deleteCodeEnvFile rethrew every error except 404. A deployment that only serves
the stateful profile answers a default-profile delete with 409
execution_profile_mismatch on every retry, so the file landed in failedFileIds,
its record and agent references stayed, and agents holding it aborted every code
turn with "Failed to provision N file(s) to the code environment". The mismatch
is now handled like an absent object: logged, and the delete proceeds. Other
409s, 5xx, auth and network errors still fail the delete.
* feat: Add Lia's Behavior Engine
A framework-free engine that plays Lia's actions step by step, moves her along a platform, and lets her choose her own activities. An attention budget keeps her quiet while the user types and saves bigger routines for when they step away. The catalog holds feelings, speech, gestures, dances, travel, small routines and page reactions.
* fix: Let a Click Wake Lia From a Nap and Run One Animation Loop
* fix: Keep Lia Stopped When Stopped Mid-Frame and Snap Walks Under Reduced Motion
* fix: Wake Lia When Typing Interrupts a Nap and Keep Her Walks on the Current Platform
* test: Drive Lia's Engine on Real Animation Frames in a Browser
* fix: Keep One Animation Loop Across a Mid-Frame Restart and Place Lia Before Her First Move
* fix: Play Every Step of an Action the Host Chains From onAction
* test: Require Lia's Reduced-Motion Chooser to Pick Something
* fix: Keep Lia Free to Act Before the First Keystroke and Skip Pointer Walks Without a Pointer
* fix: Hold Lia's Step While She Has No Layout and Replace Actions Without a Gap
* fix: Keep the Walk of an Action Started When the Previous One Ends
* fix: Wake Lia From a Nap Played Above Reaction Priority
* fix: Pause Lia's Screen and Face Clocks Along With Her Step
* fix: Stop Lia's Own Clock While She Has No Layout and Ignore Empty Obstacles
* 🪜 feat: Light the Fold Path to the Hovered Activity Row
- Every fold that holds the hovered row lights its rail from its header down to that row; the innermost turns into the row with an elbow, so depth reads as the number of lit rails.
- One pointer listener on the outermost fold writes the path onto the rails directly (a data attribute and a length variable), coalesced to one frame: hovering repaints rails, never rows.
- Rows are found by the shared glyph slot every row line starts with, so no row component marks itself.
- The path clears over a rail (which shows its own collapse hover), on touch, on a press and when the pointer leaves.
* fix: Invalidate Fold Hover Paths and Reuse Glyph Lists
* fix: Scope Fold Paths to Agent Lanes and Register Subagent Glyphs
* fix: Re-Hit-Test Queued Fold Paths After Scroll
* fix: Clear Fold Paths After Ancestor Layout Shifts
---------
Co-authored-by: Lia <lia@librechat.ai>
* 🔭 feat: Recover Stale Chunks Without the Error Page and Ship Client Logs to RUM
Stale chunks: a tab left open across a deploy that opens a code-split panel
requests a hashed chunk the new deploy removed. Vite's preload handler hands
the failure to the stale-asset recovery, which reloads, but the import then
resolves to `undefined` and `React.lazy` throws a TypeError into the route
error boundary, so the error page flashes before the reload. Lazy imports now
go through `lazyWithRecovery`/`importWithRecovery`, which turn that result into
a `ChunkLoadError`; the route boundary recognizes chunk failures (and any error
while a recovery reload is underway), asks the page recovery once, and renders
a localized "Updating to the latest version…" status. When the loop guard
declines, the normal error page with its reload action shows instead. The
page recovery is now idempotent within a page, so concurrent callers join the
in-flight reload rather than tripping the guard.
Client logs: with RUM in proxy mode, the client logger's warn/error calls,
error-boundary catches and stale-asset recovery events are exported as OTLP
log records through the authenticated `/api/rum/v1/logs` proxy. Records carry
allowlisted, scrubbed fields only, follow RUM session sampling, dedupe repeats
with a count, are budgeted per minute and per page, batch with backoff, flush
with keepalive on page hide, and stop after persistent failures. The proxy
now refuses non-OTLP content types and applies a per-user rate limit.
`RUM_CLIENT_LOGS=false` turns the export off.
* 🔒 fix: Keep Client Logs Opt-In and Drop Application Error Messages
- RUM_CLIENT_LOGS now defaults to off, like console capture, so enabling
RUM proxy mode never starts a new log stream on its own.
- Error messages are exported only for engine, browser and library
generated types (TypeError, SyntaxError, DOMException, AxiosError,
chunk failures); application errors keep type, status and stack only,
since their messages can echo prompt or response text. All quoted
fragments are now redacted.
- Batches are measured in encoded UTF-8 bytes of the exact request body,
and page hide drains every queued record that fits the keepalive quota.
- Queued stale-asset events are delivered as soon as the log exporter
starts, independent of the RUM SDK loading, and marked so they are
never logged twice.
* 🔒 fix: Export Only Known Generated Error Messages and Ack Queued Events
- Error messages are exported only when the whole text matches a known
engine, browser or library message shape (property access on
undefined, "x is not a function", fetch/abort/axios status, chunk and
CSS preload failures). The error class alone no longer qualifies, since
any code can throw a TypeError carrying arbitrary text.
- Queued stale-asset events are marked logged only after the collector
accepts the record carrying them, so an undelivered event replays on the
next page; in-flight events are skipped by the SDK path.
- A fractional RUM_PROXY_USER_MAX below one falls back to the default
instead of flooring to a limit of zero.
* 🔒 fix: Hold Unacknowledged Asset Events, Re-send In-Flight Logs on Hide
- The RUM SDK flush no longer drops queued stale-asset events whose client
log record is still unacknowledged: they stay queued and persisted,
flagged so their SDK action is not repeated, and leave the queue once
the collector accepts the record.
- Page hide re-sends the records of an unanswered regular request with
keepalive; each record belongs to the latest request carrying it, so a
late response to the original request neither marks nor requeues it.
- Client-log export stops when the authenticated layout unmounts.
- The proxy refuses /v1/logs while client logs, console capture and replay
are all off, so turning RUM_CLIENT_LOGS off also stops tabs that still
hold an older startup config.
* 🔒 fix: Export Only Complete Stack Frames, Never the Message Header
Stack reduction now strips the V8 `name: message` header (including every
line of a multi-line message) and keeps only lines with a complete V8,
Gecko or WebKit frame shape that ends in a URL or absolute path with a
line number. A message ending in a source-position-like suffix can no
longer pass as a frame.
* fix: centralize RUM delivery and asset recovery lifecycle
* fix: preserve RUM page lifetime and collector acknowledgements
* fix: retain page transport quotas and recover functional chunks
* fix: scope RUM client sessions to authenticated identity
---------
Co-authored-by: Lia <lia@librechat.ai>
* fix: Separate workspace admission and transport budgets
* fix: Validate workspace run reserves and retry test types
---------
Co-authored-by: Lia <lia@librechat.ai>
Each mirrored skill now records a sync fingerprint in its sourceMetadata, hashing the mapping version, the upstream identity (source coordinates, skill path and SKILL.md blob id) and the stored definition. When a later run finds the fingerprint still matching and the stored files matching the discovered blobs, it neither downloads SKILL.md nor re-reads and updates the skill; the public viewer grant, name-conflict cleanup, stale-skill reconciliation and status counts run as before. Mirrors without a fingerprint take the full path once and record it.
- Running chats ring their avatar instead of a trailing spinner, so the row menu stays reachable mid-run.
- A reply that lands unseen lifts its chat into a Finished group under Running until it is opened; chats marked unread by hand keep their date.
- Running and Finished headings show their counts and announce them.
- Project chats now list under Chats with their folder badge; a persisted Show project chats toggle in the chat list menu restores the project-only view.
- Badges read project names from the loaded project list before fetching one by id.
- Spinner takes a strokeWidth for the thin ring.
* feat: Add Lia's Pixel Art Renderer
Lia is a small CRT monitor mascot with a face on its screen. This adds the canvas renderer: the body, feet and arm poses, body poses, 62 expressions built from eyes, mouths, brows and extras, and the built-in screens for booting, sending and crashing. Nothing renders it yet.
* test: Assert Each Built-In Lia Screen Draws Its Own Picture
* test: Include the Cached Face Image in Lia's Drawing Signatures
* fix: Clip Lia's Glitch Strips to Her Screen and End the Hide Pose Fully Hidden
* test: Check Lia's Pixel Art in a Real Browser Canvas
* test: Share the Page That Serves Lia's Engine to Browser Specs
* test: Bound-Check Where Lia's Images Land, Not Just Her Rectangles
* test: Check Lia's Face Bounds for Every Gaze Direction
* fix: Drop Lia's Sunglasses In From the Top Row of Her Screen
* test: Check Where Lia's Glitch Strips Are Copied From, Not Only Where They Land
* perf: Batch Prompt Group Permissions in the Sidebar
* perf: Skip the Prompt Permission Batch When the List Is Empty
Add scenario coverage for batched prompt row permissions.
* test: Retry Opening Prompt Row Menus Until Their Items Render
* fix: Warn When interface.webSearch Blocks Native Search in Model Specs
interface.webSearch syncs the WEB_SEARCH permission, which also gates
provider-native search since #15875. A deployment that set it to false to
hide LibreChat's search tool silently lost native search on its model
specs. Startup now names the affected specs, describes the native-only
configuration, and notes that removing the key keeps the stored permission.
Related to #16673
* fix: Warn When a Stored Role Denies Native Search With interface.webSearch Unset
* perf: Trim Per-Token and Per-Render Work in Streaming and Lists
* test: Assert Rendered Diff Rows Alongside Render Counts
* refactor: Read the Active Sidebar Row Through the Existing Index Selector
Drops the new Recoil selector: conversationIdByIndex(0) already yields a primitive id, so rows still skip re-renders on every conversation write. The DiffView row test keeps the real cn behind a spy and derives the per-row cost from the first render.
* test: Cover the Sidebar Active Row Across Chat Switches and New Chat
The Settings dialog is a Headless UI panel at z-50, not an OGDialog, so it set no dialog depth and its Select lists kept their z-40 layer, opening invisibly behind the panel. A new DialogLayer counts it as one dialog level, so popovers inside it take the depth-aware z-index they get inside an OGDialog.
* feat: Size Roles for the Select List
The Select list drew a fixed max-h-96 and min-w-[8rem], the last hard-coded sizes among the ten shared primitives. They are now the listMaxHeight and listMinWidth appearance roles, defaulting to 24rem and 8rem. The ClickHouse theme sizes the list by its trigger and keeps the 24rem cap, since Click UI caps it only at the viewport's available height.
* test: Resize the Select List From a Reference Theme
* fix: persist approval provenance in the resume claim and publish the pre-claim paths from any unapplied abort
The claim CAS now records the provenance it replaces. An abort that lands
before the resumed provider crosses its start CAS publishes those paths, so
schedule deletion, account deletion and user stops no longer label
model-authored chunks as user-submitted.
* test: add approval provenance scenario on the shared approval e2e helpers
* test: seed an earlier started segment before the unapplied abort case
* fix: label abort provenance by the published content and persist the same selection
A path the approval claim added is user-authored only once its tool call
carries output in the abort content, so an abort between the provider
start CAS and the resumed tool keeps the pre-claim labels, and an ask
answer stamped by the abort route keeps its own. AbortResult.jobData
carries the selection so the persisted row matches the final event.
* fix: keep claimed non-tool provenance paths on an abort
Only tool-call paths wait for their resumed call to complete; a steer the
claim already found in the seed content needs no decision to apply.
* fix: count a completed tool call with an empty output as applied in abort provenance
* fix: check abort provenance on the unfiltered content and remap it onto the published parts
* test: type the abort provenance overrides with the job store field names
* test: assert the preserved paths when an unanswered ask aborts
* refactor: Move Select and Menu Pointer Events to Classes
DropdownPopup and SelectContent re-enabled pointer events inside modal dialogs through inline style, which kept a no-inline-styles suppression on both primitives. Both now apply pointer-events-auto as a class and keep only the runtime z-index inline. The Select item indicator box takes the icon size role instead of a fixed h-3.5 w-3.5.
* test: Cover Popover Picks Inside Modal Dialogs
* test: Open Settings From the Drawer and Tag Indicator Scenarios
* test: Let Each Verify Project Set the Color Mode
* test: Measure the Select Indicator After the List Settles
* test: Drop the Settings Select Pick That Never Reached a Dialog
* fix: Read Theme Roles for the Last Artifact and Agent Chip Colours
The artifact mobile backdrop now reads surface-overlay like the dialog scrims, the refresh overlay over the rendered preview reads surface-media-overlay, and the agent builder's provider chip reads surface-primary and text-primary like the conversation icon chip. Monaco's built-in vs-dark canvas joins the third-party surfaces allowlist entry, and the allowlist no longer points at a closed sweep issue.
* test: Cover Theme Roles on the Provider Chip and Artifact Overlays
* 📍 feat: Show the Current Text Where a Missing Workspace Edit Belongs
Render the current-text excerpt that LibreChat-AI/code-interpreter#300 workers append to a missing edit's EDIT_CONFLICT diagnosis: a strict parse keeps it only when every row is well formed (contiguous numbers matching the stated range, a known mark, a bounded single line), and LibreChat renders it as a numbered block under its edit with a summary of which lines differ. The excerpt passes the file-content policy like a read_file result and reaches only the model; logs keep the excerpt-free message.
Also stop rendering every 409 as a text mismatch: a Code API rejection such as WORKSPACE_QUARANTINED now passes through with its own reason, and the worker's repetitive-candidate refusal is recognized.
* 🔒 fix: Name Other Attached Edit 409s by Code Instead of Forwarding Their Body
A 409 whose code is not EDIT_CONFLICT (a quarantined workspace) now gets a host-worded message naming only its validated code, with recovery guidance for WORKSPACE_QUARANTINED, and logs keep a body of just that code.
* 🧾 fix: Present Quoted Edit Excerpts as Partial Context, Not Text to Copy
A worker excerpt can window a long region and shorten wide lines with an ellipsis, so the guidance now asks the model to correct old_text against it and use read_file for lines it leaves out or shortens, here and in the attached edit_file descriptions.
* 🔐 fix: Quote Edit Excerpts Only Where the Workspace Allows read_file
A workspace that offers edit_file without read_file now never receives the worker's current-text excerpt, so a failing edit cannot read file content the operation policy withholds.
---------
Co-authored-by: Lia <lia@librechat.ai>
* feat: Remember Tool Approvals Per Conversation
Adds an opt-in "Always allow" choice to the tool approval card. When
endpoints.agents.toolApproval.allowAlways is true, the server marks eligible
paused calls with allow_always; approving with scope "session" stores that
exact tool name on the conversation, and later runs in the same conversation
fold the stored names into the policy as exact-name allow entries.
The server enforces every step: the resume route rejects a session scope the
pause did not offer, eligibility is rechecked against the live policy before
storing, the list is written only through an owner-scoped bounded method, and
generic saves and imports cannot set or clear it. Admin deny and ask rules and
programmatic hooks still win, stored names are ignored under dontAsk or when
the flag is off, and native code tools stay under the code approval picker.
The flag defaults to false, so existing deployments keep prompting.
* fix: Share the Effective Approval Policy and Persist Allows After Resume Fences
Offering, resume re-check, scheduled admission and createRun now read one
policy from buildEffectiveToolApprovalPolicy: endpoint rules healed against
MCP alias spellings, then the conversation's remembered tools. A call held by
a healed ask rule is no longer offered "Always allow", and a scheduled fire
whose known tools are all remembered is no longer rejected as pause capable.
The remembered-tools cap and name length cap move to
toolApproval.allowAlwaysMaxTools and allowAlwaysMaxToolNameLength (defaults
64 and 256). The card stops offering the choice for new tools once the cap is
reached instead of silently approving once.
The resume route stores remembered tools only after the schedule and project
context fences pass; a storage failure still approves the call once.
* fix: Offer Always Allow Only When It Will Hold
Offer allow_always only when storing the name would auto-approve the next
identical call on the run's own path: every MCP spelling of the tool passes
the healed static policy and no programmatic or plugin approval hook can
apply to it (the pause payload does not record which hook asked).
Remembered names now expand to every alias spelling of the same tool, and a
deny or ask on any spelling keeps the tool ineligible. A table-driven spec
walks the invariant through the real hook wiring.
* fix: Recheck Approval Hooks Before Remembering a Resumed Approval
The resume-time eligibility recheck now applies the same programmatic and
plugin PreToolUse hook check as marking, with a hook context built from the
resume request and the live plugin hook source. A hook registered after the
pause turns the choice into a one-time approval: nothing is stored and the
resume continues.
Document that toolApproval.allowAlways should be enabled only once every
replica supports it, since older replicas ignore the choice.
* fix: Judge Always Allow Offers Against the Run's Live Aliases and Active Cap
* fix: Charge Only New Tool Names Against the Remembered Approval Cap
* fix: Recheck Remembered Approvals Against the Paused Run's Aliases
* chore: Sort Imports in Agent Run After Rebase
* 📝 docs: Describe Attached-Workspace Lanes, Sandbox Scratch, and Skill Files in Tool Descriptions
- Lane-mode `cwd` now states that any call outside `.worktrees/<name>` (with or
without cwd, and file tools on such paths) is checkout-wide: it waits for
running worktree calls and blocks new ones until it finishes, so checkout-wide
calls are reserved for creating, pruning or removing worktrees.
- `command`/`cwd` state that the command starts in cwd, so the model does not
also cd into it; the lane-mode cd hint says the call ran checkout-wide.
- Workers advertising the native SRT sandbox profile (`anthropic-srt`,
`anthropic-srt:<preset>`) get a bash bullet that / and /tmp are read-only and
scratch files belong in $TMPDIR or the workspace. Carried as
`codeWorkspace.nativeSandbox` into tool definitions and the bash tool.
- The attached skill-aware `read_file` description says skill files are not on
the attached machine and must be written into the workspace to run there.
- Tightened the attached bash description so the native variant stays within
the 1024-character tool description limit.
* 📝 fix: Keep the Skill-Copy Warning on the Command-Only Workspace's Skill read_file
A workspace that offers execute_command but not read_file falls back to the
skill-only read_file definition. It now carries the same sentence as the
attached skill-aware definition, so the model knows bash_tool cannot reach
skill files by path there.
* 📝 fix: Name bash_tool in the Skill-File Warning Only When It Is Registered
The skill-copy sentence on attached read_file definitions now depends on the
attached bash_tool being registered beside it (includeBash and an advertised
execute_command), for the attached skill-aware definition, its custom
line-limit variant, and the skill-only fallback. Workspaces without commands
keep the plain skill description.
Attached workspace file tools now append a short, specific explanation when
the worker rejects a path with 422:
- NOT_FOUND from a current worker, or INVALID_PATH carrying its
"Workspace path does not exist" message through an older Code API, says
the path does not exist: wait for a command still writing it, or list the
directory to find the right path. For create_file it names the missing
parent directory.
- An older worker's bare INVALID_PATH on read_file, create_file or
edit_file, which it also returns for a missing file, lists the possible
causes: not written yet, a directory or symlink, or a missing parent
directory for create_file.
Other rejections are unchanged.
* 🍴 feat: Separate Machine, Workspace, and Git Controls
* fix: Preserve Recorded Checkout Status During Recovery
* fix: Preserve Keyboard Focus Across Machine and Workspace Menus
* fix: Align Linked Worktree Status with Checkout Selection
* fix: Reset Machine Browsing on Keyboard Workspace Opens
---------
Co-authored-by: Lia <lia@librechat.ai>