state contrast with the package. The lightened
  `surface-active` fills only work because a perimeter marks the selection, and
  that perimeter lived in the SPA stylesheet, which `tsdown` never publishes —
  so `TimePicker`, `SelectDropDown`, `MultiSelect`, `InputWithDropDown`, `Badge`
  and `DataTable` gave an external consumer a 1.28:1 fill and nothing else. The
  perimeters and the placeholder ink now live in
  `packages/client/src/theme/highContrast.css`, imported by `ThemeProvider`;
  `style.css` keeps only what is genuinely SPA-only. The Ariakit selector also
  widened from four menu roles to `[data-active-item]`, and picked up
  `[data-state='selected']` for package data-table rows.
- Give Monaco selections a contrast-safe highlight: a focused selection now
  inverts the canvas (`text-primary` on `surface-primary`), matching
  `::selection`, because Monaco draws no perimeter. Weaker emphases share
  `surface-hover-alt` behind the ink find-match border.
- Populate Mermaid's separate `git0`-`git7` branch palette from the series ramp.
  Eight slots against a seven-slot ramp, so the last takes the ink rather than
  repeating branch one; labels take the canvas, since the ramp sits on its far
  side.
- Debounce the two appearance controls separately. They are independent
  settings, so light -> high-contrast dark is one flip of each, and the shared
  500ms window swallowed the second click.
- Route sharing principal badges (USER/GROUP/ROLE) through series-1/7/6.
- Apply the selected contrast mode to Office artifacts. The backend document
  picks its palette from `prefers-color-scheme` alone, so an explicit mode
  inverted whenever the OS disagreed. `withOfficeContrast` re-declares its
  `:root` tokens and pins `color-scheme` from inside the head.

* fix: address seventh-round review findings

chatgpt-codex-connector:
- Restore the citation highlight. The perimeter rule skips `span` so a wrapped
  passage is not boxed per fragment, which left the inline highlight on the
  lightened fill alone. It inverts now instead — the same treatment the mode
  gives `::selection`, 21:1 and with no geometry to wrap.
- Underline non-prose links. The 1.4.1 rule only reached `.markdown` and
  `.prose`, so the auth resend actions and in-dialog help links were colour-only
  against ink that is under 3:1 from them. It now also covers the shared
  `text-link` role, including the ones rendered as a `button`, and moved to the
  package stylesheet since that role is shared. MCP server descriptions style
  their anchors through an `[&_a]` variant rather than a class of their own, so
  they get the underline at the call site — running text, so it applies in every
  theme, not just the contrast ones.
- Route the selected tool card through `status-success` instead of
  `emerald-500/60` over a 6% fill, which composited to about 1.7:1 and left a
  selected card with a weaker edge than an unselected one.

* fix: keep helper hints opaque in the contrast modes

chatgpt-codex-connector: the `opacity-40` default-value hints composite pure
ink to about #999 on the white canvas (2.85:1) and #666 on the black one
(3.66:1), so small active labels missed even AA under a palette that contracts
for AAA. They take `high-contrast:opacity-100` now — the variant this branch
already added and `MessageNav` already uses — rather than a new token: every
text role is the same pure ink in these palettes, so there is no muted colour
to fall back to and the hierarchy is carried by size and the parentheses.

The mobile artifact drag handle gets the same treatment: it is a control, so
1.4.11 applies, and a 40% ink bar does not clear 3:1. The two remaining
`opacity-40` sites are left alone deliberately — `AssistantTool` dims a
disabled row, which WCAG exempts, and the `MarketplaceCatalog` empty-state
glyph is `aria-hidden` decoration.

* style: sort imports in the two parameter controls this branch touched

* fix: soften the temporary-chat accent outside the contrast modes

Moving the accent to `series-6` swapped a 60%-alpha `violet-800` edge for a
full-strength one, and series-6 is a saturated #7e23cd on the light palette and
#ab68fe on the dark one — the composer read as a warning rather than a quiet
mode hint. Half alpha in the standard palettes, opaque only under
`high-contrast:`, which is the one place the edge has to clear the 3:1
non-text floor.
2026-09-05 22:20:33 +02:00
Ravi Kumar L
000349681e
🩻 feat(insights): Scope Insights to Authorized Agents (#15549)
Some checks failed
Docker Dev Images Build / publish (push) Has been cancelled
Sync Helm Chart Tags / Ignore non-main push (push) Has been cancelled
Sync Helm Chart Tags / Sync chart tags (push) Has been cancelled
* feat(insights): add agent-scoped access

* fix(insights): exclude unattributed assistant messages

* fix(sharing): key role insight toggles by principal

* fix(sharing): preserve insight grant snapshot

* fix(permissions): preserve insight bits atomically

* fix(permissions): guard insight audit rollback

* fix(insights): keep admin grant controls accessible

* fix(permissions): reconcile duplicate principal updates

* fix(insights): show automatic admin access

* fix(permissions): harden insights access updates

* fix(insights): simplify authorization and preserve client state
2026-09-04 20:16:58 -04:00
Danny Avila
c302ae6f35
🪁 fix: Honor Endpoint Drop Params in Parameter Panels (#15610)
* Remove parameters from UI when parameter is dropped through dropParams

* Potential fix for pull request finding

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

* Addressed copilot and codex review suggestions

* 1. Normalize drop parameter names in client/src/components/SidePanel/Parameters/Panel.tsx
  Admin-configured dropParams for Azure/OpenAI-compatible custom endpoints use the effective backend field names (maxTokens, topP, frequencyPenalty, presencePenalty), but the panel's filter compared them against the UI's
  snake_case keys (max_tokens, top_p, etc.), so the controls stayed visible and silently discarded whatever the user set. Added a dropParamsBackendToUIKey map (in packages/data-provider/src/parameterSettings.ts) and normalize
  each dropParams entry through it before filtering.

  2. Apply the same drop-parameter filtering in client/src/components/SidePanel/Agents/ModelPanel.tsx
  The Agent builder's model panel built its parameter list independently and never consulted endpointsDropParamsMap at all, so it kept offering controls for Azure/custom providers whose values the backend drops. Reused the
  same resolution logic (including the new backend→UI key normalization) there, sourced from useGetStartupConfig.

  3. Remove any escapes from Azure fixtures in packages/api/src/app/config.test.ts
  Three tests built azure groupMap/modelGroupMap fixtures via ... as any as AppConfig['endpoints'], bypassing type checking for that shape. Replaced them with two typed helpers, createAzureGroupMap and createAzureConfig, built
  from the existing TAzureConfig/TAzureGroupMap/TAzureModelGroupMap types, so the fixtures are now fully type-checked with no any.

* HAL-1081 Deploy librechat open

  1. Gate the backend-name alias to OpenAI-compatible parameter sets (packages/data-provider/src/parameterSettings.ts)
  dropParamsBackendToUIKey had been applied unconditionally, rewriting a dropped topP to top_p even for a custom endpoint whose defaultParamsEndpoint is anthropic/google (or
  a native bedrock-* endpoint) — where topP is the UI key, so the rewrite broke hiding the control. Replaced the plain map with resolveDropParamsUIKeys(dropParams,
  endpointKey), which only aliases backend names for OpenAI-compatible endpoint keys (openAI, azureOpenAI, custom, openRouter) and passes native-provider keys through
  unchanged otherwise. Updated both Panel.tsx and ModelPanel.tsx to call it with overriddenEndpointKey, and added a resolveDropParamsUIKeys test suite in
  parameterSettings.spec.ts.

  2. Prune stale model_parameters when a control becomes hidden (client/src/components/SidePanel/Agents/ModelPanel.tsx)
  When a parameter an agent already had set gets added to its endpoint's dropParams, the control disappeared but the value stayed in model_parameters and was saved unchanged
  by composeAgentUpdatePayload — invisible to the user and silently reactivated if the endpoint later stopped dropping it. Added a useEffect, mirroring the conversation
  panel's existing pruning effect, that strips any model_parameters key no longer present in the currently-visible parameters list. Also fixed a pre-existing test regression
  (missing useGetStartupConfig mock) and added two tests covering pruning-on-drop and value retention when the control stays visible.

* fix: Preserve parameters for unknown agent providers

* fix: Sanitize agent parameters on submission

* fix: Preserve agent model overrides during pruning

---------

Co-authored-by: Marc Amick <MarcAmick@jhu.edu>
Co-authored-by: MarcAmick <5194465+MarcAmick@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
2026-09-04 13:52:23 -04:00
Marco Beretta
10c5c91594
🎨 feat: Theme-Adaptive SVG Support for Custom MCP and Group Icons (#13817)
* feat: theme-adaptive SVG support for custom MCP and group icons

Custom icons (MCP server iconPath, model spec groupIcon) were rendered as
plain <img>, so monochrome SVGs kept fixed dark colors and were nearly
invisible in dark theme.

Introduce a shared CustomIcon component that detects monochrome SVG glyphs
and tints them with currentColor so they follow the active theme, while
multi-color SVG logos and raster images keep their original colors. The
monochrome decision parses the SVG's color tokens; content is fetched once,
cached, and any failure falls back to the original image. Monochrome SVGs
render via CSS mask, never inlined, so no SVG markup reaches the DOM.

Apply across all custom-icon surfaces: MCP settings cards, the chat MCP
dropdown, stacked MCP icons, tool-call headers, and model group icons.

Also support SVG in the MCP avatar uploader: add SVG to the accepted file
types and sanitize uploaded SVGs with DOMPurify before storing them, and
make the dialog preview theme-adaptive via the same component.

Add unit tests for SVG detection, monochrome analysis, sanitization, and
CustomIcon rendering.

* fix: reset adaptive icon tint state when the source changes

A reused CustomIcon instance kept the monochrome verdict from a previous
source until its effect re-ran, so switching to a raster image or a new
multi-color SVG could briefly render it as a currentColor silhouette.

Key the resolved verdict to the current source and reset it synchronously
during render (seeding from cache when available), so a stale verdict can
never tint a different icon. Add hook tests covering the source change.

* fix: do not tint custom SVGs that have an opaque background

A grayscale SVG with an opaque full-canvas background (for example an
exported logo with a white background rect and a black glyph) passed the
monochrome check and was drawn through a CSS mask. Masks key off the alpha
channel, so the opaque background filled the whole area with the tint color
and the icon collapsed into a solid block.

Detect a full-canvas opaque background rect and exclude such SVGs from
tinting, rendering them with their own colors instead. Transparent
single-color and multi-shade glyphs remain tintable. Add tests for the
background cases.

* fix: detect opaque SVG backgrounds without a viewBox

Opaque background detection only read canvas dimensions from the viewBox, so
an SVG that declares width and height on the root element but omits the
viewBox slipped through and was tinted into a solid block.

Fall back to the root svg width and height when no viewBox is present, and
match attribute names exactly so stroke-width is not mistaken for the canvas
width. Add tests for the no-viewBox cases.

* fix: parse comma-separated SVG viewBox values

The viewBox regex only accepted whitespace between values, so a valid
viewBox like "0,0,24,24" failed to parse and an opaque background went
undetected, tinting the icon into a solid block. Accept commas and
whitespace as separators. Add a test for the comma-separated case.

* refactor: detect SVG tintability with DOMParser instead of regexes

The monochrome/tintable decision scraped SVG markup with regexes, which kept
missing edge cases (opaque backgrounds, missing or comma-separated viewBox,
stroke-width vs canvas width, embedded raster images).

Parse the SVG once with DOMParser and inspect real elements and attributes:
reject embedded <image>/<foreignObject> content, detect a full-canvas opaque
background rect, read the canvas size from the viewBox or root width/height,
and gather paint colors from attributes, inline styles, and <style> blocks.
Unparseable input is treated as not tintable. Tests cover these cases.

* fix: only tint single-tone grayscale SVGs

A grayscale SVG can draw its background as a full-canvas path (not just a
rect), e.g. a white background path plus a black glyph. The rect-only
background check missed that, and the icon flattened to a solid currentColor
block under the CSS mask.

Tint only when the SVG resolves to a single grayscale tone. Any second tone
(a background shape drawn as a path or rect, an accent, or a second shade)
now preserves the icon's own colors, which covers full-canvas path
backgrounds without per-shape geometry parsing.

* fix: count default black fill when detecting tintable SVGs

* fix: treat currentColor as a tone and stroked closed shapes as filled

* fix: ignore non-rendering template paint and count filled open paths

* fix: scan defs and symbol paint rendered through use references

* fix: resolve currentColor against fixed color and match style fills for default paint

* fix: treat alpha-zero fills as transparent in SVG tint detection

* fix: resolve inherited and CSS fill for backgrounds and default paint

* fix: resolve CSS currentColor and default-black use instances

* fix: count default-filled polylines when detecting tintable SVGs

* fix: resolve inherited fill and style/CSS opacity in SVG tint detection

* fix: ignore display:none shapes and stop counting CSS color as paint

* fix: only count CSS paints that match rendered elements

* fix: match root for CSS currentColor and skip unreferenced defs tones

* fix: count paint only on rendered shapes and filter hidden CSS currentColor

* fix: resolve use-instance currentColor and skip hidden use default fills

* fix: resolve CSS paints onto rendered painters and skip overridden use fills

* fix: keep checking ancestor opacity after resolving paint opacity

* fix: resolve CSS paints by specificity and skip transparent default fills

* fix: apply CSS over presentation attributes and follow nested use targets

* fix: strip CSS comments, honor visibility, and inherit paint through nested use

* fix: skip transparent svg use references

* Account for SVG filter colors

* fix: propagate outer use color through nested currentColor references

* fix: preserve url() id casing and record every nested use chain

* fix: preserve SVGs whose filters recolor the source

* refactor: detect monochrome icons via canvas pixel sampling

Replace the SVG source parser with offscreen-canvas pixel sampling: draw the icon, read pixel data, and treat it as monochrome only when every non-transparent pixel is grayscale within tolerance and the image has some transparency (a fully opaque image would mask to a solid block). Verdicts are cached per src.

An explicit monochrome flag on CustomIcon skips detection; sampling is the fallback. Image load failures and canvas taint from non-CORS cross-origin icons fall back to untinted rendering instead of throwing.

* harden custom SVG icon handling across security, a11y, and perf

Sanitize user-provided MCP iconPath server-side; the client-side
DOMPurify pass was bypassable by posting iconPath straight to the API.
Adds sanitizeMcpIconPath in @librechat/api (allowlist SVG sanitizer that
preserves case-sensitive names) and runs it in the create/update
controllers, plus a length cap on iconPath in the shared schema.

- gate theme detection to same-origin/data sources so a remote icon is
  no longer auto-fetched from every viewer's browser
- restore forced-colors (High Contrast) visibility and the broken-icon
  onError fallback for tinted icons
- mark redundant icons decorative to avoid screen-reader double reads
- base64-encode inlined SVG data URIs and bound the verdict cache
- harden the client SVG sanitizer and guard FileReader failures
- cover the tinted render branch, sanitizer allowlist, and edge cases

* fix: preserve local SVG references while stripping external hrefs

The hardened sanitizers dropped <use> and every href outright, blanking
self-contained exporter output like <defs><path id="p"/></defs>
<use href="#p"/> and gradients inheriting stops via href="#g".

Re-allow <use> and href/xlink:href on both sanitizers, restricted to
same-document fragments: the client uses a dedicated DOMPurify instance
with an afterSanitizeAttributes hook (kept off the shared default
instance), and the server mirrors the rule with a sanitize-html tag
transform plus an empty allowedSchemes list as a second layer. External,
protocol-relative, relative-path, and javascript: hrefs are still
stripped, leaving the element inert.

* fix: keep xlink namespace and safe filters in the server icon sanitizer

The server allowlist stripped xmlns:xlink from the root while keeping
xlink:href, leaving an unbound prefix that is a parse error once the
icon is stored as an image/svg+xml data URI, so SVG 1.1 exporter output
went blank after save. It also dropped <filter>/fe* primitives that the
client sanitizer's svgFilters profile allows, so icons relying on
effects like feDropShadow lost them between preview and reload.

Allow xmlns:xlink and mirror the client's filter allowlist (elements
plus their presentation attributes); feImage passes through the existing
fragment-only href transform, so external references are still stripped.

* fix: require empty pixels for tinting and block external url() references

Two issues from review of the icon sanitizer and monochrome detector:

- scanMonochrome flagged any pixel below full opacity as transparency,
  so an SVG that fills the whole canvas with a semi-transparent grayscale
  wash was tinted and rendered as a solid currentColor block. Require a
  genuinely empty pixel (alpha at or below the paint threshold) instead.

- The fragment-only reference rule only covered href/xlink:href, so a
  url() in filter/fill/mask/clip-path/style kept external targets like
  filter="url(https://evil/f.svg#f)". Both sanitizers now strip any
  attribute carrying a non-fragment url(), preserving local url(#id).

* fix: preserve internal SVG stylesheets while scrubbing their references

Exporter SVGs that store multi-color paint in an internal <style> block
(e.g. .red{fill:#e00} with class="red") lost all styling because the
sanitizers dropped <style> outright, so multi-color logos rendered with
default paints after upload.

Re-allow <style> on both sanitizers and scrub its CSS: strip comments
and @import at-rules and rewrite any non-fragment url() to none, keeping
local url(#id) paints. These icons only render in inert <img>/CSS-mask
contexts isolated from the host page (no script, and no subresource
loads once external refs are scrubbed), so a script-free stylesheet is
safe to keep; a premature </style> close still has its trailing markup
re-sanitized by the parser.

* fix: normalize icon value before deciding it is not an SVG data URI

A posted iconPath with leading whitespace or C0 controls (e.g.
"\n data:image/svg+xml,...") slipped past the anchored data-URI check,
so sanitizeMcpIconPath returned it raw while a browser trims the prefix
and still renders the SVG, keeping its external <image>/url() and other
stripped content. Normalize the value the way URL parsing does, trimming
leading/trailing C0 controls and spaces and removing embedded
tab/newline, before the SVG check so these variants are sanitized too.

* fix: cap sanitized icon length and re-encode as compact base64

The schema length check ran on the input, but sanitizeMcpIconPath then
re-encoded the markup with encodeURIComponent, which expands SVG by well
over the base64 input size (a 235KB base64 icon became a 374KB percent
string). create stored that over-limit value, and editing the server
resubmitted the prefilled value and failed MCPServerUserInputSchema.

Re-encode the sanitized SVG as base64 (far more compact for angle-heavy
markup, which sanitizing can itself grow via explicit close tags) and
drop anything still over MAX_MCP_ICON_PATH_LENGTH, so a stored icon can
never exceed the cap that a later edit re-validates against.

* fix: resolve CSS escapes before stripping external SVG references

A literal url(/@import matcher missed CSS-escaped references: a value
like style="fill:u\72l(https://attacker/x)" or an escaped @import
survived sanitization because the browser un-escapes \72 to r at
tokenization time while the regex never saw a url(...). The external
reference was then persisted and served to other clients.

Un-escape CSS the way a browser does (single pass over \hex and
\char escapes) before the url()/@import matchers run, on both the client
and server sanitizers. Also scrub the multi-declaration style attribute
in place instead of dropping it, so co-located local paint rules survive
while escaped external references are neutralized.

* fix: re-sanitize after scrubbing style blocks to block markup reintroduction

scrubStyleBlocks runs after sanitizeHtml and splices un-escaped CSS back
as raw markup, so an escaped sequence like \3c/style\3e\3cimage/\3e
(inert text during the first pass) became a real </style><image href>
element after unescaping, past the allowlist. Re-run the allowlist over
the scrubbed result whenever a <style> block was rewritten, stripping any
element the un-escaping surfaced while leaving legit local rules intact.

The client sanitizer is unaffected: it sets the scrubbed CSS as a DOM
text node, which serializes with < and > escaped, so no element is
reintroduced. Added a regression test pinning that behavior too.

* fix: make forced-colors icon tint override the inline background color

The forced-colors rule for .custom-icon-tint had no !important, so it
lost the cascade to the element's inline `background-color: currentColor`
and the masked glyph still blanked out in Windows High Contrast — the
exact regression the rule was meant to prevent. Verified in a real
forced-colors render: without !important the tinted icon is invisible;
with it the glyph paints in CanvasText and stays visible.

* fix: enforce icon length cap in the sanitizer, not the schema

The schema `.max()` on iconPath rejected the whole update at parse time.
A server whose stored icon predates the cap (previous versions allowed
large data-URI images) re-submits that value from the edit dialog, so
the user was locked out of changing any field or clearing the bad icon.

Drop the schema `.max()` and enforce MAX_MCP_ICON_PATH_LENGTH in
sanitizeMcpIconPath for every value type: an over-cap SVG is compacted
or dropped, an over-cap raster/URL is dropped. Editing a server with a
pre-existing oversized icon now succeeds and clears the icon instead of
failing validation.

* fix: decode XML entities and parse quoted URLs in SVG CSS scrubbing

Two more bypasses of the CSS reference scrubber, both live once the
stored image/svg+xml is parsed by a viewer:

- A <style> block reaches the scrubber as raw text, so an entity-encoded
  reference like &#64;import (or &#x40;import) was never seen as @import;
  the browser decodes it at render time. Decode XML character references
  (numeric + the five predefined named) before the matchers run.
- The url() regex excluded ')' even inside quotes, so a valid quoted URL
  like url("https://x/a)b") slipped through. Parse quoted and unquoted
  URL tokens separately so a ')' inside quotes stays part of the target.

Applied to both the server (sanitizeMcpIconPath) and client (sanitizeSvg)
scrubbers. Entity decoding also catches an entity-encoded </style> markup
breakout, which the re-sanitize pass then strips.

* fix: close custom icon review threads

* fix: transpile sanitize-html's ESM dependency chain in api jest runs

* fix: preserve marker definitions and fragment-only marker references

* fix: keep object-cover sizing when tinting masked icons

* fix: preserve textPath labels and currentColor color scopes in icon sanitizer

* fix: preserve pattern coordinate-system attributes in icon sanitizer

Keep patternUnits, patternContentUnits, and patternTransform so patterned
SVGs do not fall back to objectBoundingBox after server-side sanitize.

* refactor: sanitize icons with one shared DOMPurify policy

The server re-implemented DOMPurify's SVG profiles as two hand-written
sanitize-html allowlists, so every element and attribute the client kept had to
be transcribed by hand, and each miss surfaced as an icon that previewed
correctly and then lost its markers, patterns, filters, or xlink binding once
stored. Run DOMPurify on jsdom instead and share the config and reference hook
with the client, which drops both allowlists.

Restrict url() values to same-document fragments on every attribute rather than
on the marker properties alone, so external references in fill, stroke, filter,
mask, and clip-path are stripped as well.

Restore feDropShadow casing after sanitizing. DOMPurify has to parse as HTML,
its XML mode drops filter primitives and gradients, and the HTML tag-name
adjustment table predates that primitive, so it survived lowercased and left its
filter empty once the stored icon was re-parsed as image/svg+xml.

* fix: defer jsdom load until MCP icon sanitization

jsdom and dompurify were imported at module scope, so every
require('@librechat/api') paid their startup cost. Load them on first
SVG sanitize instead.

* fix: sort type import in MCP icon sanitizer

The static-check import sorter wants value imports before type-only
imports. Moving the DOMPurify type import after librechat-data-provider
clears the drift.

* fix: address PR review bot findings

chatgpt-codex-connector:
- Resolve icon URLs through new URL() instead of trusting a leading slash, so
  /\attacker.example/icon.svg is no longer treated as same-origin
- Check every url() token in an SVG attribute rather than only the first, and
  reject values carrying a CSS escape that would decode into one
- Forbid animateTransform, animateMotion, animateColor, and mpath, which the
  SVG profile admitted despite the policy intending to remove animation
- Preserve the radial-gradient fr attribute the profile omits
- Reject multi-tone grayscale in scanMonochrome, which a mask would flatten
- Sample at 128px so a small chromatic accent is not averaged away
- Settle a stalled icon load after 10s so its in-flight entry is released
- Bound the decoded SVG before building a jsdom tree for it
- Reject an oversized icon upload with a visible error instead of accepting it
  and silently clearing it on save

* fix: address Codex findings and trim icon sanitizer comments

- Bound icon files by size before reading or sanitizing them
- Track tone spread across every painted pixel, not only solid ones
- Settle monochrome detection on any sampling throw
- Use named DOMPurify/JSDOM types and shorten PR comments

* fix: percent-decode SVG data URI bodies before base64 decoding

* fix: declare the SVG namespace on sanitized icons and keep icon rejections out of form validity
2026-09-04 10:49:43 -04:00
Marco Beretta
bd31ecd129
✨ feat: add morphing state icons across the client (#14697)
* feat(client-ui): add morphing icon support

* feat(client-ui): morph shared state icons

* feat(chat-ui): morph chat action icons

* feat(client-ui): morph remaining action icons

* fix(client-ui): type morph icon test factory

* test(client): type agent detail fixture

* test(client-ui): guard morph icon accessibility defaults

* test(chat-ui): assert mermaid view toggle via morph icon

* style(chat-ui): sort mermaid header imports

* test(api): wait for rebuilt MCP session to finish connecting

The SSE conflict recovery test treated a new session id as a finished
reconnect. initialize assigns that id before connectClient emits
connected, so isConnected() can still be false. Wait for a live
connection on the rebuilt session instead.

* fix(client): clean up rebase fallout
2026-09-04 03:27:14 +02:00
Danny Avila
aa7539909b
🪶 feat: Enable Standalone Skill Authoring (#15480)
* feat: enable standalone skill authoring

* test: align skill e2e flow with enable toggle

* fix: preserve legacy skill state on version restore
2026-09-01 18:39:05 -04:00
Danny Avila
ccd673bf05
🧰 fix: Keep File-Backed Agent Capabilities and Their Saved Tools in Agreement (#15418)
An agent with knowledge files showed File Search as selected in the builder while
the save omitted `file_search` from its tools. At runtime the tool was gone, the row
still looked selected so re-adding it changed nothing, and the next save re-confirmed
the broken state. Recovery needed a MongoDB write.

The builder's model for file-backed capabilities is stated in ToolsSection: they stay
selected while they hold files, and removal is routed to the file manager rather than
the capability flag — mirroring the file-only `context` built-in. Three places
implemented that model separately and drifted:

- display resolved it from the flag OR the file count,
- persistence resolved it from the flag alone,
- the removal guard lived inline in the tools list and was missing entirely from the
  marketplace dialog.

That last gap is how the bad state was reached. `ToolsMarketplaceDialog` toggled a
selected File Search straight to `file_search: false` with no file check, producing
flag-off-with-files — a state display cannot show and persistence silently drops.

`items/capabilities.ts` now owns the capability table and the one rule, and display,
removal and persistence all derive from it:

- persistence resolves the same way display does, so a file-holding capability keeps
  its tool (this also heals already-broken agents on their next save),
- both toggle paths route removal of a file-holding built-in to the file dialog,
- the removal guard takes file counts rather than the agent, so a component can pass
  the hydrated entries it already watches while submit reads the form — the rule is
  shared, the source of the count need not be.

Submission also stops mutating `data.tools` in place and dedupes the result.

Supersedes the persistence half of #15380, which fixed the save but left the
marketplace path able to recreate the state and kept the rule in a third copy.

Fixes #15368
2026-08-31 09:47:32 -04:00
Danny Avila
c1cb591d49
🗿 feat: Add Attached Stateful Code Environments (#15352)
* feat: add attached stateful code environments

* fix: include code environment in lazy agent type

* fix: harden stateful environment routing

* fix: complete code environment route isolation

* fix: declare code environment map type

* fix: preserve configured code execution routes

* fix: harden stateful environment updates

* test: assert route-scoped sandbox readiness

* fix: harden stateful environment lifecycle

* fix: isolate migrated code sessions

* test: assert route-qualified code sessions
2026-08-30 15:38:45 -04:00
Danny Avila
03fa98eb37
🕸️ feat: Complete Keenable Web Search Provider (#15288)
* feat: wire Keenable web-search provider into config, schema, and UI

Keenable landed as a search provider in @librechat/agents (#285, shipped in
3.2.58+), but LibreChat did not yet expose it. This adds the config/schema/UI
glue so it can be selected, mirroring the existing Tavily provider.

- data-provider: add `keenable` to SearchProvider type + SearchProviders enum,
  keenableApiKey/keenableApiUrl schema fields, and a keenableSearchOptions block
  (maxResults, site, attributionTitle, timeout).
- data-schemas: register keenable in webSearchAuth.providers and default the
  key/URL placeholders in loadWebSearchConfig.
- api/web: pass keenableSearchOptions through to the provider and handle
  Keenable's keyless model. Unlike other providers it authenticates with no key
  (the public endpoint), picking up an optional key/URL when set; the URL
  override is SSRF-preflighted like other user-provided URLs.
- client: add Keenable to the provider dropdown with an optional API-key input.
- docs: document KEENABLE_API_KEY/KEENABLE_API_URL in .env.example and a
  webSearch example in librechat.example.yaml.
- tests: keyless + keyed auth resolution, config defaults, and schema parsing.

* fix: ESLint no-unused-vars and clarify Keenable yaml example

- Remove the now-unused RerankerTypes import in data-schemas web.ts (the lint
  job runs with --max-warnings 0 on changed files, so this latent warning failed
  CI once the file was touched).
- Note in the librechat.example.yaml Keenable stanza that a scraper (and
  reranker) is still required for web search to load, and include a Firecrawl
  scraper in the example.

* chore: fix import order drift (sort-imports)

* feat: add Keenable as a keyless scraper and select it without a pinned provider

The Keenable scraper landed in @librechat/agents#337, so wire the scraper
category the same way the search provider already is: `scraperProvider:
keenable` reads pages through Keenable's public fetch endpoint with no key
(a key only lifts rate limits, and the endpoint is overridden with
KEENABLE_FETCH_URL). Paired with `rerankerType: none` this makes a fully
keyless web-search stack possible for the first time.

Also closes the Codex finding on this PR: because none of Keenable's auth
fields are required, the generic auth loop skips it whenever it isn't pinned,
so a key submitted through the API-key dialog (which cannot pin a provider)
left the providers category unauthenticated. Keenable is now selected in that
case, gated on one of its values actually being present so installs that
configured nothing keep their current behavior. The scraper gets the same
fallback, additionally gated on Keenable being the resolved search provider,
so it never silently scrapes for another provider.

* fix: select the Keenable scraper from a supplied key, not only for Keenable search

The API-key dialog submits credentials and cannot pin a provider, so choosing
Keenable as the scraper while search stays on Serper/SearXNG/Tavily had no
effect: the unpinned-scraper fallback required Keenable to also be the resolved
search provider.

A supplied Keenable value now triggers it as well, which is the only signal the
dialog can send. The fallback still runs only when no keyed scraper
authenticated, and with neither trigger the category stays unauthenticated, so
a deployment that never configured Keenable is unaffected.

Note the fully keyless choice still cannot be expressed through the dialog:
Keenable's key is optional, so picking it with no key submits nothing at all.
librechat.example.yaml now documents pinning scraperProvider: keenable for that
case.

* style: Sort Keenable imports

* fix: Harden Keenable auth resolution

* fix: Preserve Keenable selection intent

* fix: Fail closed on invalid web search auth

* fix: close keenable auth gaps

* style: sort web auth imports

* fix: preserve web search selection integrity

* fix: isolate web search auth ordering

* fix: silence expected credential misses

* chore: bump agents sdk

* fix: preserve web search preference ownership

* fix: forward cleared Keenable endpoint

* style: sort web search hook imports

* fix: require intent for credential clears

---------

Co-authored-by: Ilya Bogin <ilya.bogin@keenable.ai>
2026-08-27 14:29:20 -04:00
Danny Avila
8b1fcc0fc2
🌐 fix: Expose Gemini Models to Vertex AI Agents (#15234)
* 🌐 fix: Expose Gemini Models to Vertex AI Agents

* ♻️ refactor: Resolve Shared Vertex Model Catalogs

* fix: Preserve Exact Vertex Model Catalogs

* style: Format Agent Model Selection

* test: Preserve Native FS in Stable Diffusion Spec
2026-08-27 06:50:23 -04:00
Marco Beretta
227a99ede8
🧮 fix: Render Google Settings From the Shared Schema and Bound Them Per Model (#14989)
* fix: render Google settings from the shared schema and bound them correctly

Google was the last endpoint hand-rolling its own sliders. The schema was
already there and already wired, only the frontend never used it, so
rendering from it replaces 315 lines with the body OpenAI, Anthropic and
Bedrock share.

That closed a functional gap rather than only moving code: the old form
exposed six fields where the schema declares fifteen, so Resend Files,
Thinking, Thinking Budget, Thinking Level, Grounding with Google Search,
URL Context and File Token Limit were unreachable from a Google preset.
resendFiles is added to the Google schema so its switch round-trips, and
the endpoint type is resolved from the endpoints config, since a preset
for a Google-compatible endpoint need not carry endpointType and would
otherwise blank the panel.

Sharing the controls also meant inheriting their gaps, which this fixes:

- Number settings declared a range that nothing enforced, so a value past
  the provider's ceiling was persisted and rejected later. clampSettingRange
  applies it, and generateDynamicSchema validates the same rule, so the
  definition is the single source of truth for both.
- Thinking budget bounds are per model. The generic range capped 2.5 Pro
  below its documented 32,768 and admitted Flash values above 24,576.
  positiveMin carries the documented floors while -1 stays typeable as the
  decide-automatically sentinel.
- Ranges the model narrowed are marked modelSpecific, so a switch to a
  model that ignores the parameter cannot rewrite a value set for another.
- useDebouncedInput rebuilt its debouncer every render, because neither
  setOption nor the inline setter is memoized, so pending edits were never
  really superseded and a flush reached an instance holding nothing. The
  callbacks move to refs, and the text and slider controls flush on blur or
  value commit so Save and Export cannot read a stale preset.
- Controls reset to their definition default on a conversation or preset
  change and only recovered ~560ms later, which showed saved values as
  defaults and could write the default back.

The debounce regression test fails against the previous memo dependencies
and passes with the refs, so the flush is verified rather than assumed.

* fix: keep the context token bounds on the Google setting

The bounds came from the hand-rolled Google editor, but they were added to
the shared definition every endpoint renders, so blurring the field clamped
OpenAI, Anthropic, Bedrock and custom endpoints to a window that is only
Gemini's. Custom endpoints may declare context windows outside it.

* fix: agree with the generated schema across the sentinel gap

A stored value between range.min and zero passed through clampSettingRange
unchanged, though the schema admits only the sentinel or the positive floor,
so normalization could preserve a value the provider then rejects. Validate
a configured default against the same rule.

* fix: keep positiveMin on configured parameter definitions

The runtime schema for customParams.paramDefinitions retained only min, max
and step, so a configured positive floor was stripped before the UI saw it
while the shared SettingRange type advertised it.

* fix: commit a double-click slider reset immediately

The browser dispatches dblclick after the second pointer release, so the
value commit has already flushed and the reset sat in the debouncer. Saving
or exporting inside that window read the value the slider no longer showed.

* fix: normalize an out-of-range stored value on mount

The applied-range ref started at the first range, so the effect returned
immediately and a budget saved under the shared range stayed displayed and
savable when the selected model no longer allowed it.

* fix: normalize on navigation and keep sliders out of the sentinel gap

The parameters panel stays mounted across conversations, so a legacy budget
could arrive under a range that never changed; keying the normalization on
the conversation or preset identity as well catches it. After a navigation
the local value still belongs to the conversation being left, so the
incoming stored value is what gets normalized.

A slider steps straight through the gap between a sentinel minimum and its
positive floor, which the generated schema rejects, so the committed value
is clamped. It is also set before the flush: the keyboard path commits
before it reports the change, so the flush alone had nothing to write.

* fix: close the remaining paths into the sentinel gap

Applying a preset over the open conversation replaces the stored value
without changing the conversation id or the model, so normalization now also
triggers on a stored value that arrives differing from the local one. A
value the user typed reaches the conversation through this same field and
matches by the time it lands, so it stays on the blur-clamped path.

The slider's adjacent number input only flushed on blur, so a typed value
could sit in the gap the track is now kept out of.

A configured positiveMin above the maximum admits nothing but the sentinel
while the clamp maps every non-negative input onto a maximum the generated
schema rejects, so both the config schema and the definition validator
refuse it.

* fix: keep a non-negative sentinel and a loadable slider default

The minimum is the sentinel whatever its sign, and the generated schema
admits it outright, so a range like { min: 0, positiveMin: 10 } no longer
has its 0 lifted to the floor by the clamp.

The synthesized slider default took the midpoint of the whole range, which
for a sentinel range lands in the gap the validation added alongside it, so
an otherwise coherent custom definition failed to load. It now takes the
midpoint of the admissible interval.

---------

Co-authored-by: Danny Avila <danny@librechat.ai>
2026-08-25 21:21:28 -04:00
Danny Avila
e9936b8ad2
🎢 fix: Restore Schedule Dialog Scrolling So Save Stays Reachable (#15225) 2026-08-25 19:58:45 -04:00
Marco Beretta
bf6144c9e1
🎛️ fix: Withhold the Seeded Model Catalogue Until Models Resolve (#15035)
`useGetModelsQuery` seeds from a static fallback config, so `modelsQuery.data`
describes a hardcoded model list both before the mounted fetch resolves and
after it fails outright. The agent builder read that seed as authoritative and
offered models the active server configuration never exposed.

Blank the catalogue until the mounted fetch actually succeeds, surface the
failure in the model panel instead of silently falling back to the seed, and
refuse to create an agent against a provider/model pair the resolved catalogue
does not offer.

Also wires the builder's orphaned `htmlFor` labels to the controls they name.
2026-08-25 08:58:02 -04:00
Ravi Kumar L
3d2da403ce
🥛 fix: Drop Stale Saved Model Defaults in Builder Forms (#15179) 2026-08-25 08:18:17 -04:00
Danny Avila
4b113697b5
🔌 feat: Background Execution Toggles for Actions & Plugin Tools (#14407)
* 🧵 feat: Background Execution Toggles for Actions & Plugin Tools

* 🩹 fix: Resolve action background opt-in across encoded-domain forms and scope it per action

* 🧹 refactor: Resolve action domain in a single pass

* 🧩 fix: Merge Normalized Action Background Options

* 🪢 fix: Reconcile Action Background Aliases

* 🧭 fix: Harden Action Background Compatibility

* 🕰️ test: Allow Settled Task TTL Expiry

* 🧬 fix: Merge Refreshed Action Tool Registrations
2026-08-25 08:13:13 -04:00
Marco Beretta
649e68170e
🖼️ refactor: Consolidate Provider Icons Into a Single Registry (#15148)
* test: make useIsActiveItem observer assertions deterministic

The two attribute-flip tests mutated inside act() and then raced a 4 second
waitFor against MutationObserver delivery, so they failed once the client
workspace gained enough suites for a worker to stall past that budget.

Wait on actual observer delivery instead. The hook registers its observer on
mount, so it is ahead of the test's in delivery order and has already reacted
by the time the promise resolves. The new helper filters on data-active-item
because React writes data-active onto the same element when it re-renders, and
an unfiltered observer would resolve on that write instead.

This removes the last wall-clock dependence in the file, so the 20 second
jest timeout is no longer needed.

* feat: add canonical ProviderId vocabulary and resolver

* feat: resolve custom endpoint provider identity at config load

* feat: add provider icon registry data

* feat: add ProviderIcon and ProviderAvatar components

* feat: add provider icon resolution hook

* refactor: migrate direct icon lookups to the provider registry

* refactor: migrate composite endpoint icons to the provider registry

* refactor: render message provider icons from the registry

* refactor: remove the duplicated endpoint icon maps

The model selector was the last consumer of the icons map, so it now
resolves art through the provider registry like every other icon call
site. That leaves getIconKey with no callers, and the five icon map
types it depended on with no references, so all of them go too.

* fix: address Codex review findings on provider icons

Move brand tile colors onto theme tokens, accept relative image paths,
pass endpoint config into message icon resolution, keep Cohere padding
on landing only, render configured image URLs in provider-only
consumers, preserve the Gemma label, and publish provider assets with
the shared client package.

* fix: address remaining Codex findings on provider icons

Keep monochrome art white on branded avatar tiles, inline provider
assets as module data URLs so ProviderIcon works outside the SPA, and
recognize api.cohere.ai when resolving custom endpoint brands.

* fix: address the latest Codex review notes

Stop inlining provider logos into the shared bundle, keep agents and
assistants marks on group icons, reject CSS appended to brand
gradients, give brand tokens hex fallbacks for package consumers, and
treat data image URLs as configured artwork.

* fix: honor native provider and theme-controlled avatar contrast

Use an explicit custom-endpoint provider when host branding misses,
keep agents and assistants marks on model specs, and drive branded
avatar foreground from a theme token instead of a raw white class.

* fix: tighten brand validation and inherit SVG fill color

Forward the computed color class into provider SVGs, accept only a
single balanced gradient for brand backgrounds, keep provider
foreground hex-only, recognize relative image fragments, and preserve
percentage sizing in URLIcon fallbacks.

* fix: keep EndpointIcon hook-free and accept protocol-relative icon URLs

useMentions.ts invokes EndpointIcon({...}) as a plain function in seven
places, inside useMemo mappings and a React Query select callback, so the
useProviderIcon call added to it ran a hook outside a render and threw
"Invalid hook call" as soon as the mention list was built. It now uses the
hook-free resolveProviderIcon, and a spec pins the imperative-call contract
those call sites depend on.

isImageURL explicitly rejected protocol-relative URLs, so an endpoint or
model group configured with //cdn.example.com/provider.png fell through to
provider resolution and rendered the generic mark, where the removed
UnknownIcon rendered any nonempty custom iconURL. A leading // followed by
a host is now an image; a bare // or /// still is not.

The ConvoIcon spec's two cohere conversations move to one shared fixture,
since ProviderId.cohere is not an EModelEndpoint and a single-step
assertion to TConversation failed the client type check.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MWvn6ezgLmN8D5GDmFVnwv

* fix: annotate themeBrandTokens for isolatedDeclarations

packages/client compiles with isolatedDeclarations, under which
`as const satisfies` is not an explicit type annotation, so the emitted
declaration could not be produced from the initializer alone.

This never surfaced before because the "Type check @librechat/client"
step only runs after "Type check @librechat/api", which was failing on
dev's Agents SDK issue and skipping it.

Annotated as readonly (keyof IThemeBrands)[] and frozen, matching
themeColorTokens directly above it. Both consumers only call .includes()
and .map(), so no literal tuple type is lost.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MWvn6ezgLmN8D5GDmFVnwv

* fix: keep nested provider SVGs at their span's size

ProviderIcon sizes component art with an outer span carrying an inline
width/height, then rendered the SVG with cn('h-full w-full', classes).
Because cn is twMerge, a caller's own sizing class won that merge, so the
fraction applied twice: Landing passes size={41} with h-2/3 w-2/3, ConvoIcon
scales to a 27px span, and the SVG then took two thirds of that again, ~18px
where it used to be ~27px.

Only component-backed providers regressed. The asset branch has no wrapping
span, so its fraction still resolves against the 40px container.

Reordering the merge makes the span's size authoritative while leaving every
other caller class in place, including the [color:inherit] that branded
avatars forward. The img branch keeps resolving against its parent, so its
size is unchanged.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MWvn6ezgLmN8D5GDmFVnwv

* fix: close the image-format and provider-host tables

Two allowlists that the refactor narrowed, fixed as sets rather than one
entry at a time.

isImageURL's extension list had grown by patch four times, each round
restoring one form the old renderer accepted. It now carries every format
browsers actually render, so avif joins apng, bmp, cur, jfif and the jpeg
spellings in a single pass.

The host table had no Azure entry, so an OpenAI-compatible endpoint on
team.openai.azure.com fell through to the generic mark; the custom schema
cannot express provider: azure, so host was its only signal. Both supported
Azure suffixes are added, and enumerating ProviderId against the table
surfaced Google as the same gap, which is added too.

Bedrock, mlx and ollama are the remainder and cannot be host-resolved:
bedrock's hostname is region-scoped under a shared AWS suffix, and the other
two are served from the operator's own machine. That is now recorded next to
the table and pinned by a test, so a provider added later without a host
fails rather than silently rendering the generic mark.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MWvn6ezgLmN8D5GDmFVnwv

---------

Co-authored-by: Claude <noreply@anthropic.com>
2026-08-24 13:00:38 -04:00
Danny Avila
8773b36eec
🎽 fix: Commit Subagent Roster Selections to Form State (#15154)
* fix: persist subagent selections synchronously

* test: verify subagent roster form state

* style: sort subagent roster test imports
2026-08-24 03:21:48 -04:00
Marco Beretta
dd85c6d6d0
🪹 feat: Shared Empty State for Side Panels (#15123)
* feat: shared empty state for side panels

Bookmarks and Memories each hand-rolled the same empty state: the same bordered
card, the same circular icon surface, the same title and caption sizes, written
out twice. Schedules had none at all, so an account with no schedules got a bare
list with nothing to explain what the panel is for.

One EmptyState primitive in packages/client, taking an icon, an optional title
and description, and an optional action. Bookmarks and Memories move onto it with
no visual change and no copy change. Schedules gets a real empty state, and an
error state with a Retry action, so a panel that failed to load offers a way out
instead of looking empty.

A description with no title takes the title's size rather than the caption's:
where it is the only line, it IS the message.

* fix: drop the create hint for roles without schedule create access

The panel already hides its create button behind hasCreateAccess, but
the empty state still told a USE-only viewer to create a schedule it
offers no way to create. The invitation now renders only when the
capability does.

* fix: suppress the create hint when the quota already blocks creation

A maxPerUser of 0 disables the create button on an empty list, so the
empty state must not say to create one either; the hint now follows the
same effective gate as the button.
2026-08-23 18:56:34 -04:00
Marco Beretta
8a118c7cb3
⏱️ feat: Shared Time Picker for Schedule Times (#15122)
A schedule's time was three dropdowns side by side: hour, minute, meridiem. That
is three controls for one value, it cannot be read at a glance, and the minute
list was a fixed set of four with the stored value bolted on, so a schedule
already running at :07 could be kept but never chosen.

They become one TimePicker: hour, minute and, where the clock format calls for
one, meridiem, as scrollable columns behind a single trigger showing the selected
time. An hourly cadence gets MinutePicker, the same control with its other
columns dropped, so it reads as the same widget rather than a different one. Both
live in packages/client with their wording passed in as props, so the primitive
carries no translation keys of its own.

Not `<input type="time">`: the browser owns its rendering, and it cannot be
brought in line with the rest of the form.

`hour12` is a required prop rather than a locale-derived guess. The app has
already resolved its Clock format setting, and re-deriving the answer inside the
picker would let it disagree with the summary printed beside it.

The trigger names its selected value as well as its field: `aria-labelledby`
replaces a button's child text, so pointing it at the label alone announced
"Time" and left a screen reader user unable to tell what was selected without
opening the columns and reading them. The columns are a roving-tabindex
radiogroup, arrow keys wrap, and the selected row is scrolled to the middle of
its column on open.

The popover is deliberately not portaled. A Radix dialog sets `pointer-events:
none` on the body while open, so a popover portaled out of it renders correctly
but receives no clicks or wheel events, and its focus trap puts the content out
of tab order too.

Hour and minute are set in one change. Behind separate fields a half-applied edit
could submit a time the user never picked, and the form now carries the hour as
the 0-23 value the cadence stores rather than a 12-hour value plus a meridiem it
has to recombine.
2026-08-23 22:47:33 +00:00
Marco Beretta
7834ebab33
🕰️ feat: Clock Format and Week Start Preferences (#15121)
* feat: clock format and week start preferences

Times were written in whatever convention the browser locale implied, and the
week always started on Sunday. Neither is right for a large part of the user
base: most of Europe reads a 24-hour clock and starts the week on Monday, and a
user running an English interface in a region that does either is currently
given the American convention with no way to change it.

Two General settings, Clock Format (System / 12-hour / 24-hour) and Week Starts
On (System / Sunday / Monday). Their System branch reads the runtime locale
rather than `i18n.language`, which is normalized down to a translation bundle:
`en-GB` and `en-AU` both become `en`, which is exactly the regional part these
two settings depend on, and reading it would report a 12-hour clock and a Sunday
week to a British user.

Week start is typed on the same 0-6 Sunday-first scale the schedule cadence uses
rather than being narrowed to Sunday/Monday, because the System branch reports
whatever the locale says and several (ar-EG, fa-IR) start the week on Saturday.
Engines without `Intl.Locale.prototype.getWeekInfo` fall back to a short list of
Sunday-first regions with Monday, the ISO 8601 default, otherwise: this is a
display default the toggle can always override, so an imperfect fallback degrades
rather than breaking.

Both settings are stored per browser. They describe how this device reads a
clock, which is a property of where someone is sitting rather than of their
account, and a user who moves between a European desktop and a US phone wants
each to read its own way.

Applied to message timestamps, the schedule dialog and card, key expiry and
refill dates, prompt and agent version dates, memory dates, and project chat
lists. The weekday order also drives the schedule dialog's day pills and the way
a weekly cadence reads back, so a wrap-around selection of Sat+Sun+Mon reads
"Monday, Saturday, Sunday" in a Monday-first week instead of "Sunday, Monday,
Saturday".

Dropdown now names its selected value as well as its field label. `aria-labelledby`
REPLACES the trigger's own text, so pointing it only at the caller's label left
the selected value unannounced, which these two settings are the first consumers
to hit.

* fix: teach the week-start fallback the Saturday-first regions

The no-week-data heuristic could only answer Sunday or Monday, folding
ar-EG to Sunday and fa-IR to Monday when CLDR says both start on
Saturday, and the selector offers no explicit Saturday override to
recover with. It now carries CLDR's Saturday-first territories, and the
UAE moves off the Sunday list to the Monday default, where CLDR put it
when its weekend moved to Sat-Sun. The fallback tests delete the
engine's week data for their duration, so they exercise the heuristic
on every engine instead of skipping wherever getWeekInfo exists.

* fix: infer likely regions for bare language tags and stop rebuilding clock formatters

A runtime that reports a language-only locale (bare ar or fa) carried no
region for the week-start heuristic, so those users fell to the Monday
default even though maximize() knows their likely region starts the week
on Saturday. The heuristic now maximizes before defaulting.

The runtime locale and each locale's meridiem answer are also cached at
module scope: every message timestamp mounts useClockFormat, so the
uncached path built a fresh Intl.DateTimeFormat per rendered message,
hundreds in a long conversation, even when the preference ignores the
locale entirely.

* fix: keep the Maldives on Friday in the week-start fallback

CLDR's lone Friday-first territory was in neither fallback set, so
dv-MV (and bare dv, which maximizes to MV) fell to Monday on engines
without week data, with no Friday override in the selector to recover
with. The three per-day sets consolidate into one region-to-day map.

* fix: complete the Sunday-first fallback from CLDR week data

The hand-picked ten Sunday-first regions left the System preference on
Monday for en-IN, id-ID, bn-BD, ur-PK, th-TH and the rest of the long
tail on engines without week data. The list is now every territory whose
und-XX week does not start Monday per CLDR, deprecated codes included,
with a note on how to regenerate it when CLDR moves a territory.

* fix: mock message context across markdown test suites and prevent global plugin cache leak
2026-08-23 18:41:14 -04:00
Marco Beretta
b55222b6f4
📆 feat: Weekly Schedules on Several Days (#15120)
* feat: weekly schedules on several days

A weekly cadence has always stored `daysOfWeek` as a list, and the API has always
accepted several, but the dialog offered a single day-of-week dropdown. That
picker could only hold `daysOfWeek[0]`, so a multi-day schedule created through
the API read back as running on one day, and the form needed a preservation rule
to avoid collapsing the rest of the set on an unrelated edit.

The dropdown becomes a row of toggles, one per day, so the control can represent
what the cadence already stores. The preservation rule goes with it: there is
nothing left to preserve once an untouched picker shows the real set.

Each pill is a toggle button rather than a checkbox because it renders as one,
and carries the long weekday name as its accessible label since "Mon" reads fine
at a glance but poorly aloud. The submitted set is sorted, so two schedules
picked in a different order are the same cadence.

Weekly with nothing selected is expressible in the form but not on the wire, so
it blocks submit with a message rather than silently saving as Monday.

* fix: keep the weekday pills one line tall and honest about empty sets

Seven 3rem pills wrapped to a second line inside their md cell, spending
height the dialog's no-scroll budget does not have; they now share the
row equally with the locale's narrow weekday labels, composed on the
shared Button so the pills carry its focus ring rather than a bare
feature-styled element. Both labels are built once per locale instead of
fourteen Intl constructions per keystroke, and the cadence summary no
longer describes the Monday fallback while the form says to pick a day.

* fix: name each weekday pill outright on hover

The narrow labels repeat within a week and read by position; until the
week-order preference lands, the position is fixed Sunday-first, so the
title gives a sighted user the full day name without relying on it.
2026-08-23 15:21:05 -04:00
Marco Beretta
1724489024
🧭 feat: Choose a Schedule's Time Zone (#15119)
* feat: choose a schedule's timezone

A schedule's timezone was whatever zone the browser reported when it was created,
and nothing in the dialog could change it afterwards. That is wrong for anyone
who travels, for a shared account, and for a team schedule that should follow an
office rather than whoever happened to open the dialog.

The zone becomes a picker over every IANA zone the runtime knows, with the user's
own zone and UTC pinned first. `Intl.supportedValuesOf` is unavailable on older
engines, so that pinned pair doubles as the fallback list: a user who cannot
browse zones can still keep the one their schedule already uses. Each option
carries its current offset, since a name alone does not tell two similar zones
apart.

A zone change on its own is a timing edit, so it is submitted like one. The
server recomputes the next run whenever the timezone changes and measures the
interval floor against the effective pair, which is what makes `0 0,12 * * *` a
12-hour gap in UTC and an 11-hour one in America/New_York on the day it springs
forward. The dialog now mirrors that: the cron field validates against the
selected zone and the floor is measured in it, so a cadence cannot be accepted
here and refused by the API.

* fix: keep zone-only edits out of the cadence and the zone list findable

A timezone-only edit rode the cadence dirty flag, so the PATCH carried a
cadence rebuilt from the form and could overwrite stored fields the
pickers cannot represent, an API-created hourly's nonzero hour for one.
The floor still validates a zone change as the timing edit it is, but
only touched cadence controls put a cadence on the wire, and the spec
now pins that down instead of only checking the zone.

The picker list also gains the modern IANA names supportedValuesOf
omits (it reports CLDR's legacy canonical forms, Asia/Calcutta for
Asia/Kolkata), each probed against the engine before inclusion, and the
per-zone offset labels are cached per locale so reopening the dialog
stops rebuilding ~400 Intl.DateTimeFormat instances.

* fix: carry the full tzdb rename set into the zone picker

Five names covered the famous renames but Node still accepts and omits
fourteen more modern canonical identifiers (the Argentina provinces,
Indiana and Kentucky city moves, Kathmandu, Asmara, Faroe, Chuuk,
Pohnpei, Kanton, Atikokan). The set is now the tzdb rename list, still
probed per engine and deduped; deprecated links like US/Eastern and the
sign-inverted Etc/GMT forms stay out deliberately, since they duplicate
zones already listed under their canonical names.
2026-08-23 15:02:41 -04:00
Marco Beretta
44d97f859d
⏰ feat: Custom Cron Cadence for Scheduled Chats (#15084)
* feat: custom cron cadence for scheduled chats

Scheduled chats could only be built from four fixed presets, each pinned to a
single hour and minute, so anything outside that shape (twice a day, every 15
minutes, the 1st of the month) was not expressible. This adds a Custom cadence
that takes a raw five-field cron expression.

The cadence schema becomes a discriminated union on `frequency`. A cron row
carries `expression` instead of the hour and minute it cannot represent, since
there is no single hour for `0 9,17 * * 1-5`, and the Mongo schema requires each
field only for the shape that has it: a blanket `required` would reject every
cron write, and dropping it entirely would let a structured cadence silently
fire at 00:00 with a missing hour.

Five fields only. croner also reads a six-field form carrying seconds and a
seven-field form that pins a year, and both are refused. Seconds would promise a
precision the engine does not keep, since it polls on a thirty-second tick and
offsets each schedule by up to two minutes of jitter. A pinned year makes a
cadence that runs out, and every place that computes a next run reads "no next
occurrence" as a cadence it cannot read.

Compilation, validation, next-run previews and interval measurement live in
packages/data-provider so the dialog and the engine share one parser and cannot
drift. The dialog previews the next occurrences, enforces the admin interval
floor and disables its own submit from the same functions the server validates
with, so it cannot offer a Create the API answers 400 to.

The interval floor now covers cron, and measures it twice, taking the smaller.
The nominal gap is probed in UTC and discounted by the same worst-case DST
allowance the structured branches carry, which keeps `0 9 * * *` reporting
exactly what the Daily preset reports. Real elapsed time is then measured in the
schedule's own zone across each of that zone's transitions, because
spring-forward compresses a gap that straddles one: `0 0,12 * * *` in
America/New_York is 11 hours that day, not 12, and a floor between the two would
otherwise be bypassed. The floor ships with the schedules list so the dialog can
mirror it rather than surfacing it as a 400 after submit.

Radio gains a wrap variant, since five frequency segments no longer fit one row
in a phone-width dialog and a translated label can push even a desktop one over.
Its indicator follows the selection across rows; the single-row default is
unchanged.

* fix: mark the cron input invalid when the interval floor rejects it

A floor-violating expression disabled Create and rendered the cadence
message, but the input itself still said aria-invalid=false and its
aria-describedby never reached that message, leaving a screen reader
user with a disabled Create and no stated reason.
2026-08-23 14:35:18 -04:00
Marco Beretta
69bf8ffdf1
📏 fix: Size Agent Builder Tool and Skill Rows (#15134)
The info and remove buttons on each Tools/Skills row passed `size-6 p-0`
through className without a `size` prop. tailwind-merge 1.14.0 has no
`size-*` group, so `size-6` never conflicted with the default size
variant's `h-10 px-4 py-2` and the buttons rendered at 40px. Because they
sit at opacity-0 until hover, the rows read as 52px of mostly empty
padding.

Pass size="icon-xs" so the recipe replaces the default outright. Rows go
from 52px to 40px.
2026-08-23 10:36:01 -04:00
Danny Avila
67b7b441b2
🛂 feat: Filter Model-Bound Content by Source (#14425)
* feat: introduce optional content protection seam

* feat: enforce source-aware content filters

* feat: complete source-aware content enforcement

* test: activate skill file-text fail-close fixtures

* fix: harden source-aware content filters

* fix: harden model-bound content filtering

* fix: preserve legacy filters and generated files

* fix: inspect shared scalar metadata

* test: align mocks with current dev dependencies

* feat: add persisted content filter safeguards

* feat: complete source-aware content filter enforcement

* fix: move resume content preflight into TypeScript

* fix: close content inspection edge cases

* fix: harden content protection boundaries

* fix: complete content protection safeguards

* test: align persisted memory filter coverage

* fix: reconcile content protection with current dev

* fix: reconcile content protection with latest dev

* fix: close content protection review gaps

* fix: enforce source-aware provider boundaries

* fix: preserve legacy PII preflight semantics

* test: stabilize stored branch preflight fixture

* fix: defer agent writes until protected model admission

* perf: harden source-aware model-bound filtering

* fix: canonicalize provider lineage before validation

* fix: satisfy model-bound callback type checks

* perf: Bound content protection filtering work

* fix: Bound submission array traversal

* fix: Stabilize bounded content snapshots

* fix: Scope model-bound traversal overflows

* fix: Preserve scoped content inspection

* fix: Accumulate aggregate traversal scopes

* fix: centralize content policy boundaries

* test: align deferred tool policy context

* test: align controller policy mocks

* style: normalize content protection imports

* fix: close content policy review gaps

* fix: narrow active skill policy config

* fix: address content protection review boundaries

* fix: retain exact provenance overflow sentinel

* fix: preserve literal and scoped provenance updates

* fix: narrow persisted edit provenance

* fix: isolate exact overflow attribution

* fix: centralize stored prompt protection

* fix: fail closed on incomplete transcript evidence

* fix: align canonical transcript routing

* refactor: centralize content policy preflights

* fix: isolate upload policy error typing

* style: sort policy preflight imports

* refactor: centralize content policy boundaries
2026-08-21 22:43:32 -04:00
Marco Beretta
199de92c51
⚡ perf: Warm Feature Catalogs in the Background After First Paint (#15047)
* perf: warm feature catalogs in the background after first paint

Prompt groups and MCP server/tool queries no longer fire on the app
startup path. A catalog warmup store releases them after first paint
on browser idle, staggered with jitter so a fleet of clients does not
burst the API all at once. Panels opened before warmup activates
their catalog immediately and fall back to their existing loading
states. The prompts list endpoints now also run their independent
access lookups in parallel instead of in three serial rounds.

* perf: gate MCP icon observers and re-arm warmup across sessions

MCP icon/name observers mounted from rendered messages now wait for the
warmup gate like every other server-catalog consumer, so conversations
with MCP tool calls no longer pull the server list onto the first-render
path. The warmup schedule resets on logout so a second login in the same
tab warms on its own stagger instead of releasing every catalog at once.
Panel mount activations now require a visible sidebar, since a persisted
active panel stays mounted while hidden.

* perf: void stale warmup callbacks and gate the agent panel tools query

Reset now bumps a generation captured by every idle callback and its
stagger timer, so callbacks pending across a logout can no longer release
catalogs into the next session. The agent form's MCP tools query keeps
its own readiness gate so a hidden persisted panel cannot pull the tools
request ahead of its stagger once the server list resolves.

* perf: reset warmup on Root unmount and honor the insights route collapse

Root can unmount in the same render that flips authentication on logout,
so the warmup effect now resets from its cleanup as well as the
unauthenticated branch. Panel activations mirror UnifiedSidebar's
panelExpanded condition, treating the insights route as collapsed instead
of reading the raw sidebar atom.

* test: re-expand the approval tool card the saved message remounts

The helper opened the card once and then waited on its body. Saving the
response swaps the placeholder message id for the persisted one, which
rekeys every part in the turn: the card remounts collapsed, its body
unmounts, and the output assertion waits out its timeout against a
disclosure nothing is going to reopen. The redis transport lane pays a
round trip per stream event, so its finalization lands late enough to
catch the helper mid-assertion.

Wait for the closing model turn before expanding anything, then re-open
the group and the card on each attempt until the scoped output is on
screen.

* Update AgentPanelContext.tsx import order

---------

Co-authored-by: Danny Avila <danny@librechat.ai>
2026-08-21 20:33:29 -04:00
Danny Avila
8ae94afa91
🪡 fix: Thread Parent Message ID Through MCP Request-Scoped Bodies (#15095)
* fix: Unify MCP request-scoped headers

* fix: address request-scoped MCP review findings

* test: preserve request scope on status errors

* fix: treat authorized on-demand MCP servers as ready

* refactor: separate MCP readiness from connection state

* fix: preserve on-demand MCP readiness labels

* test: satisfy OpenAI conversation ownership guard

* fix: keep MCP action predicates boolean

* fix: close deferred MCP request context gaps

* fix: preserve on-demand MCP configuration actions

* fix: fail closed on unavailable MCP parent context

* test: complete MCP connecting-state mocks

* fix: preserve missing MCP parent on continuations

* fix: align native MCP request identities

* fix: preserve edited MCP parent identity

* test: use scoped Agent initializer fixture

* test: expose MCP request body helper

* fix: preserve MCP turn identity across resume

* style: sort stream metadata imports

* fix: carry normalized MCP identity to execution
2026-08-21 16:33:01 -04:00
Dustin Healy
f02ce63d57
✂️ fix: Strip Redundant Server-Name Prefixes from MCP Tool Keys (#14732)
* ✂️ fix: Strip Redundant Server-Name Prefixes from MCP Tool Keys

MCP servers that prefix every tool with their own name produce model-facing keys that embed the server twice once the _mcp_<server> suffix is appended, pushing long tool names past provider 64-character function-name limits. Tool keys now drop a leading <normalizedServerName>_ prefix (case-insensitive, skipped when a sibling tool already owns the stripped name). The original upstream name is recorded as serverToolName on the cached definition and is always what tool calls send to the server, and runtime lookups also try the stripped spelling of persisted pre-strip keys so existing agents keep resolving.

* 🩹 fix: Keep Stripped MCP Tool Keys Provider-Safe and Collision-Free

Assistant writers submit catalog entries verbatim, so the internal serverToolName mapping is now removed from provider-facing definitions before they reach create/update payloads. Prefix stripping is collision-guarded over the resulting name set rather than raw siblings only, which also covers case-variant prefixed pairs under the case-insensitive match.

Assistant payload healing now rewrites a pre-strip persisted key to the stripped catalog key when that key actually exists in the loaded definitions, and legacy agent references keep their persisted spelling as the runtime instance name so per-tool options stay applied while the upstream call still uses the matched entry's raw name.

* 🧷 fix: Harden Stripped MCP Tool Keys Against Heal, Collision, and Cache Edges

The pre-strip heal now resolves the key boundary against both raw and normalized server spellings, mapping back to the raw name for the shadow and membership guards, so keys persisted after server-name normalization heal too. Collision detection iterates to a fixpoint so a fallback to a raw name cannot silently collide with another sibling's stripped result, and a stripped remainder equal to a synthetic marker (wildcard or server pin) is never produced.

MCP catalog cache slices are versioned so replicas that predate serverToolName never read stripped entries during a rolling deploy; stale slices expire on their own.

* 🔎 fix: Resolve Pre-Strip Keys in Event-Driven Definitions and Reinspect Persisted Catalogs

The event-driven definitions loader now tries the stripped spelling of a persisted key when the exact lookup misses, keeping the persisted name so it matches the runtime instance, which stops legacy agents from failing initialization with expected tools unavailable. The registry storage schema version is bumped so followers rebuild persisted toolFunctions instead of republishing pre-strip definitions into the versioned catalog namespace.

The assistants heal also fails closed when a normalized-suffix reference lands on a contested server-name slot, since rewriting persisted data must not bind an ambiguous reference to the tie-break winner.

* 🛰️ fix: Reserve the Synthetic OAuth Name and Heal User-Owned Server Keys

A stripped remainder equal to oauth would make the client stream handlers treat a real tool call as a synthetic authentication prompt, so it joins the reserved remainders alongside the wildcard and pin markers.

The assistants heal now audits the FULL accessible server set on every run instead of operator config names only, since assistants reference user-owned servers whose catalogs the definitions loader already resolves; an unavailable audit still skips healing entirely.

* 🧬 fix: Verify Upstream Identity for Legacy Keys and Reserve Sibling Raw Names

Stripped results now reserve every sibling's raw name even when that sibling itself strips, so a stripped key can never shadow another tool's pre-rollout persisted references within the same snapshot. Every legacy fallback (runtime lookup, event-driven definitions, assistants heal) accepts a stripped-spelling match only when the entry's recorded serverToolName proves the same upstream tool, so a stale key for a removed tool degrades to unavailable instead of calling a different sibling.

To keep that identity visible to the heal, assistant tool definitions retain serverToolName and the controllers sanitize entries through toProviderToolDefinition at the provider submission boundary instead. The agent editor migrates pre-strip persisted ids the same identity-verified way, with the upstream name exposed on the MCP tools payload.

* 🧭 fix: Heal Wildcard Tool Options and Reserve the OAuth Namespace

Wildcard-expanded catalogs rename stripped tools without any agent.tools entry to preserve the spelling, so buildToolClassification now aliases persisted pre-strip tool_options keys onto the current instance names in place, identity-gated on the definition's recorded upstream name and never overriding an explicit entry. Both loading modes flow through it: instances carry mcpServerToolName from createToolInstance and event-driven definitions thread serverToolName from the catalog.

stripServerNamePrefix also reserves the entire oauth namespace rather than the exact name, since the client stream handlers classify every oauth-prefixed key as a synthetic authentication call.

* 🛡️ fix: Derive the Full Reserved Namespace and Heal Approval Policies

The reservation guard now covers every namespace consumers classify by prefix: the wildcard and server-pin markers alongside oauth, plus the server-scoped mcp_ pluginKey namespace that pre-strip keys could never enter. Stripping also never produces a key whose isActionTool classification differs from the raw key's, since a server whose normalized name contains _action_ would otherwise see a real MCP tool routed down the OpenAPI action path past MCP authorization.

Admin toolApproval globs written against upstream tool naming keep applying: pattern lists are healed at run wiring with the current names of tools whose pre-strip spelling matches, list-level so deny, ask, and allow precedence is unchanged and a non-matching deny can no longer fail open. The MCP tools wire type also declares serverToolName end to end.

* 🪪 fix: Alias Both Key Spellings for Approval Policies and Hook Matchers

Identity aliases are now collected once at tool classification, in both directions: a stripped instance aliases its pre-strip spelling and a legacy-named instance aliases its current catalog spelling, with the current name recorded on legacy matches by the runtime lookup and the event-driven definitions loader alike. The aliases ride the agent config through both loading modes, so approval pattern healing applies to deny rules written against either spelling, closing the bypass where a rule targeting the current name missed an unedited agent's legacy instance.

Programmatic approval hook matchers get the same treatment: each hook is additionally registered under an anchored exact-name pattern for tools whose other spelling its regex matches, keeping the admin's matcher semantics intact while argument, user, and tenant specific deny or ask decisions keep executing for renamed tools.

* 🔁 fix: Alias Tool Options in Both Spelling Directions

Options aliasing now consumes the same bidirectional alias pairs as policy healing and hook matchers, so options the editor migrated to the current catalog spelling still reach a legacy-named instance retained by an unedited agent.tools entry. The previous serverToolName-only derivation skipped exactly that case since the legacy key equals the instance name there.

* 🤝 fix: Reserve the Agent Handoff Namespace Before Stripping

The client renders any lc_transfer_to_ prefixed call as an agent handoff and the background and intent passes exclude such names, so a stripped remainder inside that namespace would misclassify a real upstream tool. It joins the mcp_ pluginKey namespace as a bare-prefix reservation, which pre-strip keys could never enter.

* ⚡ fix: Reuse the Loader's Server Snapshot and Index the Editor Catalog

getAssistantToolDefinitions now returns the accessible-server snapshot from the same merged registry read that resolved the catalogs, and the heal consumes it instead of repeating the app-config and registry round trips on the assistant write path; without a snapshot the heal still fetches and fails closed as before.

The agent editor's id migration uses a memoized tool_id map, so the per-key form heal does constant-time lookups instead of scanning the catalog per option.

---------

Co-authored-by: Danny Avila <danny@librechat.ai>
2026-08-21 14:30:29 -04:00
Danny Avila
d602452c05
🪪 fix: Support MCP Server Titles With Hyphens (#15094)
* fix: Support MCP Display Titles With Hyphens

* fix: Preserve Legacy Regex Target Compatibility
2026-08-21 12:44:31 -04:00
Danny Avila
876a087558
🛰️ feat: Show Child Agent Activity in a Side Panel (#15075)
* feat: add parent-scoped subagent thread reads

* fix: tighten child thread read bounds

* perf: project child activity messages

* test: update child activity route fixtures

* test: satisfy response mock types

* fix: bound child activity reads at storage

* style: sort child activity imports

* fix: bound child activity storage reads

* feat: show child activity in a parent-owned panel

* fix: preserve side panel identity

* fix: refresh child activity safely

* fix: follow the selected child task

* test: update child panel fixtures
2026-08-21 11:10:41 -04:00
Danny Avila
8c14f03432
🗂️ feat: Scope Scheduled Chats to Chat Projects (#15056)
* 🗂️ feat: Scope Scheduled Chats to Chat Projects

Adds an optional chat-project destination to a schedule, plus the operator
config to require one — or to pin every scheduled run to a specific project.

Feature:
- `chatProjectId` on the schedule row, accepted on create/update, projected on
  the wire, and carried into the run's conversation through the durable trigger
  envelope's `run` context.
- `interface.schedules.requireProject` refuses schedules that are not filed
  under a project; `interface.schedules.projectId` pins every run to one
  project and implies the requirement.
- Dialog gains a project picker (required when configured, a read-only row when
  pinned); the card shows the destination and the new disabled reasons.

Invariants:
- ONE resolver (`resolveScheduleProjectId`) decides the destination for the
  write handler, the fire path, and the wire projection alike, and an operator
  pin OUTRANKS the stored id in all three. Tightening the config therefore
  redirects — or stops — existing schedules instead of grandfathering where
  their runs land. A pin implies `requireProject` for the same reason: without
  it, a row created before the pin would keep firing with no project at all.
- Create/fire precheck symmetry, mirroring `resolveAgentFireAccess`: a write
  this handler accepts is one the next fire also accepts. Any edit leaving a
  schedule ENABLED re-validates its EFFECTIVE (possibly stored) project, like
  the existing stored-agent and cadence-floor rechecks. A DISABLING edit skips
  the requirement, or a schedule auto-disabled for `project_required` could
  never be turned off.
- Fire-time enforcement auto-disables rather than filing runs loose, matching
  agent_deleted: new `project_required` (requirement raised after creation) and
  `project_deleted` (gone, or pinned to a project this owner does not have)
  reasons, both refused BEFORE a billed generation is dispatched, and both
  advancing so a schedule can never wedge on the occurrence.
- `computeCreateDigest` appends the field only when present, so a payload
  without a project digests byte-identically to one from before this change —
  a create retried across the upgrade still matches its own row instead of
  reading as key reuse.
- Project reads are scoped to the owner, so ownership and existence are the
  same lookup; a read ERROR propagates instead of failing closed, so a Mongo
  blip retries the fire rather than auto-disabling the schedule.

The trigger idempotency key hashes principal/event/target and never
`envelope.run`, so the added run field cannot destabilize delivery identity.

* 🗜️ fix: Keep the Schedule Dialog Inside Its Height Budget

The project picker landed as a new ROW in the schedule dialog, which broke the
e2e edit spec: `md:overflow-visible` turns off the template's scrolling from
`md` up, so the dialog's content must fit the viewport. The extra row pushed the
footer's Save button below a 720x1280 window, where Playwright reported a
visible, enabled button it could never click — 226 scroll-into-view retries and
a 2-minute timeout, on all three attempts.

Measured against `dev` at 1280x720 (Save button's bottom edge, viewport 720):
  dev              688   (32px slack)
  project row      ~790  (off-screen, CI failure)
  3-column row     704   (16px slack — half the budget spent)
  this commit      686   (34px slack, 2px better than dev)

The identity row is now three columns — name, agent, project — and its caption
moved out of the agent cell to sit full width beneath the row: at a third of the
dialog that sentence wraps an extra line, and the row is the tallest thing
competing for the budget. The caption is grouped with the row rather than left
to the form's own 4-unit rhythm, which spent more height on the gap than the
caption occupies.

The e2e spec now asserts the button is in the viewport before clicking it, so
the next field that overflows this dialog says so in one line instead of a
two-minute timeout on a visible element. FOLLOWUPS.md records what the planned
dialog controls (multi-day weekly, timezone, attachments) need first: give
`ControlCombobox` the `portalElement` prop `Dropdown` already has, portal the
popovers into the dialog content, and let the form scroll again.

* 🧹 fix: Address Codex Review on Scheduled Chat Project Scope

Four P2 findings, all real.

Unreachable clearing path (ScheduleDialog). The picker only held live projects, so
`com_ui_schedule_project_none` was a PLACEHOLDER — nothing selectable. Once a
schedule had a project the owner could never take it away, leaving the server's
`chatProjectId: null` path reachable only by API. The picker now carries a real
"No project" option whenever a project is optional, and omits it when one is
required, where there is nothing valid to select.

Placeholder shown for a real project (ScheduleDialog). A stored or pinned project
outside the first loaded page had no name in the paged map, and the combobox
renders its placeholder for an empty display value — telling the owner a scoped
schedule had no project. That one project is now read by id, with the raw id as a
last resort: a poor label, but an honest one.

Project policy skipped at the resume boundary (service.ts). `claimScheduleResume`
re-applied the schedules gate, the revision fence, the kill switch and
SCHEDULES:USE, but not the project policy this PR added. Approving a paused run
whose project was deleted — or whose owner now sits under a requirement or a pin
it no longer satisfies — billed a continuation the very next scheduled fire would
refuse and auto-disable the schedule for. The effective-project resolution now
runs there too, refused before the lease and the capacity slot so a policy refusal
costs nothing and leaves no state to unwind. NOTE: agent access and balance are
still not rechecked on resume; that gap predates this PR and is left alone.

Per-card project derivation (ScheduleCard). Every card ran the projects hook and
rebuilt the full option array, name map, and one icon element per project, to use
a single name — O(schedules x projects) per render and per project-list refresh.
The hook is split: `useChatProjectNames` (map only, for the panel, which resolves
every card's name once and passes it down) and `useChatProjectPicker` (options and
pagination, for the dialog's one combobox). The panel skips the query entirely
until some schedule actually has a scope.

Tests: four at the resume boundary (verified failing without the gate) and four in
the dialog spec. The picker selection in one existing test now goes through the
search field — the popover's VIRTUALIZED renderer sizes its window from a scroll
height jsdom always reports as 0, so with three options it materialized only two.
Full schedules e2e re-run green against the rebuilt client.

* 🎯 fix: Settle Project-Policy Refusals and Keep Create Retries Idempotent

Second Codex round, four P2s. Three were consequences of the resume gate added in
the previous commit, which was half-built: it admitted where it should not and
stranded the run where it refused.

Project policy moves from `claimScheduleResume` into `isScheduleLive`'s `policy`
branch. Both entry points consult that branch FIRST, and both already route its
refusal through abort-and-settle — so a policy stop now settles the occurrence
instead of answering a bare 409 while the job stays `requires_action`, the card
keeps reading "Needs approval", and every retry repeats the same 409 until expiry.
No change to resume.js: the existing branch does the work.

The rule is deliberately NARROW. It refuses only where no valid destination is
left — the requirement is on with nothing satisfying it, or the schedule's own
project is gone (which also unset it on the conversation). It does NOT refuse
because an operator's pin moved: the paused conversation cannot be rebound
(`chatProjectId` is excluded from the resume context and the continuation reuses
the same conversationId), so refusing would strand a pending approval over a
destination it can never reach, for a pin that governs only where the NEXT run
lands — which the fire path already redirects.

Create retries are idempotent again. Project policy had been applied BEFORE the
`clientRequestId` replay lookup, so a raised requirement, a deleted project, or a
moved pin could answer 400 for a create that already committed — pushing the
client to rotate its key and create a DUPLICATE schedule, the exact failure the
key exists to prevent. Policy now applies only to a genuinely new insert, and the
digest is computed from the CLIENT's payload rather than the resolved destination,
so today's policy can no longer re-digest a genuine retry into a mismatch.

An explicit `chatProjectId: null` under a pin is refused rather than silently
resolved to the pin. The payload contract defines `null` as clearing the scope;
answering 201 while filing under the pin reported success for the opposite of what
was asked. Only an OMITTED field takes the pin silently.

Tests: five on the policy branch (both refusals verified failing without it, plus
guards that a moved pin and a live project still admit) and two on the handlers
(the pinned explicit clear, and a committed create recovered by retry after the
policy tightened — verified answering 400 without the reordering).

* 🧭 fix: Converge the Stored Project on the Destination a Fire Resolved

Third Codex round, four P2s.

The root confusion behind the resume findings: an operator pin outranks the stored
id at fire time, `fireSchedule` sends the pin in the trigger envelope, and the row
keeps its old value. The row therefore LIED about where that occurrence's
conversation went, and every later re-validation — the resume boundary above all —
checked a project the conversation was never filed under. A schedule storing A,
pinned to B, with B later deleted and A still live, was admitted for resume into a
conversation that had just been unscoped.

Fixed at the source rather than at each reader: a fire that resolves a destination
different from the stored one writes it back, claim-token fenced like every other
worker-side write and deliberately WITHOUT a configRevision bump — this is the
server reconciling itself to policy, not an owner edit, and a bump would fence an
in-flight occurrence off its own run. Written only AFTER the destination validates,
so an unusable pin never lands in the row, and best-effort: the envelope already
carries the right destination, so a failed write costs accuracy on a later recheck,
never the run itself. The wire projection already reported the pin, so this also
stops the row and the UI disagreeing.

An explicit `chatProjectId: null` under a pin is now refused on the DISABLING edit
path too. The pin check and the requirement are independent rules, and folding them
together let `{enabled: false, chatProjectId: null}` skip the pin check entirely,
unset the row, and answer with a wire projection still naming the pin. Only the
requirement is waived for a disabling edit.

The dialog no longer requires a project for an edit that leaves a schedule DISABLED.
The server waives the requirement there precisely so a row auto-disabled for
`project_required` can still be renamed or tidied up; requiring it in the form made
that unreachable, and an owner with no projects could not edit the stopped schedule
at all.

FOLLOWUPS.md records the two residual gaps with their exact triggers: the sub-second
deletion race inside the resume claim window (which needs the effective project
persisted per OCCURRENCE plus a distinct policy conflict routed through
abort-and-settle), and the fact that convergence happens only when a schedule fires.

Tests: four on convergence (pin written, no write when unchanged, no write for a
destination that failed validation, fire survives a failed write), two on the
disabling-edit pin rules, two in the dialog. Schedules e2e re-run green.

* 🔑 fix: Keep an Explicit Project Clear Out of an Omitted Field's Digest

`computeCreateDigest` appended `chatProjectId` on `!= null`, so an OMITTED field and
an explicit `null` produced the same digest. Because the replay lookup and
`matchesCreateIntent` deliberately run before project policy, a request could reuse a
pinned create's `clientRequestId` while explicitly sending `chatProjectId: null` and
receive 201 describing the pinned row — success reported for the opposite of what it
asked, and the pinned-clear refusal the normal create path applies never reached.

Now `!== undefined`: an omitted field still digests byte-identically to a payload from
before project scope existed, so a create in flight across the upgrade still matches
its own row, while an explicit clear is a distinct intent and digests differently. A
pre-scope client never sent the field at all, so nothing legacy can carry an explicit
null.

* 📍 fix: Validate a Paused Run Against the Project Its Own Occurrence Used

The schedule-wide convergence from the previous commit was not enough, and the
reason is the single-active run index: it covers `status: 'started'` only, so a
PAUSED run does not block the next occurrence. While run 1 sat paused in project A,
a pin move plus one later fire rewrote the schedule row to B — and the resume
policy then validated B while run 1's conversation was still filed under A. Delete
A and the continuation was admitted into a conversation that had just been unscoped.
That window lasts as long as the pause, not the sub-second race the previous commit
documented.

The reservation now records the destination THIS occurrence used, and
`isScheduleLive` validates that record when given the occurrence's `scheduledFor` —
which `resume.js` already reads two lines above the call. No new conflict type and
no settlement-path surgery: the refusal rides the abort-and-settle branch that check
already has.

An ABSENT record falls back to the schedule-level resolution rather than reading as
unscoped. A pre-scope occurrence, or one whose row is gone, must never be treated as
evidence to stop a run — the fallback keeps legacy paused runs behaving exactly as
they do today.

Schedule-wide convergence stays: it keeps the row honest for the UI and for every
check that has no occurrence in hand.

FOLLOWUPS.md now describes the one remaining gap accurately — a deletion inside the
claim window, which needs a distinct policy conflict routed through abort-and-settle
rather than the bare 409 an `inactive` conflict produces.

Tests: three on occurrence-vs-row precedence (the decisive one verified failing
without the lookup), two on what the reservation records, and the resume controller
spec now pins `scheduledFor` in the policy call. Schedules e2e re-run green.

* 🏷️ fix: Tell a Deliberately Unscoped Occurrence From an Unrecorded One

The occurrence fallback added in the previous commit conflated two different
absences. A post-upgrade run that deliberately went unscoped omitted the field
exactly like a row written before the field existed, so both took the fallback —
and a paused unscoped run was then validated against the schedule's CURRENT
project. Under a requirement or a pin added while it sat paused, that admitted a
billed continuation into a conversation satisfying no present policy.

The reservation now ALWAYS records its decision, `null` for unscoped, and the read
reports `recorded` from key PRESENCE rather than truthiness. Only an unknown record
— a pre-scope row, or no row at all — falls back to the schedule-level resolution;
a recorded null is the genuinely unscoped occurrence it says it is, and is refused
once a project becomes required.

The distinction rests entirely on a stored `null` surviving as a present key while a
never-written field stays absent, so that is asserted against real Mongo rather than
assumed: if it ever stopped holding, unscoped runs would silently start being
validated against the schedule's current project again.

Tests: two against mongodb-memory-server (recorded null vs never-written vs missing
row), plus refusal of a recorded-unscoped occurrence under a new requirement and the
preserved fallback for a pre-scope one. Schedules e2e green.

* 🔁 fix: Validate an Initial Scheduled Start Against Its Own Occurrence

The initial-start policy check in `request.js` called `isScheduleLive(..., { policy:
true })` without `scheduledFor`, so it fell back to the schedule-level resolution
even though the run row already carries the occurrence's recorded scope. An
occurrence reserved unscoped, with a pin introduced while its loopback request sat
queued, was therefore admitted against the new pin — producing a billed unscoped
conversation under a requirement it does not satisfy, from an envelope already built
without a project.

`scheduledFor` was already in scope there. Passing it makes the initial start and the
resume validate the same way: against the destination the occurrence itself recorded.
2026-08-21 11:09:04 -04:00
Danny Avila
276f5f88fe
🗓️ fix: Hide Unsupported Schedule Variables (#15053) 2026-08-20 18:33:00 -04:00
Danny Avila
c5276fc63d
⏱️ feat: Run Scheduled Chats Through Durable Agent Triggers (#14939)
* feat: Scheduled Chats — agent-centric scheduled runs creating real conversations

Squash of the full review-hardened branch (PR 14540, supersedes 14373) onto
latest dev, preserving the exact verified tree. History prior to this commit
lived on the pre-squash branch; every invariant below survived 25 Codex review
rounds plus two external audit rounds (R26) with regression tests that fail
without their fixes.

Feature:
- Schedules CRUD + side-panel UI (cadence dialog, run cards, Run Now), roles/
  permissions (SCHEDULES:USE), interface.schedules availability, per-user limits
  and capacity slots, timezone-aware cadence with DST-conservative floors and
  misfire grace.
- Engine: single-process claim/fire loop with leases, loopback POST dispatch
  (signed schedule-fire JWT claims, per-occurrence idempotency key inside the
  route's clientRequestId charset), overlap/balance/capacity/duplicate skip
  policies with auto-disable streaks (too_many_failures, insufficient_balance),
  reconciliation from retained terminal-job evidence, erasure sweep.
- Scheduled runs create real conversations through the resumable agents chat
  path: HITL pauses surface on the card (requires_action), resumes re-apply the
  fire boundary's admission policy (revision fence, enabled, global kill switch,
  SCHEDULES:USE, availability) before continuing a billed generation.

Correctness invariants (the audit surface):
- Settlement discipline: every persistence-producing write happens-before a
  run's terminal outcome write; Stop/complete/pause race through single-winner
  terminal CAS claims (dev's TerminalJobClaim substrate) with retained,
  completedAt-less evidence for scheduled fires plus an owner-intended outcome
  stamp (scheduleOutcome) the reconciler prefers over re-derived success —
  round-tripped through the Redis hash mapper.
- Swallowed generation failures (client error content parts) classify to
  error/skipped_balance instead of success on both initial and resumed paths;
  stale stamps are refreshed evidence-first when persistence plus the Mongo
  outcome write both fail.
- Abort honesty: delivery judged by generation ownership and the CAS's actual
  from-status; republication escalates to the transport's acknowledged variant;
  the Stop route settles only genuinely paused runs.
- Account deletion: one-way barrier (deletionRequestedAt) with auth-cache
  tombstone-before-stamp, boundary rechecks across all auth strategies, quiesce
  of scheduled + interactive work with durable per-stream abort fences
  (positive-evidence acknowledgement only), owner-side finalization markers for
  post-terminal billed writes, deferred-deletion sweep (explicitly ensured
  partial index) that completes cascades autonomously. Remote OpenAI-compatible/
  Responses requests are documented as outside the quiesce and tracked in issue
  14594.
- Store compatibility: finalization markers optional on the legacy IJobStore
  contract with coherent degradation (registration fails -> synchronous title
  fallback; count reads 0).

* fix: fence terminal response persistence from deletion; retain stale-pause evidence

Two blockers from the third external review round.

Terminal persistence visible to account deletion:

- The finalization marker was registered only when post-terminal TITLE work was
  possible, but every persistence-owning terminal CAS opens the same window: the
  claim drops the job out of the active set BEFORE the response save (and the
  background user-message/convo saves), so a deletion quiesce landing there saw
  neither an active job nor a marker and could cascade while the admitted request
  could still recreate messages. Both controllers now register the marker before
  every persistence-owning claim — the fresh-turn path and the HITL resume — and
  release it only after their pending saves (and any post-terminal title) have
  landed, on success and failure paths alike. The TTL bounds a crash.
- settleAbortFence no longer clears a complete/error fence while
  `terminalPersistencePending` is set: terminal at the CAS is not settled while
  the owner is still persisting. The job facade now surfaces the flag.
- The marker trio is REQUIRED by the runtime store contract (assertJobStoreV2
  refuses a store without it at configure time, keeping the failure loud and
  deterministic) while remaining optional on the legacy public IJobStore type for
  source compatibility. The silent degrade path from the previous round is gone —
  it was not deletion-safe.

Stale-pause recovery retains scheduled evidence:

- Three crash/timeout recovery paths — ApprovalLifecycle.failStalePausePersistence
  and the InMemory/Redis stale-pause cleanups — unconditionally stamped
  `completedAt`, putting a scheduled fire's failed-pause error terminal on the
  short completed TTL. A Mongo outage longer than that TTL erased the evidence and
  the reconciler recovered the run as `interrupted` instead of `error`. All three
  now follow the controller-observed path from the previous round: scheduled jobs
  omit `completedAt` (retained-evidence TTL) and stamp the error outcome.

The PR description now explicitly narrows the deletion guarantee for the remote
OpenAI-compatible/Responses paths (tracked in issue 14594).

* fix: deletion-fence marker protocol — generation-scoped, atomic, fail-closed, all terminal paths

One consolidated pass over the finalization-marker mechanism, per the fourth
external review round. The invariant it establishes: NO persistence-owning
terminal CAS runs without a durable, generation-qualified marker covering the
window it opens, and every consumer treats a pending terminal as unsettled.

- Generation-scoped markers. Entries were keyed (userId, streamId), so a
  Stop-superseded generation finishing late could clear the marker its
  replacement registered on the same conversation. Marker fields are now
  qualified by the generation's createdAt; clears must present the same
  identity, and an unqualified legacy clear cannot drop a qualified entry.
- Atomic Redis registration. HSET-then-EXPIRE loses the fresh marker when the
  user's existing hash expires between the two commands (or the process dies
  there); registration is now a single Lua script carrying both.
- Fail closed everywhere. Registration failure (after one retry) now REFUSES
  the terminal CAS instead of proceeding uncovered: the completion claim throws
  into the error path, the error path skips completeJob and leaves the job
  ACTIVE — deletion-visible by itself, recovered by the stale-running reaper —
  and abortJob returns a new retryable `fence_unavailable` failure the Stop
  route answers with 503 and the deletion quiesce treats as an unacknowledged
  stop (fence kept). The previous round's log-and-proceed is gone.
- Every terminal path enrolled. abortJob now owns its window (register before
  the abort CAS, clear in its finally — the Stop route's checkpoint prune and
  partial save run inside beforePublish, between CAS and publication); the
  interactive and background generation-error paths register before their
  completeJob; the resume controller's error finalization registers before its
  completeJob. A lost or thrown claim releases the marker after pending saves
  flush instead of holding the user's deletion behind the TTL.
- Every pending terminal unsettled. settleAbortFence defers on
  terminalPersistencePending for ALL statuses — including `aborted`, whose
  route-side persistence the previous guard missed.

Also: a direct Redis regression for scheduled stale-pause retention (the P2
test gap), and the PR description no longer claims to carry every commit.

* fix: lease-token lifecycle fences — same-generation isolation, admission fence, undelivered-Stop retention, legacy abort enrollment

Fifth external review round; four P1s, handled as the requested consolidated
lifecycle-fence pass.

- Lease tokens. Marker fields were (streamId, createdAt), shared by every
  contender on the same generation — completion and Stop, or two racing Stops —
  so a losing contender's cleanup erased the winner's still-live marker. Every
  registrant now carries a unique lease token in the field and may only ever
  clear its own lease; unqualified legacy clears cannot touch qualified entries.

- Admission fence. Authentication can pass before the deletion barrier goes up,
  and the durable createJob is several async steps later — a deletion quiesce in
  that window saw neither an active job nor a marker and could cascade before
  the admitted request created its job. The controller now registers an
  admission lease and THEN rereads the deletion barrier: the ordering guarantees
  either this request observes the barrier (403, lease released, slot/claim
  cleanup) or the quiesce observes the lease and defers. Held until createJob is
  durable; released on every refusal and initialization-error path. Fail closed
  when the lease itself cannot be registered (503 retryable).

- Undelivered-Stop retention. abortJob released its lease in a finally even when
  delivery AND publication had provably failed — the job reads terminal
  (invisible to active-set scans), a user Stop writes no durable abort fence,
  and the remote owner keeps generating and will persist its abort-catch writes
  whenever the signal finally lands. The lease is now retained in exactly that
  case, and each resignal attempt heartbeats a fresh lease so the fence outlives
  the TTL for as long as delivery is still being driven. The abort-winning
  turn's own loser-side pending saves are additionally fenced in the controller
  catch (best-effort — those writes are already in flight).

- Legacy abort enrollment. abortMiddleware (assistants abort route fallback for
  non-assistants endpoints) awaited abortJob and then spent usage and saved the
  stopped response AFTER the abort's lease was released. Both writes now run
  inside `beforePublish`, between the abort CAS and publication, covered by the
  same lease as every other abort.

Barrier tests, each verified to fail without its fix: same-generation lease
isolation (store), racing two-Stop loser cleanup (manager, stale-read forced
CAS race), undelivered-Stop lease retention, resignal heartbeat, admission
refusal with lease-before-reread ordering plus release-on-durable-create, and
legacy-abort persistence inside beforePublish.

* fix: heartbeat-backed owner-lifecycle leases close the settlement handoff races

Sixth external review round: the remaining P1 interleavings were one structural
problem — lease handoffs that were not atomic — resolved as the requested
consolidated lifecycle-lease pass.

- Quiesce reads leases BEFORE the active-job scan. The admission-lease -> durable
  -job handoff is only atomic against a reader in the OPPOSITE order of the
  writer: writers hold the lease strictly until the job is active-set visible,
  so leases-first shows every interleaving either the lease or the job.
  Jobs-first allowed a request to create its job after the scan and release its
  lease before the count — hiding both, cascading, and letting the new
  generation persist into a deleted account.

- The abort acknowledgement is fenced by the owner-lifecycle lease. Redis ACKed
  the moment the owner's AbortController tripped; the stopping side released its
  lease on that ACK while the owner's asynchronous abort-catch persistence was
  still ahead. The transport now awaits a manager-installed pre-ACK hook that
  registers a DETERMINISTIC owner lease (exactly one owner exists per
  generation, and determinism is what lets the signal-time registrant and the
  owner's catch-side release agree across processes) before the acknowledgement
  is persisted or published; an owned same-replica abort bridges to the same
  lease before tripping its local controller. Both generation-owner catches
  (fresh turn, resume) release it once their writes land.

- A failed replacement handoff no longer orphans the predecessor. The atomic
  replacement removes it from active storage, and an unconfirmed handoff
  terminalizes the replacement too — leaving nothing a quiesce could discover
  while the predecessor's provider may still be generating. Its owner lease is
  now retained at the point the receipt fails delivery; the owner replica renews
  it through the pre-ACK fence when the signal finally lands.

- Leases HEARTBEAT while held. The five-minute store TTL only bounds a crashed
  holder; live persistence — a stalled save, a long deferred title — must never
  outlive its own fence. holdUserFinalization registers and renews every minute
  until released; the controllers' completion/error/admission leases all hold.
  The undelivered-Stop retention moved to a deterministic `stop` lease that
  every resignal attempt renews and the first successful one clears (no more
  opaque leases accumulating to TTL), and a THROWN abort transition releases the
  contender lease instead of leaking it.

- The user-document abort-fence mutations now invalidate the auth user-doc
  cache, matching every other user-doc write.

Barrier tests, each verified to fail without its fix: quiesce lease-scan
ordering (plus the observed-lease defer), pre-ACK fence ordering at the
transport, owned-abort owner-lease bridging, replacement-handoff predecessor
retention, held-lease heartbeat past the TTL, and failed-then-successful
resignal reaping the retained stop lease.

* fix: one manager-owned owner-lease span across every abort delivery path

Seventh external review round; four lifecycle-fence gaps, closed by making the
owner-lifecycle lease a single manager-owned, heartbeat-held span.

- Fail-closed acknowledgements. The pre-ACK hook registered a one-shot lease and
  the transport ACKed even when it failed; a same-replica owned abort likewise
  swallowed registration failure. The hook now acquires a HELD owner lease
  (heartbeat until the owner's catch releases it via releaseOwnerLease) and a
  rejection SUPPRESSES the acknowledgement — the stopping side stays retryable
  behind its retention lease, and every resignal re-drives the handler. The hook
  also stopped gating on `job.createdAt === generationId`: during a replacement
  handoff the store holds the replacement while the abort targets the
  predecessor, and that gate silently skipped exactly the generation being
  acknowledged (the store job is owner identity, never a generation gate). A
  local owned abort acquires the same held lease before tripping its provider;
  post-CAS the trip cannot be withheld, so acquisition failure downgrades
  delivery and the retention handoff keeps the user fenced.

- Committed-but-lost-reply disambiguation. A thrown abort transition released
  the contender lease as if nothing had happened, but a Lua CAS can commit and
  lose its reply — an aborted job invisible to active-set scans whose provider
  was never signalled, with no fence left. The throw path now re-reads the exact
  generation: only a job still live under the caller's identity proves no
  commit; committed or ambiguous outcomes hand the fence to the deterministic
  stop lease (kept on the contender lease if even that fails) before rethrowing.

- Replacement handoff covered end to end. A LOCAL replacement abort acquires the
  predecessor's held owner lease before the trip (failure reports the receipt
  undelivered, engaging retention). Failed-handoff retention is no longer a
  swallowed one-shot: it heartbeats with the durable acknowledgement proof as
  its renewal predicate — acquisition failures keep retrying for as long as the
  fence is needed, and the retainer stands down (without clearing the shared
  field) once the remote owner ACKs and thereby holds its own lease.

- A LOCAL resignal delivery hands off to the owner lease BEFORE clearing the
  retained stop lease, and keeps the stop lease when that handoff fails.

Barrier tests: hook rejection suppressing the ACK, commit-then-lost-reply
retention with its provably-uncommitted counterpart, local-resignal owner
handoff ordering, pre-ACK owner lease held past the store TTL until release
(and provably stopped after), and failed-handoff retention retrying on its
heartbeat — verified fail-before/pass-after by stashing the fixes.

* fix: finish scheduled chat lifecycle hardening

* fix: close scheduled chat review follow-ups

* fix: generation-fence abort recovery evidence

* test: wait for settled approval tool output

* test: preserve scheduled init reconciliation option

* refactor: rebuild scheduled chats on durable agent triggers

* test: reset MCP cache mock between cases

* test: isolate scheduler startup in server specs

* fix: harden scheduled run lifecycle

* fix: normalize schedule capacity conflicts

* test: type schedule collision fixture

* fix: re-fence scheduled resume and expiry

* fix: fence scheduled resume handoffs

* fix: release superseded manual schedule leases

* fix: release failed run-now claims

* fix: release superseded engine claims

* fix: repair schedule dialog interaction and rework its form

The agent picker was unusable: ControlCombobox portals its popover to the
body by default, which lands it outside the dialog's Radix focus trap. Clicks
passed through it, it could not be tabbed into, and the trap fighting Ariakit
for focus locked the page up on selection. The prop is documented for exactly
this case — pass `portal={false}` and give the dialog `overflow-visible`, as
ProjectButton already does. The time and day dropdowns defaulted the same way.

Alongside that:

- Extract the agent builder's instructions editor (special-variable menu plus
  expand-to-fullscreen) into a controlled `VariableEditor` and use it for the
  schedule prompt. Insertions now route through `onChange`, so react-hook-form
  sees them — the schedule PATCH is built from `dirtyFields`, and a `setValue`
  that skipped dirty tracking would have dropped an inserted variable silently.
- Replace the hand-rolled frequency buttons with the shared `Radio`. They marked
  the selection with `bg-surface-hover` on an outline button whose hover is the
  same token, so the selected option was indistinguishable from a hovered one;
  `Radio` is also a real radiogroup rather than four `aria-pressed` toggles.
- Wrap the fields in a real `<form>` and associate the footer button by id, so
  Enter submits. Group the frequency, day and time controls in fieldsets.
- Add placeholders for name and prompt, match the textarea fill to the other
  fields, and label the hourly case as minutes past the hour.
- Widen the dialog to `md:max-w-3xl` and pair name with agent so the form fits
  without scrolling on desktop.
- Move scheduled chats below skills and above prompts in the side nav.

The new dialog spec fails when the portal fix is reverted.

* test: cover scheduled and subagent deletion drains

* refactor: own the form-control appearance in the client primitives

Addresses the codex finding on ScheduleDialog: a feature-local `FIELD_CLASS`
restated the `Input` primitive's border, radius, height and background so it
could be pasted onto the schedule dropdowns, leaving those controls with no
connection to the primitive they were imitating.

Move that appearance into `packages/client/src/components/Field.ts` as the
single source `Input` and `Textarea` now compose, and give `Dropdown` and
`ControlCombobox` a `variant="field"` that applies it. The schedule dialog
passes the variant and carries no class strings of its own.

This also repairs a break the dev merge would otherwise have introduced: the
newer `Dropdown` splits `className` (wrapper) from `triggerClassName`, so the
old pasted classes would have landed on the wrapper and left the triggers
unstyled.

The semantic-token guard now watches the shared module and asserts each
primitive still composes it, which covers more than the two files it read
before.

* fix: keep an explicit schedules disable from becoming an opt-in

`use` is two things at once for a dual-purpose runtime interface field: a
permission bit, which DB overrides strip, and the runtime disable signal that
`getLimits` reads. Stripping it from `{ use: false, maxPerUser: 2 }` leaves an
object, and `getLimits` treats any object without `use: false` as enabled — so
an admin override written to stop scheduled billing for a role or user started
it instead.

Collapse an explicit disable to the boolean form before the strip, on both
paths that accept it: the `interface.schedules` field patch, which admitted the
object wholesale because bare runtime paths deliberately bypass the permission
gate, and the overrides merge, which reached the composite-field branch and
kept `maxPerUser`. Objects that only narrow limits are untouched, so a
principal can still be given a smaller cap.

Both regressions fail without the normalizer.

* fix(schedules): Wave A — null-balance CAS, atomic paused-card clear, clustered erasure sweep

Slice 1 (thread r3804518381): route the existing-null balance initialization
through a { user, tokenCredits: null } compare-and-set instead of a blind $set,
so a concurrent initializer/charge landing between the preflight read and the
write is never handed back its spent starting balance. On a CAS miss the
preflight re-reads the winner. The absent-record $setOnInsert path and the
credited-record refill-config sync are unchanged. Adds the initializeNullBalance
adapter (no upsert) and regression coverage for winner/miss/sync cases.

Slice 2 (thread r3804518388): updateScheduleById now drops a `requires_action`
lastRun projection atomically with the configRevision bump. Any pause present at
edit time was projected under the pre-edit revision and can never be replaced by
its own revision-fenced terminal outcome — a disabling edit would strand the
card on "Needs approval" forever. Implemented as classic-operator CAS branches
(DocumentDB rules out a conditional pipeline $unset), fenced on the card STILL
being the pause so a terminal outcome or newer occurrence that races in is
preserved. Terminal history survives untouched.

Slice 4 (thread r3803826204): expose initializeScheduleErasureSweep from the
schedule runtime facade and start it in every clustered (experimental) worker
after Mongo is up. It re-drives eraseScheduleIfDrained for soft-deleted rows so
a hidden prompt cannot outlive its drain when the delete/erase-on-settle
attempts miss. It arms nothing else and never infers owner death from a
process-local missing job (isTopologySafeToArm gates that).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

* fix(schedules): Wave B.1 — reversible account-deletion schedule suspension

Thread r3804518383. Account-deletion quiesce marked every schedule `deleting`,
disabled it, and cleared nextRunAt destructively. When a later cascade step (or
the drain itself) failed, the controller cancelled the user-deletion fence —
restoring the user — but the schedules stayed `deleting` and were erased by the
sweep, silently losing all of a live user's scheduled prompts.

Replace the destructive marking with a REVERSIBLE, token-fenced suspension:

- suspendUserSchedulesForDeletion(userId, token) snapshots each schedule's prior
  enabled/nextRunAt under a per-attempt token, then fences firing (disable, clear
  nextRunAt, rotate claimToken). It never sets `deleting`, so a suspended row is
  not erasure-eligible. Snapshotting reads then bulkWrites (a classic update
  cannot copy field values under DocumentDB), fenced per row so an already-
  suspended/soft-deleted/edited row is left alone; idempotent per token.
- restoreUserSchedulesFromDeletion(userId, token) reverses it, re-enabling and
  re-arming only rows still carrying the exact attempt token and not independently
  deleted — so an owner-deleted or newer-attempt-suspended schedule is never
  resurrected.
- deleteUserController generates the attempt token, passes it to quiesce, and on
  any failure that cancels the user-deletion fence restores the suspended rows. A
  successful deletion hard-deletes them (and their snapshots) via the existing
  cascade and never restores.

Adds a `deletionSuspension` embedded field (excluded from the wire schedule),
data-method regression tests (suspend/restore/fence/idempotency/no-resurrect),
and controller tests (restore on drain-false and post-quiesce cascade failure,
no restore on success).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

* fix(schedules): Wave B.2 — wire the interactive Stop persistence protocol

Thread r3804255932. The durable Stop primitives (requestRunAbort 'stop',
getScheduleRunAbortState, markRunAbortPersisted) already existed but production
only used requestRunAbort(..., 'deletion'). An interactive Stop flipped the job
to `aborted` and then persisted its partial message + checkpoint inside
`beforePublish`, without ever stamping the schedule Stop or acknowledging it —
so reconciliation, the generation owner, or a concurrent schedule/account
deletion could terminalize the run and release its capacity (and erase data)
mid-write.

Wire the request -> persist -> acknowledge -> settle barrier through the
schedule runtime (the route never touches raw Mongo):

- Expose beginScheduledStop / acknowledgeScheduledStopPersistence on the service.
- The abort route stamps the Stop BEFORE signalling abortJob (a serialized
  'in_progress' loser returns 409 STOP_IN_PROGRESS without a second abort),
  acknowledges only after beforePublish persistence succeeds, and on a
  persistence failure leaves the barrier unresolved so the run stays preserved
  (client retries; stale-owner timeout is the bounded recovery). A failed abort
  releases the stamp it placed so a replacement/retry is never blocked through
  its predecessor.
- recordScheduleOutcome (the owner settlement path) now waits, bounded, for the
  Stop acknowledgement before terminalizing; a resolved/non-stop/stale marker
  proceeds immediately. The paused Stop settles only after its own ack.

Adds service-layer barrier tests and route-level ordering/persistence-failure/
in-progress tests; the data-layer serialization is already covered.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

* fix(schedules): Wave C.1 — reconcile durable trigger delivery with the reservation

Threads r3803826192 (manual limiter) and r3804255924 (PII/moderation), plus the
independently-found long-Retry-After race. fireSchedule reserves a `started` run
and a global capacity slot BEFORE the durable trigger delivery reaches the chat
route, where an interactive limiter (manual Run Now), PII, or moderation can
reject it before any generation job exists — dead-lettering the delivery while
the run sat `started` until the 30-minute orphan sweep mislabeled it interrupted.
And a valid delivery deferred by Retry-After (up to 24h) could be orphan-settled
and have its capacity released, then fire anyway.

Translate durable delivery state into the schedule outcome:

- Store the deterministic trigger deliveryKey on the ScheduleRun reservation
  (computed from the envelope BEFORE enqueue, so an ambiguous commit still has it).
- Add a getTriggerDelivery engine dep (wired to the merged trigger service's
  getDelivery) that reads the durable delivery by key.
- Schedule reconciliation, for a jobless `started` run: staging/pending/leased →
  admission is live, never orphan; dead → record `error` from the durable
  lastError and release capacity promptly (no 30-minute wait), through the
  ordinary outcome/auto-disable path; succeeded or no record → the existing
  legacy orphan policy (interrupted only past the cutoff); a delivery lookup
  failure defers rather than orphaning a possibly-live delivery. Limiter/PII/
  moderation middleware writes no schedule state.

Adds reconcile state-mapping tests (dead/pending/leased/staging/succeeded/none/
lookup-failure) and a fire test that the reservation's deliveryKey equals the
enqueued delivery's idempotency key.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

* fix(stream): Wave C.2 — durable retry/ack for terminal host lifecycle actions

Thread r3804518375. Approval expiry won the `requires_action → aborted` CAS and
then invoked the host hook best-effort: `runApprovalExpiredHandler` swallowed a
failure, and because later sweeps enumerate only `requires_action` jobs, the now-
aborted job was never offered again. In the clustered entrypoint (no schedule
reconciler) the ScheduleRun stayed `requires_action` and its retained job
persisted indefinitely.

Make the host lifecycle work durable rather than schedule-specific:

- Add a generic `terminalHostActionPending` marker, set ATOMICALLY in the same
  terminal transition (ApprovalLifecycle.expireWithIdentity), only when a host
  adapter is installed.
- Retain and index such jobs: both stores keep them out of terminal reaping and
  expose getTerminalHostActionJobs(); Redis adds a set + extended (24h-bounded)
  TTL, in-memory a bounded 24h retention so a permanently-failing hook cannot leak.
- The manager clears the marker only after the adapter acknowledges success,
  fenced by generation identity (clearTerminalHostAction), so a replacement
  generation can neither clear nor execute its predecessor's action.
- cleanup()/expireStaleApprovals() enumerates unacknowledged terminal host actions
  across restarts and replicas and retries the idempotent hook; the relay only
  re-invokes while the marker is unacknowledged, so a successful ack prevents
  duplicate work. Store-won expiry marks it too, so a loser-replica relay still
  crosses the hook.
- Terminal SSE notification continues regardless of host-hook outcome.

Covers in-memory behavior (retry after failure, restart/other-replica retry, ack
prevents duplicates, identity fence, terminal notification on failure, no marker
accumulation for non-scheduled jobs) and updates the Redis cluster-membership
contract test for the new index.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

* style(schedules): satisfy import sorting in fire.ts and fire.spec.ts

CI "Static checks" failed on IMPORT_SORT for the two files Wave C.1 added imports
to (the AgentTriggerEnvelope type import and getAgentTriggerIdempotencyKey).
Applied scripts/sort-imports.mts to exactly those files — imports-only reordering,
no behavior change. Other files reported by a repo-wide check are pre-existing on
dev and deliberately left untouched so this PR is not widened.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

* fix(schedules): repair the deletion CLI and order restore before the fence release

Addresses three findings from the fresh Codex review.

P1 — config/delete-user.js called methods.disableUserSchedulesForDeletion, which
Wave B removed in favor of suspendUserSchedulesForDeletion. The file is
@ts-nocheck and its spec mocked the removed name, so neither typecheck nor tests
caught it; the real CLI would throw a TypeError before deleting anything and then
only unwind the fence. The CLI now uses the tokenized protocol: it mints a
suspension token, suspends with it, and restores that exact attempt's rows in its
finally block when the deletion does not commit. Its spec mocks the real methods,
so the breakage can no longer hide.

P2 — both the HTTP controller and the CLI released the user-deletion fence BEFORE
restoring schedules. That fence is what refuses new schedule writes/claims, so the
gap let an owner PATCH edit a still-suspended row and have its enabled/next-run
state overwritten by the older snapshot, and let a second deletion attempt
re-suspend under a new token — making the first restore a no-op and stranding the
disabled snapshot permanently. Restore now runs first, while writes are still
fenced.

Hardening for the same defect class across a crash: suspendUserSchedulesForDeletion
now ADOPTS an existing suspension's snapshot when re-suspending a row abandoned by
an earlier attempt, instead of re-capturing the row's current (already-suspended)
state. Without this, an attempt that died before restoring would have its
successor snapshot "disabled, no next run" and permanently strand the schedule.

Tests: CLI restore-before-fence ordering and no-restore-on-success; the same
ordering assertion on both controller post-quiesce failure paths; a data-method
regression that a second attempt adopts the abandoned snapshot and restores the
original enabled/next-run state.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

* fix(schedules): converge dead deliveries in topology-safe maintenance

Codex finding: the `dead` delivery mapping added in Wave C.1 lives only inside
startScheduleEngine's reconciler, but the clustered entrypoint arms no engine — it
runs erasure-only maintenance. A delivery queued before a restart into clustered
mode and then rejected before generation creation (interactive limiter, PII,
moderation) dead-letters while its ScheduleRun stays `started`, holding a global
capacity slot indefinitely for an ordinary non-deleting schedule.

Add a dead-delivery convergence pass to the erasure sweep, so every topology that
runs schedule maintenance settles it. The pass is POSITIVE-EVIDENCE-ONLY and is
therefore safe where absence-based reconciliation is not: a `dead` delivery is
durable shared state proving no generation owns the reservation. It settles only
when the job is confirmed absent or identity-mismatched (an identity-matched job
still owns the run), defers on an unknown job lookup, on an in-flight abort, and
on an in-flight resume hand-off, ignores legacy reservations with no deliveryKey,
and applies a short grace so an accepted delivery still creating its generation is
never settled mid-handoff. Auto-disable policy is deliberately left to the armed
engine; this path records the failure and frees the slot.

Deliberately does NOT touch api/server/experimental.js — the clustered entrypoint
already starts this sweep, so the convergence arrives through the existing
initializer and the shared-file footprint stays as-is.

Tests: settles a dead delivery as error under an explicitly UNSAFE topology,
leaves live deliveries alone, never settles under an identity-matched running
generation (delivery is not even consulted), defers an in-flight abort, and
ignores a reservation with no deliveryKey.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

* fix(stream): refresh host-action retention on each retry attempt

Codex finding: unacknowledged terminal host-action evidence was capped at the
24h pause TTL, so a host dependency (Mongo) unreachable for longer than that let
the Redis key — and its pending marker — expire with no generation-fenced
acknowledgement, stranding the ScheduleRun where no reconciler is armed.

Measure retention from the LAST retry rather than from the terminal transition:
enumerating a pending host action IS the retry attempt, so both stores refresh
its retention as they hand it to the hook (Redis re-EXPIREs the job key; the
in-memory store stamps terminalHostActionRefreshedAt and bounds from it). Evidence
therefore survives as long as some replica is still actively retrying, while a
deployment that stops sweeping entirely still lets it age out — so this does not
reintroduce the unbounded leak the cap existed to prevent.

Test: after a failed hook, a later cleanup pass keeps the marker pending and moves
its retention basis forward.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

* fix(schedules): defer settlement when the Stop barrier times out

Codex finding: waitForStopPersistence returned after its 5s poll budget even when
the Stop was still fresh and unacknowledged, and recordScheduleOutcome then went
straight on to terminalize the run — releasing its capacity, deletion, and erasure
barriers while beforePublish may still have been writing. Slow checkpoint cleanup
is indistinguishable from a dead route on that signal, so the timeout was being
treated as if the barrier had been satisfied.

The poll budget now means "undecided", not "clear". On timeout with a fresh,
unacknowledged Stop the barrier DEFERS: recordScheduleOutcome returns false
without recording, leaving the run active/preserved. Settlement then happens
either when the route acknowledges, or once the existing stale-owner cutoff
(ABORT_OWNER_PRESUMED_ALIVE_MS) authorizes a later attempt — which the loop
already treats as clear-to-settle. Callers with durable retry (the approval-expiry
host action, reconciliation) re-drive it, so a deferral converges rather than
stranding the run.

Test: a fresh Stop that never acknowledges within the budget reports not-settled
and records no outcome.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

* fix(schedules): require definite delivery failure, extract its message, converge deferred Stops

Third Codex round. Three findings in code from this closeout, plus one pre-existing
P1 that is a one-line operator-safety fix.

P1 — dead-delivery settlement demanded too little evidence. `dead` does not prove a
request was rejected: the trigger host marks response timeouts and invalid success
responses `certainty: 'ambiguous'`, and the engine dead-letters those once retries
are exhausted. The erasure sweep treated every dead letter as positive evidence, so
an ambiguous one sitting over a generation a peer had accepted could terminalize the
run and release its capacity mid-flight. It now settles only on a DEFINITE rejection,
unless job absence is deployment-authoritative (safe topology), where the
confirmed-absent job is itself the evidence.

P1 — `lastError` is an `AgentTriggerDeliveryFailure` object, not a string. A
duplicated local interface declared it `string` (against CLAUDE.md's no-duplicate-
types rule), so both the sweep and the engine reconciler passed the object into the
String-typed run/schedule `error` fields; Mongoose would reject the cast, the per-row
catch would swallow it, and the run would keep its global capacity slot. The dep type
now reuses the canonical `AgentTriggerDeliveryFailure` and both call sites pass
`.message`. Re-typing immediately surfaced a stale test that had asserted a string.

P1 (pre-existing) — the base-config global stop is honored in `getLimits` via
`isRuntimeDisabled` rather than a literal `=== false`. The stop has two shapes, and
deepMerge turns base `{ use: false }` plus a principal override of `true` into
`{ use: true }`, so the literal check reported the feature enabled and Run Now
dispatched straight through fireSchedule, bypassing the operator's emergency stop.
Now the same predicate the engine gate already uses.

P2 — a Stop whose settlement DEFERRED past the poll budget had no convergence path
where no reconciler is armed. `acknowledgeScheduledStopPersistence` now optionally
re-drives the terminal outcome once the barrier clears; `recordRunOutcome` is
match-guarded and idempotent, so an owner that already settled makes it a no-op. The
abort route passes it for a running generation; a paused job still settles explicitly.

Tests: ambiguous dead letters refused under unsafe topology but settled when absence
is authoritative, definite rejections settled either way, the failure message carried
through, and the abort route's re-drive present for running / absent for paused.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

* fix(schedules): converge terminal runs in clustered workers, retry suspension restore

Fourth Codex round on 74f7af8. Both findings are in code from this closeout.

P2 - a clustered worker never settled a run whose generation finished but whose
outcome write failed. `recordScheduleOutcome` retries three times and then returns
false; the owner honors that by PRESERVING the terminal job as the only surviving
evidence, and the armed engine's reconciler replays exactly that. The clustered
entrypoint arms no engine, and its sweep only covered deleting schedules
(settleAbandonedRuns) and dead deliveries - an identity-matched job was skipped
outright. So an ordinary live schedule's run stayed `started`, held its GLOBAL
capacity slot, and kept a preserved job that carries no `completedAt` and is
therefore invisible to the store's finished-job sweep: both leaked until store
expiry.

The live-schedule pass now also converges from a retained terminal job, mirroring
the reconciler branch it stands in for: honor the owner's stamped outcome over the
generic status (so a balance refusal still walks its streak rather than resetting
it), clear the reserved conversationId when the generation never emitted its
created event, and delete the retained job only AFTER the outcome write is durable.

This stays positive-evidence-only and safe in every topology. Presence, not
absence, is the evidence: an identity-matched job is authoritative wherever it is
observed - a shared store shows the real generation, a process-local store can
only be showing this process's own - which is why it needs no
canInferOwnerDeathFromMissingJob fence, unlike the absence-based paths. The
in-flight abort/resume fences still defer, so an `aborted` job cannot settle a run
whose owner may still be persisting. Both cases now share one pass over one window
rather than two, and `retainedOutcome` moved to types.ts so the sweep reuses the
engine's mapping instead of duplicating it (and stays independent of the engine).

P2 - a failed restore stranded a live account's schedules. Cancelling an account
deletion restores the suspended rows while the deletion fence is still armed and
then releases that fence; nothing re-drives the restore afterwards, so one
transient write failure left the user with silently disabled, next-run-less
schedules. The restore is now retried at the single choke point both the HTTP
controller and the CLI share. Retrying is safe because each attempt re-reads only
the rows STILL carrying the token: a partially-applied unordered write converges
on exactly the stragglers, and a fully-applied one finds nothing.

The fence is still released when every attempt fails, deliberately: retaining it
would refuse the live account's schedule writes AND make beginAgentTriggerUserDeletion
report `in_progress` forever, blocking the retry that is the convergence path (a
later attempt adopts this snapshot, so its cancel restores these exact rows). Both
callers now log the user id and suspension token so the state stays recoverable by
hand if that never happens.

Tests: retained terminal job settled and its evidence released, stamped outcome
preferred over the generic status, stamped failure reason carried, reserved
conversationId cleared for a never-created conversation, identity-mismatched
terminal job ignored, abort-in-flight deferred; restore retried past a transient
write failure and converging on a partially applied restore.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

* fix(schedules): converge unprojected pauses and release replayed bookkeeping jobs

Two of the four findings I flagged as pre-existing to my closeout commits. Both are
in code this PR introduces, so merging would have shipped them; both are the same
capacity/evidence-leak class the last four rounds have been closing. The remaining
two (slotless rows escaping the per-user cap, and the non-rotating `started`
reconciliation bucket) are genuinely latent and deliberately left alone.

An UNPROJECTED PAUSE held a global capacity slot forever. The pause projection is
what moves a run row off `started`; `recordScheduleOutcome` already retries it, but
the request controller discarded the result, so three failed attempts were dropped
silently. The armed engine's reconciler replays that state, but the clustered sweep
did not: a paused job is not terminal, so the retained-job path returned without
settling, and the dead-delivery path never inspects an identity-matched job at all.
The row stayed `started` with no cutoff that would ever clear it.

The call site now surfaces the failure, and the sweep converges it, mirroring the
reconciler's pause branch: project `requires_action` (which frees the slot) but do
NOT release the job's evidence — unlike a terminal job it is still live, awaiting an
approval. The resume hand-off fence still defers, so re-projecting cannot release a
slot a continuation just claimed.

The BOOKKEEPING REPLAY pass leaked its retained job. A run reaches that pass only
because its owner crashed before bookkeeping — which is also before it could release
the job it retained for exactly this recovery. The active-run pass clears its own;
this one never did, and a preserved job is deliberately kept WITHOUT `completedAt`
so the store's finished-job sweep cannot reap it early, so nothing else ever would.
It now clears after `finalizeBookkeeping` succeeds — identity-guarded, a no-op when
no job is retained, and skipped entirely when the replay itself failed, since the
retained job is the only surviving evidence in that case.

Tests: pause projected and the slot freed while the live job's evidence is kept,
paused job deferred during a resume hand-off, retained job released once replayed
bookkeeping is durable, and retained job kept when the replay fails.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

* fix(schedules): fence the clustered pause replay against a concurrent resume claim

Codex round 5, on code I added in f735341. The finding is correct and the exposure
is one I introduced.

The pause replay lives in a sweep that runs in EVERY clustered replica, so several
sweepers can observe the same unprojected pause. Its guards were all derived from an
in-memory row SNAPSHOT: hasResumeHandoffInFlight reads the snapshot's
resumeClaimedAt, and recordRunOutcome's pause branch matched any row currently in
`started`/`requires_action` with no fence of its own. The race: sweeper A projects
the pause and frees the slot; the owner's approval then claims a fresh one
(markRunResumeClaimed takes the row to `started` WITH resumeClaimedAt in one write);
sweeper B, still holding the pre-projection snapshot, passes its hand-off check and
replays — `$unset: { capacitySlot, resumeClaimedAt }` under a continuation that is
already running. The run reverts to `requires_action` while its generation proceeds
outside global capacity.

The engine's reconciler makes the same call and has the same snapshot-derived guard,
but v1 arms exactly one engine, so it has no concurrent racer; the sweep is the first
thing to run this transition in parallel. Left the engine alone rather than widening
the change: its `requires_action` re-affirmation is deliberate and single-writer.

Fixed where the race is, in the write itself. `recordRunOutcome` takes an optional
`requireNoResumeClaim`, which adds `resumeClaimedAt: { $exists: false }` to the pause
filter, and the sweep sets it. Because the stamp is written in the SAME update that
moves the row to `started`, its absence is atomic proof no resume owns the row. The
flag is deliberately NOT set by the generation owner: its own re-pause legitimately
clears the stamp as the hand-off's completion signal.

The fence cannot block the recovery it exists to enable: markRunResumeClaimed only
matches `requires_action`, so a genuinely stuck `started` row can carry no resume
claim — it is in fact blocking its own approval until this replay frees it.

Tests: a replay from a stale snapshot leaves a resume-claimed row's status, slot, and
claim stamp intact, while a stuck row with no claim still recovers; and the sweep is
asserted to send the fence.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

* fix(schedules): fence stale resume claims by age, not existence

Codex round 6, on the fence I added in ee00c60. Correct again, and the defect is
the mirror image of the one it fixed.

The fence was an EXISTENCE check (`resumeClaimedAt: { $exists: false }`) while its
caller's guard is a FRESHNESS check (hasResumeHandoffInFlight, bounded by
RESUME_HANDOFF_STALE_MS). They agree while a claim is fresh and disagree once it is
abandoned: a worker that dies after markRunResumeClaimed takes the row to `started`
and stamps resumeClaimedAt — but before the continuation resumes or
releaseRunResumeClaim rolls it back — leaves the stamp set forever. Past the bound
the sweep correctly stops deferring and tries to recover the row, but the write
rejected it purely because the field still existed. The row stayed `started` holding
its global capacity slot, and its approval was unresumable for good, since
markRunResumeClaimed only matches `requires_action`. That is exactly the stuck state
this replay exists to clear, so the fence had reintroduced it for the crashed-resume
case.

`requireNoResumeClaim: boolean` becomes `resumeClaimStaleBefore: Date`, and the
filter matches a row with no claim OR a claim older than that cutoff. The sweep
passes the SAME bound its in-flight check uses, so the two can no longer disagree.
A genuinely racing claim is by construction fresh — it is created after the sweeper's
snapshot — so the race from round 5 stays closed.

Tests: a row whose resume claim was abandoned by a dead worker now recovers (status,
slot and stamp all cleared), alongside the existing two — a fresh claim still repels
a stale-snapshot replay, and an unclaimed stuck row still recovers.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CZwZPDBgxy4mCWWg6zQSjq

---------

Co-authored-by: Claude <noreply@anthropic.com>
2026-08-20 11:51:30 -04:00