Commit graph

830 commits

Author SHA1 Message Date
Danny Avila
0d85af0a77
🧫 chore: Keep Jest Specs Out of Tool Discovery (#15766)
* fix: Skip test modules during structured tool discovery

* fix: Keep tool discovery validation in TypeScript
2026-09-08 23:58:04 -04:00
Danny Avila
ec15075dc8
🚚 feat: Route Attachments and Provision Tools Lazily (#15763) 2026-09-08 23:28:52 -04:00
Danny Avila
ab6cde0c79
⏳ fix: Separate Temporary and General Chat Retention (#15734)
* fix: separate temporary and general chat retention

* fix: inherit retention when branching messages

* fix: harden retention lifecycle boundaries

* fix: preserve subagent retention context

* fix: trust stored chat retention for uploads

* perf: overlap generated file retention lookup

* fix: derive retention from trusted chat state

* fix: preserve retention across the full turn lifecycle

* test: accept the conversation retention read adapter

* fix: carry retention deadlines through durable job saves

* fix: resolve retention at cached and new chat admission

* fix: resolve direct tool retention before execution

* fix: preserve retention across assistant writes and direct tools
2026-09-08 11:22:54 -04:00
Danny Avila
968950a4bd
🪫 fix: Gate Programmatic Bash on Stateful Worker Capabilities (#15726)
Some checks are pending
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
* fix: Gate Programmatic Bash by Worker Capabilities 🧩

* fix: Resolve Capability Credentials From Deployment Config

* fix: Require Bash Runtime for Programmatic Tool Calling
2026-09-08 08:23:38 -04:00
Dustin Healy
cdc3339a78
🪬 fix: Fence Agent Checkpoint Ownership and Cleanup Recovery (#15680)
* 🔒 fix: Isolate Checkpoint Cleanup by Generation Ownership

* 🔒 fix: Harden Checkpoint Cleanup Ownership

* 🔒 fix: Retain Checkpoint Cleanup Evidence

* 🔒 fix: Persist checkpoint cleanup receipts

* 🔒 fix: Bound checkpoint cleanup receipt lifecycle

* fix: Bind checkpoint cleanup to durable ownership and deletion intent

* fix: Bind event actor checkpoint storage to authenticated owners

* fix: Resolve exact actor checkpoint ownership before lifecycle cleanup

* fix: Preserve actor storage compatibility with durable ownership and prune work

* fix: Defer checkpoint erasure until conversation deletion commits

* fix: Reclaim orphan actor ownership and batch deletion lookups

* fix: bind actor checkpoint ownership to payload rows

* fix: isolate actor readers and await account persistence

* fix: gate owned suspensions and drain complete account state

* fix: unify checkpoint deletion lifecycle and storage authority

* fix: Replay checkpoint deletion from durable owner state

* fix: retain checkpoint cleanup identity before topology loss

* fix: complete checkpoint catalog capture and retirement lifecycle
2026-09-07 22:04:59 -04:00
Danny Avila
f1bd85498e
🪫 fix: Recharge Direct OpenID MCP Bearer Sessions (#15704)
* fix: recover direct OpenID MCP bearer sessions

* fix: harden direct bearer recovery lifecycle

* fix: close direct bearer lifecycle races

* test: align direct bearer checks with current dev

* fix: unify direct bearer recovery budget

* fix: infer direct bearer recovery from placeholder

* fix: preserve bearer precedence and recovery budget

* fix: propagate bearer recovery across joiners

* fix: preserve direct bearer fallback without session provider

* fix: close direct bearer recovery race windows

* chore: sort bearer recovery test imports

* fix: align bearer recovery with OAuth cleanup lifecycle

* fix: fence request-scoped MCP creation during config mutations

* fix: retain suspended MCP catalog authentication outcomes

* fix: cancel MCP bearer recovery and reuse refreshed credentials

* fix: close MCP bearer recovery lifecycle boundaries

* fix: preserve ordered catalog authentication rejection

* fix: isolate bearer recovery across request and transport boundaries

* fix: preserve shared recovery ownership and credential precedence

* fix: wait for recovery borrowers and validated OpenID publication

* fix: observe the validated browser publication flight

* fix: cancel abandoned OpenID publication followers

* fix: fence refresh grant admission after lease acquisition
2026-09-07 21:05:26 -04:00
Danny Avila
601ed07870
🫧 fix: Retire Cancelled MCP OAuth State (#15702)
* fix: Clean Up MCP OAuth Teardown State

* fix: Harden OAuth Teardown Boundaries

* fix: harden MCP OAuth teardown races

* fix: avoid flow cache dependency cycle

* fix: bind OAuth teardown to credential generations

* test: cover active OAuth server fixtures

* style: sort OAuth cleanup imports

* fix: discard mismatched OAuth token snapshots

* fix: fence OAuth teardown across callback races

* test: align OAuth callback fixtures with server fencing

* test: complete OAuth callback server fixtures

* fix: version OAuth flow settlement

* fix: fence OAuth teardown snapshots

* test: align OAuth teardown fixtures

* test: complete OAuth snapshot fixtures

* fix: snapshot OAuth deletion lifecycle

* test: bind OAuth revocation snapshots

* test: assert per-record OAuth fence

* fix: harden MCP OAuth teardown consistency

* test: expect snapshotted OAuth readers
2026-09-07 15:43:02 -04:00
Danny Avila
29971f188a
🧺 fix: Skip Priming Denied File Search Resources (#15699)
#15691 paired every read of `AgentCapabilities.execute_code` / `file_search`
with its role grant, and skipped `execute_code` from the resend-file priming
when `RUN_CODE` was denied. It left the `file_search` half: the grant was not
available inside `initializeAgent`, so a role denied `FILE_SEARCH` still had
its prior-turn search files re-hydrated on every resend.

Priming is not free. The files are fetched, `updateFilesUsage` bumps their
counters, and `primeResources` builds `tool_resources.file_search` — all for
a tool the loader is about to drop. A deployment with the capability off paid
the same cost.

`InitializeAgentParams.fileSearchAvailable` carries the capability AND the
grant, resolved from the same single `resolveToolRoleGrants` read that already
answers `codeEnvAvailable`, so pairing the second flag costs no extra role
lookup. Every initializer that resolves grants forwards it: the chat path and
its handoff, added-convo and discovery hand-offs, both API routes, and the
OpenAI-compatible embedder route. Absent leaves priming unconditional, so an
embedder that resolves no grant keeps its current behavior.

The loop that selected the resend tool resources becomes
`resolveResendToolResources`, which makes both halves of the gate testable
rather than reachable only through a full agent initialization.

The memory agent is untouched: it passes no `conversationId`, so it never
reaches this priming.

Co-authored-by: Claude <noreply@anthropic.com>
2026-09-07 07:56:23 -04:00
Danny Avila
a3fc3b9dbf
🚧 fix: Gate File Search and Run Code by Role (#15691)
* fix: enforce FILE_SEARCH role permission server-side

The FILE_SEARCH role permission is stored, served by GET /api/roles/:name and
settable through the admin API, but nothing ever checks it.
PermissionTypes.FILE_SEARCH occurs zero times under api/server; the three
occurrences in @librechat/api are all in the interface-to-role sync, which
writes the permission rather than checking it.

Measured on v0.8.8-rc1 and confirmed unchanged in rc2: a user whose role has
FILE_SEARCH.USE = false can still upload a document with
tool_resource=file_search and gets 200 with "embedded": true. The same user
calling execute_code correctly gets 403.

Two gates, both mirroring toolAccessPermType in ~/server/controllers/tools.js:

* Upload (api/server/routes/files/files.js): a tool-resource-to-permission map
  checked before any branching, so it also covers the assistants path. Returns
  403 with the same log line as the RUN_CODE gate. execute_code is included
  because it had the same hole: the tool CALL was gated, the upload was not.

* Tool loading (api/app/clients/tools/util/handleTools.js): checked outside the
  lazy loader so a denied user never gets the tool equipped, rather than one
  that fails when called. Fails closed if the permission check itself throws.
  All four required imports were already present in that file, next to an
  existing FILE_CITATIONS check.

Both changes live in /api (JavaScript) rather than /packages/api (TypeScript)
as CLAUDE.md prefers for new backend code: each mirrors an existing pattern in
exactly these files, and a second permission gate in another language and
package would be harder to keep in step with the first.

Tests: api/app/clients/tools/util/handleTools.fileSearchPermission.test.js
covers permit, deny, the log line, and the fail-closed path.

* fixup! fix: enforce FILE_SEARCH role permission server-side

* 🔐 fix: Gate `file_search` and `execute_code` on Role Permissions

The `FILE_SEARCH` gate added in the previous commit lives in
`handleTools.loadTools`, which `loadAgentTools` only reaches when
`definitionsOnly` is false. That default is false on the chat path
(`Endpoints/agents/initialize.js`) but true on the responses and
OpenAI-compatible controllers, so on those paths a denied user still got
the tool definition advertised to the model and their files primed by
`primeSearchFiles`, and the call failed later in `loadToolsForExecution` —
the "gets one that fails when called" outcome the gate set out to avoid.

Move the decision to the capability filters that both loaders already run,
mirroring how `canUseMCP` resolves the `MCP_SERVERS` permission ahead of the
synchronous filter. `AgentCapabilities` stays the instance-wide deployment
switch; the role grant is a second condition a tool has to clear. Because
`hasFileSearch`/`hasExecuteCode` derive from the filtered list, priming is
skipped for a denied user too.

`RUN_CODE` had the same hole: `toolAccessPermType` in
`~/server/controllers/tools.js` gates only `POST /tools/:toolId/call`, not
the agent run, so a role without `RUN_CODE.USE` could still execute code
through an agent. Both tools go through the same map.

Checks use `checkAccessWithRequestCache` and run only for tools whose
capability is already enabled, so a run costs at most one role read. A check
that throws denies the tool.

Also formats the test file added in the previous commit, which Prettier
rejects as-is.

* 🔐 fix: Close the Remaining `RUN_CODE` and `FILE_SEARCH` Bypasses

Codex review of the previous commit found four ways past the gates. All four
share a cause: the permission map was copied per call site, so every boundary
that was not copied into stayed open.

Moves the maps and the check into `packages/api/src/tools/rolePermissions.ts`,
per CLAUDE.md ("all new backend code must be TypeScript in `/packages/api`",
"keep `/api` changes to the absolute minimum"), and points the four `/api`
boundaries at it.

- `handleTools.loadTools` now gates every tool in `toolRolePermissions`, not
  just `file_search`. It is the shared boundary the Assistants required-action
  flow crosses via `processRequiredActions`, which never passes the agent
  capability filter — so a legacy Assistant with a raw function named
  `execute_code` ran arbitrary code for a role denied `RUN_CODE`.

- The check is now request-cached (`checkAccessWithRequestCache`). The previous
  `checkAccess` call omitted `req`, so the agent path resolved the same grant
  twice and issued a second serial `getRoleByName` during chat startup, against
  the rule in AGENTS.md.

- `codeExecutionEnabled` is gated by the role result at all three loaders, not
  only the `execute_code` entry in the filtered list. It drives tool
  classification and the programmatic bash tool, so an agent pairing
  `execute_code` with an MCP `code_execution` tool still advertised and
  instantiated `run_tools_with_bash` for a denied role.

- The upload map covers `EToolResources.code_interpreter`, which the Assistants
  builder posts instead of `execute_code`, and native `code_interpreter` /
  `file_search` tools are dropped when an assistant is created or updated.
  Those run inside the provider and reach neither loader, so configuration time
  is the only place they can be gated.

Renames `handleTools.fileSearchPermission.test.js` to
`handleTools.rolePermissions.test.js` — it now covers both tools.

* 🔐 fix: Gate the Image Upload Route and the v1 Assistant Writers

Second Codex round found the same shape again: boundaries the map had not been
wired into. Adds two shared helpers next to the maps in
`packages/api/src/tools/rolePermissions.ts` so a new upload handler or assistant
writer gets the check by calling one function.

- `/files/images` accepts `tool_resource` and routes agent uploads to
  `processAgentFileUpload` on its own. The Code Files UI sends images there, so
  every image was a way around the upload boundary for a role denied
  `RUN_CODE`. Both upload routes now call
  `checkToolResourceUploadPermission`.

- That helper passes `req`, so the role read joins the request cache. The
  previous `checkAccess` call omitted it, and an allowed Assistants upload goes
  on to `processFileUpload` → `addResourceFileId` → `updateAssistant`, whose own
  check does pass `req` — two serial lookups per upload, against AGENTS.md.

- `/assistants/v1` is still mounted and its `createAssistant` / `patchAssistant`
  write native tools straight to the provider. Both now filter through
  `resolveAssistantToolPermissions`, which v2 also uses in place of its local
  copy.

Run-time enforcement for native assistant tools is deliberately not here: it
means overriding `body.tools` at `createRun` in both chat controllers, which
changes run semantics, and is better argued on its own.

* 🔐 fix: Gate Legacy `retrieval` and the Code-Environment File Tools

Third Codex round, same shape twice more.

- The v1 assistant builder submits `{ type: 'retrieval' }` where v2 submits
  `file_search` (`AssistantPanel.tsx:203`). Only the v2 spelling was mapped, so
  `resolveAssistantToolPermissions` treated legacy retrieval as ungated and both
  v1 writers preserved it for a role denied `FILE_SEARCH`. Both spellings now
  answer to the same grant.

- `codeEnvAvailable` in the agent initializer was capability-only, and
  `initializeAgent` rebuilds `bash_tool`, `read_file` and the workspace file
  tools from it — after the tool loader has already dropped `execute_code` for a
  denied role. The bash gate stops command execution, but those handlers still
  read, search, create and edit files in an attached code environment. The flag
  now carries the role grant, and short-circuits before the role read when the
  deployment has the capability off.

* 🔐 fix: Resolve Tool Grants Once Per Request and Pair Every Capability Gate

Four review rounds found thirteen issues, all one class: a boundary the
permission check had not been copied into. Rounds 2 and 3 each found gaps
created by the previous round's fix, and a manual sweep of tool constructors
and upload routes still missed the `codeEnvAvailable` family entirely — because
the thing worth enumerating is not "paths that reach a tool" but "reads of the
capability".

There are 28 such reads. This replaces per-site permission checks with one
resolution per request, pairs every read that is a gate, and pins the list so a
new one cannot be added silently.

- `resolveToolRoleGrants` resolves `RUN_CODE` and `FILE_SEARCH` together and
  memoizes on the request. Both perf findings disappear by construction: gates
  share one role read instead of each issuing its own.

- Newly paired gates: `codeEnvAvailable` in the OpenAI-compatible, Responses and
  memory-agent initializers (`initializeAgent` rebuilds `bash_tool`, `read_file`
  and the workspace file tools from it, and hands their handlers the code
  environment); the capability checks in agent upload processing and in
  agent-management upload purposes; and the library chat-completion service,
  behind an optional `getRoleByName` dependency so embedders are not broken.

- The v1 Knowledge upload posts `assistant_id` with no `tool_resource`, so there
  is no resource to authorize. It now reads the assistant's own native tools and
  requires their grants, instead of treating a missing resource as unrestricted.

- The startup role read moved off the critical path into the existing
  `Promise.all`, and the direct tool-call check passes `req` so the shared
  loader gate reuses its result rather than repeating the lookup.

- `toolCapabilityGates.spec.js` pins every file that reads either capability
  with a count and a reason. A new read fails the suite until it is classified,
  so the next boundary announces itself instead of waiting for a reviewer.

* 🔐 fix: Abort Assistant Writes on Role-Lookup Failure, Authorize Legacy Uploads First

Round 5 found no missed boundary — the inventory held. All three findings are
about how the gates behave.

- Failing closed is right where a denial blocks the operation, and wrong where a
  denial instead filters a payload the caller persists. A transient role-store
  outage during an assistant create or patch was silently stripping
  `code_interpreter` / `file_search` / `retrieval` and saving the result, turning
  a momentary failure into permanent config loss. `checkToolRolePermission`
  gains `throwOnError`, and `resolveAssistantToolPermissions` uses it: every one
  of its callers either persists the filtered list or rejects the request, so a
  lookup failure has to propagate.

- The legacy-assistant upload gate ran after `sanitizedUploadFn` had already
  pushed the file to the provider, so a denied role left an untracked remote file
  and got a 500 for what is a 403. Moved to the route, next to the tool-resource
  gate, before any bytes are sent.

- `resendFiles` hydration derived its resource set from unfiltered `agent.tools`,
  so a denied role still paid the thread walk and code-file reads for a tool
  about to be dropped. The `execute_code` half now keys off
  `effectiveCodeEnvAvailable`, which already carries the grant.

The `file_search` half of that last one needs the grant threaded into
`initializeAgent`, which is plumbing in the most delicate file here for a latency
win on an already-correct path — deliberately left for a follow-up.

* 🧪 fix: Declare `resolveToolRoleGrants` in the Upload Spec's Module Stub

`Tests: api (shard 2/3)` failed on `a0b38c3`: `process.spec.js` stubs
`@librechat/api` wholesale, so the `resolveToolRoleGrants` call added to the
upload gate resolved to undefined and threw. Five failures in
`processAgentFileUpload`, plus one cascading `processFileURL` assertion from the
same suite.

The gate was verified through its own spec and the routes above it, but not
through the spec of the file it lives in — `process.spec.js` was never run
locally. Every spec belonging to a file this PR touches now has been, and the
mongodb-backed ones that still cannot run here were checked statically for the
same stub shape: all but `process.integration.spec.js` spread `requireActual`,
and that one never reaches these paths.

* ⚡ fix: Skip the Role Lookup When Code Execution Is Disabled

The browser initializer already guards this; the OpenAI-compatible and Responses
controllers did not, so every request on those endpoints issued a role read whose
result the capability short-circuit then discarded. Both now start the lookup
only when the deployment has `execute_code` enabled, matching the third.

* ⚡ fix: Drop Three Redundant Reads on the Gated Paths

- The memory initializer resolved grants unconditionally, so a deployment
  without `execute_code` paid a role read for a flag that could only be false.
  Guarded like the three chat initializers.

- Agent-management upload purposes resolved grants before `checkCapability`, so
  a purpose the deployment has switched off paid a role read on the way to being
  rejected. Capability first, then the grant.

- The legacy-assistant preflight builds an OpenAI client to read the assistant's
  tools, and `processFileUpload` then built a second one — re-reading the user's
  key expiry and values. The preflight now hands its client on, and processing
  builds one only when it wasn't given one.

---------

Co-authored-by: Paul <200737214+SSIG-IT@users.noreply.github.com>
2026-09-07 07:03:22 -04:00
Danny Avila
58de77619e
🗝️ refactor: Prefer the Recorded Storage Key for Every Download Stream (#15695)
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
S3 and CloudFront records have carried `storageKey` since #12987, yet six
readers still handed `file.filepath` to `getDownloadStream` and re-derived
the key by parsing a presigned or CDN URL, while four others had grown
their own `storageKey || filepath` expression. One resolver now serves all
ten: `resolveDownloadPath` returns the recorded key when present and the
path otherwise, so records without a key (local, Firebase, Azure, code
output) behave exactly as before, and the share route keeps its local-only
query-string strip on top.

`resolveStoredS3Key` reuses the same `StoredFileRef` type. Covered by unit
cases for the resolver and an S3 case proving a record with a key streams
correctly even when its stored URL no longer parses.

Closes #15693
2026-09-07 01:28:35 -04:00
aeyeopsdev
e504128e8d
🫙 fix: Surface Document Extraction Failures (#15651)
Co-authored-by: aeyeopsdev <275853971+aeyeopsdev@users.noreply.github.com>
2026-09-06 19:09:49 -04:00
Danny Avila
b181ef0232
🪬 feat: Configure Langfuse Trace User Identity and Metadata (#15667)
* 🪪 feat: Configurable Langfuse Trace User Identity and Metadata

* 🪪 fix: Prefer createRun's conversationId and warn once on an unset userIdField

* 🪪 fix: Validate Langfuse trace field names at the point of use

* 🧹 chore: Sort identity imports

* 🪪 fix: Source the trace model label from the initializer, not a prompt-affecting option

* 🧪 fix: Build the trace context without a prototype so partial client contexts resume

* 📦 chore: Bump `@librechat/agents` to v3.8.0 for the Langfuse trace userId override
2026-09-06 17:30:23 -04:00
Danny Avila
8a85df9715
📜 feat: Add Skills Discovery and Authoring to the Management API (#15642)
* feat: expose skill discovery and authoring through management API

* fix: declare public skill management schema and handler types

* fix: preserve skill permissions and shared management error semantics

* fix: validate skill file paths before writing storage

* test: isolate execution limiter routes from skills management

* fix: align skills discovery access and file upload protections

* fix: authorize skills operations before sync and upload admission

* perf: overlap independent skills authorization reads

* style: sort skills management imports
2026-09-06 17:26:26 -04:00
aeyeopsdev
a827f35a69
🛤️ fix: Align File Search Citation Paths (#15650)
* fix: keep file search anchors aligned with citation passages

* refactor: centralize file citation selection

* test: wire citation helpers into API mocks

---------

Co-authored-by: aeyeopsdev <275853971+aeyeopsdev@users.noreply.github.com>
Co-authored-by: Danny Avila <danny@librechat.ai>
2026-09-06 17:25:47 -04:00
Danny Avila
3f9285355d
🏠 fix: Route Standalone File Writes to BYOM Workers (#15675) 2026-09-06 17:24:21 -04:00
Danny Avila
ebf70826b5
📫 fix: Surface Sandbox Artifact Delivery Failures (#15671)
* fix: Surface sandbox artifact delivery failures

* chore: Sort artifact delivery import
2026-09-06 15:31:19 -04:00
Marco Beretta
c9d208ad3b
🧰 feat: Unify Tool Activity, Reasoning, Search, and Agent Workflows (#14546)
* feat: Group Reasoning with Tool Calls and Polish Activity UI

- Treat reasoning (Thoughts) as transparent to tool grouping so interleaved
  thoughts fold into the tool group instead of splitting it
- Group a lone tool call that has reasoning (e.g. a skill) so it gets the
  same collapsible chrome as multi-tool groups
- Count only real tool calls for the group header; add a reasoning indicator
  and a singular 'Used 1 tool' label
- Rework the in-group Thoughts panel: tool-row sizing, rounded content,
  header copy button, and a floating collapse/copy bar
- Round the floating thinking-bar buttons
- Refine the agent handoff row and instructions panel with consistent
  spacing and a copy affordance

* feat: Move subagent runs to an artifacts-style side panel and polish reasoning UI

- Replace the subagent dialog with a right-side panel that shares the
  artifacts slot (currentSubagentRunId + subagentPanelVisibility atoms,
  a subagentRunsState registry, and a useOpenRightPanel cross-clear so the
  two panels are mutually exclusive)
- Rebuild the inline subagent card to be tool-consistent, preview a single
  live/result line, and open the panel (auto-focusing on stream arrival)
- Relocate the originating prompt to a collapsed, smoothly-animated
  disclosure under the panel header
- Gate the Thoughts copy button to expanded state and move the floating
  bar to a scroll-out reveal, matching the code-block pattern
- Fix the reasoning expand/collapse jump by animating the trailing gap
  inside the grid track instead of an expand-only margin

* fix(client): handle singular grouped tool labels

* refactor(client): unify tool row disclosure styling in a shared module

Extract the hover-reveal chevron and expanded panel spacing into
disclosure.ts and align every tool activity row on the same layout:
h-5 header, my-1 rhythm, full-width click target, and a 200ms chevron
rotation. Rows rendered through ProgressText now stretch across the
message column (the absolute content layer was shrink-wrapping to the
text), and the copy buttons in thinking and output panels anchor to
the row instead of the panel corner.

* fix(client): friendly labels and correct arg parsing for native tools

Map skill, read_file, create_file, edit_file, set_memory, and
delete_memory to localized display names so summaries stop showing raw
tool ids, and give the memory tools the Brain icon the Memories panel
already uses. read_file and skill rows parse the arg the backend
schemas actually emit (path, with file_path as a legacy fallback;
skillName) and fall back to a generic noun instead of rendering an
empty name in their labels.

* feat(client): outcome-based tool group summaries

Group headers now describe the activity performed: homogeneous search
groups read "Searched the web" or "Searched web and files", mixed
groups read "Ran N actions" with per-label repeat counts (Create File
x2) and a failure tally in the detail line. Generic tool rows resolve
their label through getToolDisplayLabel, the OAuth trust warning moved
into a shared ToolAuthWarning that a group renders once instead of per
call, and the group body is a flex column so row margins stop
collapsing unevenly between rows.

* feat(client): render file edits as a structured diff view

Replace the highlighted-text diff in file authoring output with a
line-based renderer: red and green tinted rows with +/- gutters,
old/new line numbers when hunk headers carry them, and diff noise
stripped (file headers, streamed old_text/new_text markers, bare @@
separators become dividers). The code window header shows the file
name with +N/-N change stats next to Copy Code, and file authoring
rows adopt the path-first arg parsing and shared panel spacing.

* refactor(client): align memory updates with the standard tool row

Rebuild MemoryArtifacts on the shared disclosure row (useExpandCollapse
grid animation, hover-reveal chevron, h-5 header) in place of its
bespoke ResizeObserver height animation, and simplify the expanded
panel: section headings dropped in favor of bold memory keys, a single
p-3 inset instead of double padding, and the now-unused heading
translation keys removed.

* fix(client): gate retrieval expansion on renderable results

The row offered an expand affordance whenever raw output existed even
when no sources parsed, opening an empty panel. Expansion now keys off
parsed results, which also covers attachment-only sources that arrive
without text output, and collapses if results disappear.

* style(client): soften the message nav hover tooltip

Rounder corners and the light border weight to match the rest of the
floating surfaces.

* fix(client): compact single-number gutter in diff view

The dual old/new gutters left one column blank on every add or delete
row, reading as a stray number floating far from its marker. Each row
now shows the one relevant line number (old for removals, new
otherwise) right beside the +/- marker, and the scroll container is
vertical-only since lines wrap.

* feat(client): dedicated card for memory tool calls

set_memory and delete_memory rendered through the generic tool card,
reading "Ran Save Memory" with raw parameter chips like an unknown MCP
tool. A new MemoryCall part gives them first-class rows: "Saved
memory" / "Deleted memory" with the key as a quiet subtitle, the
Brain icon, and an expanded panel that mirrors the memory attachment
display (bold uppercase key over the stored value, or a deletion note),
with error output shown inline when a write fails.

* style(client): gap between the web search row and its source list

The expanded source panel sat flush against the 22px favicon stack,
which overflows the 20px row. The panel now opens with a 6px top
margin while the row keeps the shared my-1 rhythm.

* feat(client): richer web search source list

The panel only showed favicon, title, and domain while the search
artifact carries much more. Source rows now include the result snippet
(clamped to two lines) and date under the domain, and a Serper answer
box renders as a lead card above the list with its title and snippet.
Streaming behavior is unchanged; the answer box is attachment-only
since it never carries citation anchors.

* refactor(client): move web search answer box into a details hover card

The inline answer card read as assistant prose sitting inside the
source list. The panel is back to sources only; an Info trigger at the
right edge of the row (hover-revealed like the chevron, focus-openable
for keyboard) opens a hover card with the search query, the source
count, and the provider answer box when present. WebSearch now
receives the tool call args to surface the query.

* feat(client): image, shopping, and place verticals in web search results

Serper vertical results were parsed into the artifact and dropped.
The expanded panel now renders them when present: an image strip of
rounded thumbnails at true aspect ratios (sized from thumbnail
dimensions, no layout shift), a shopping card carousel with product
image, title, price and store, delivery note, and star rating, and a
place list with name, category, address, and rating. All link out,
use semantic lists with localized labels, and cap item counts.
Expansion now keys off any renderable content, so a verticals-only
search still opens.

* style(client): help cursor on the retrieval relevance badge

The badge sits inside a clickable row, so it inherited the pointer
cursor while only offering an explanatory tooltip.

* style(client): strip link underlines in web search verticals

The image and shopping anchors sit inside the markdown prose scope, so
every label rendered with the blue link underline.

* feat(client): surface search verticals outside tool disclosures

Image, shopping, and place results were only visible after expanding
the web search row, and disappeared entirely inside a collapsed tool
group. The strips moved into a shared verticals module that renders
in two spots following the attachment pattern: always below the row
for an ungrouped search, and hoisted below the group (visible while
collapsed) for grouped calls, where the row suppresses its own copy
via hideAttachments. The disclosure panel is back to sources only.

* feat(client): platform-native map links on place results

Place rows were the only vertical without a destination. Each row now
links out to the user's native maps app: Apple devices (detected via
the existing isMacPlatform check) get maps.apple.com with the query
and coordinates, everything else gets a Google Maps search link. Both
are keyless https links that degrade to a web map, so a misdetected
platform still lands somewhere useful. Rows pick up the same hover
treatment as the other linked results.

* fix(client): pin the bash copy button while the command scrolls

The button lived inside the scroll container with sticky/float
positioning, so horizontal scrolling dragged it through the command
text. It now sits absolutely in a non-scrolling wrapper around the
scroller and stays in the corner on both axes.

* feat(client): add view-transition morph helper for composer surfaces

Wrap a state update in startViewTransition (with a synchronous flush) so
elements sharing a view-transition-name animate their position, size and
shape between two states instead of swapping.

The composer band gets its own named, opaque snapshot group stacked above
the morphing element: without it a surface travelling between the thread
and the composer paints over the input and bleeds through the strip below
it. Falls back to a plain update where the API is unavailable and for
reduced-motion users.

* style(client): rebuild the ask question surfaces on shared option rows

The popover and the chat card rendered the same question two different
ways: numbered rows in one, wrapped chip buttons in the other, and the
panel was painted with hardcoded white/gray instead of theme tokens.

Both surfaces now render one AskOptions list, so numbering, check badges,
rounding and hover states match wherever a question appears. The card
groups its prompt with the description, styles the answer box like the
composer, and gains a Skip button beside Submit so declining does not
require the popover. While the popover holds the question the card keeps
a hidden copy of itself in the thread, reserving the exact space the card
will need so nothing reflows when the question moves back.

* refactor(client): make one control move a question between chat and composer

The popover carried both a chevron and a close button. Both left the
question pending and both moved it into the chat, so their only real
difference was invisible: the chevron kept the composer wired up as the
answer box while the close button released it. Two icons for one visible
outcome, with the meaning hidden in a placeholder change somewhere else.

Collapsing now does both jobs at once. Moving a question to the chat hands
the composer back for normal messages, and the card's chevron re-arms it,
so the popover's visibility and the composer's role can never disagree.
Escape maps to the same path and the dismissed-ids state is gone. Both
chevrons get a tooltip and a matching label, since no glyph conveys this.

Also fixes two behaviours found along the way: the card's Skip was a
silent no-op once the question left the popover, because it routed through
a helper gated on answer mode instead of the live pause, and clicking a
multi-select row moved the keyboard highlight, leaving a row painted as
selected after it was unchecked.

* style(client): fix changed-file lint formatting

* style(client): align tool call import order

* fix(client): render handoff instructions as readable context

* test: stabilize full test suite

* test(e2e): align assertions with action groups

* fix: preserve tool run and answer state

* fix: sync subagent and answer handoffs

* fix(client): keep summary parts stable through failed summarize rounds

Failed summarize completions spliced the in-flight summary part out of the
content array, remounting every later index-keyed part (collapsing whatever
the user had expanded mid-stream) and breaking the content-position ==
step-index invariant that streaming writes rely on. Finalize the part in
place instead and scope completion to the owning step's slot so
back-to-back rounds cannot clobber a newer in-flight part.

* fix(api): disable summarization below a viable context budget

A tiny user-set maxContextTokens re-triggers summarization on every graph
step (the summary allocation rounds to nothing and history still
overflows), burning dozens of LLM calls until the recursion limit aborts
the run with an opaque LangGraph error. Fall back to plain pruning under a
1024-token floor so the run either fits or fails fast with the actionable
empty_messages breakdown.

* feat(client): render context budget errors with readable copy

empty_messages and final_context_overflow are structured errors thrown by
the agents SDK, but they were missing from the typed-error dispatch table
and surfaced as raw JSON inside the generic failure text. Map both to
localized copy: empty_messages keeps its token budget breakdown in a
details block, final_context_overflow interpolates the projected and
available token counts.

* Fix ask answer handoff and OAuth warnings

* fix(client): address Codex review findings

* fix(client): address Codex review findings on tool group and summary UI

Replace the raw green/red diff palette in DiffView and CodeWindowHeader with
the existing status-success/status-error semantic roles, so a customized or
versioned theme moves the diff colors with it.

Compose the shared Button primitive in the ask option rows instead of
recreating hover, focus-visible and disabled appearance locally, and swap the
raw text-white on the checked badge for text-on-status.

Prefer a THINK part's generated reasoning_label in the grouped rendering path,
matching the standalone Reasoning path, with the generic Thinking/Thoughts text
only as a fallback.

Derive grouped tool completion and failure from the run step's terminal status
through the shared resolveToolCallPhase helper, so a step the run closed as
failed with empty or benign output no longer reads as running and then reports
success while the card beside it shows a failure.

Carry an explicit failed flag on a summarize round that ended in error, so the
partial deltas already streamed into the slot are no longer presented under the
"Conversation summarized" label.

Also drop seven orphaned English locale keys this branch left unused, which
were failing the unused-i18n static check.

* fix(client): address second round of Codex review findings

Reuse the shared Button primitive for the three question move controls (the
chat card's chevron and both popover chevrons), which each repeated the same
hover, focus-ring, color and rounding recipe locally. Only the icon geometry
stays local.

Drop provider image results that carry neither thumbnailUrl nor imageUrl.
Both are optional in the ImageResult contract, and an unusable entry still
made images.length nonzero, so the card rendered an anchor and img with
undefined destinations instead of omitting the result.

Gate the memory deletion confirmation on a completed phase. The panel opens as
soon as the key streams in and the phase can still settle as cancelled, so
"Memory deleted" was claiming a removal that never happened while the header
correctly read Cancelled.

Key pending answer drafts by action id instead of a single shared slot. With
Save drafts disabled, editing a collapsed question in one conversation and
then editing another paused conversation's question overwrote the only slot,
and the action-scoped reader then showed the first card an empty box with its
unsent answer unrecoverable. Entries are dropped once answered.

* fix(client): address third round of Codex review findings

Reuse the shared CopyButton for the handoff instruction copy control, which
recreated hover, focus-ring, opacity, rounding and copied-state appearance
locally. The icon crossfade and tooltip now come from the shared primitive and
only the reveal-on-hover behavior stays local.

Keep unlabeled commentary out of collapsing tool groups. An unlabeled group
auto-collapses once complete, so folding commentary in hid the assistant's
prose behind a generic action header. The flush now splits the block at
commentary and groups each surrounding run on its own, restoring the shape the
previous fallback had while keeping reasoning folded in.

Drop shopping results without a link and places with nothing to name, for the
same reason the image results are filtered: every provider field is optional,
and an unusable entry still made the array nonempty, so the card rendered an
anchor with no destination that looked actionable but could not be opened.

* fix(client): address fourth round of Codex review findings

Move the disclosure chevron appearance into @librechat/client as
disclosureChevronVariants, owning the hover/focus reveal, the motion and the
open rotation. The four tool, search, memory and reasoning surfaces now pass
their expanded state and keep only icon geometry local, so a change to the
shared behavior reaches all of them. Panel spacing stays feature-local, being
genuine layout.

Require a navigation target on image results, not just an image source: a
thumbnail-only entry passed the previous guard and then rendered a styled
anchor with an undefined href, looking actionable but opening nothing.

Exclude closed run steps from the pending-auth predicate. ToolCall already
hides the sign-in button on a terminal status, so a persisted OAuth call whose
progress never reached 1 left the group showing a trust warning with no
authentication action behind it.

Copy handoff instructions through copy-to-clipboard, as the sibling
message-content controls do. navigator.clipboard is undefined on a non-secure
origin, where dereferencing it threw before the rejection handler could run and
the button silently did nothing.

* fix(client): address fifth round of Codex review findings

Compose the shared Button through HoverCardTrigger's asChild for the web search
details trigger, which recreated rounding, hover colors, focus ring and motion
locally. This also replaces Radix's default anchor with a real button. Only the
hover-card reveal and the icon geometry stay local.

Copy compact reasoning through copy-to-clipboard for the same reason as the
handoff control: navigator.clipboard is undefined on a non-secure origin, where
dereferencing it threw before anything was copied.

Document the streaming-peek fade as a narrowly scoped custom-CSS exception. It
is a mask-image stencil rather than paint, so only its alpha channel is read
and no theme role could meaningfully restyle it.

* fix(client): address sixth round of Codex review findings

Recognize the batched-edit separator pair for every edit in a diff preview.
formatEditPreview writes a `--- old_text N` / `+++ new_text N` pair before each
edit's `@@`, but the parser gated on `inHunk`, which is never reset, so only
the first pair was skipped and every later one was counted as a deletion plus
an addition and rendered as a marker row. Matched by name instead, which leaves
content lines that merely begin with `-` or `+` alone.

Compose the shared Button for the popover's type-below hint, suppressing only
the fill the way Summary's quiet text buttons already do, so the focus ring and
disabled handling come from the shared recipe.

Render an answer box that carries only snippetHighlighted. It is independently
optional from snippet, and the details card opens for any answer box, so such a
result previously showed a heading with an empty body.

* fix(client): match synthetic edit headers only in separator position

Matching the marker names alone was wrong: deleting a real `-- old_text`
comment line, as SQL and Lua use, is prefixed to exactly `--- old_text`, so the
previous guard discarded a genuine changed line from both the preview and the
statistics. The pair is now skipped only in separator position, adjacent and
immediately followed by a hunk header, which no run of genuine content
satisfies. Covered by a test asserting such a pair inside a hunk survives as
content.

Also drops the em dashes from the comments added over this review pass.

* fix(client): address eighth round of Codex review findings

Compose the shared Button for the compact reasoning disclosure, keeping only
the row geometry local so the focus ring and disabled handling come from the
shared recipe.

Lazy-mount the compact reasoning body with useLazyCollapseBody, as the
standalone panel already does. Collapsed is the default whenever thoughts are
hidden, and a streaming THINK part re-renders on every delta, so the full text
was being mounted and laid out behind an invisible panel.

Classify grouped memory failures with isMemoryFailureOutput. Those tools report
failure in prose that generic isError parsing does not recognize, so a
persisted call with no terminal status showed a failed card inside a group
reporting success with no failure count.

Announce the diff statistics to assistive technology. The +/- glyphs carry
their meaning through color and sign, so aria-hidden left screen-reader users
without the change summary; the counts are now also exposed as localized
screen-reader text.

* fix: persist failed summaries and stop double-rendering memory mutations

Stamp the failed marker on the aggregated summary part server-side, mirroring
how ON_RUN_STEP_CLOSED already stamps runStepStatus. The SDK aggregator ignores
a complete event that carries no summary, so an errored round's partial deltas
were saved unmarked and a reload re-rendered the truncated text under
"Conversation summarized".

Skip memory attachments that carry a toolCallId in the message-level
MemoryArtifacts decoration. Those calls already render inline as a MemoryCall
card with the same key, value and outcome, so one mutation appeared twice.
Attachments with no originating call keep the decoration as their only surface.

Keep a reasoning-bearing group expanded when Open Thinking Dropdowns by Default
is on. The folded-in THINK part only renders inside the group body, which is
not even mounted while collapsed, so the preference had no visible effect until
the action group was opened by hand.

Give same-named place results unique keys. Several branches of one chain share
a name and often carry no identifier, so rows whose address, rating and map
target all differ collided on one React key.

* fix: adopt the shared code surface and count background task failures

Register surface-code in the versioned theme registry and use it for the diff
preview. The token, its light and dark CSS values and its Tailwind mapping were
already in place and already consumed by CodeBar, CodeBlock and ResultSwitcher;
only the three theme interfaces, the two bundled themes and DiffView were
missing. DiffView no longer coordinates two unrelated surface roles through a
dark: override, and a custom theme can restyle the code surface centrally.

Include background-task failures in grouped outcomes. A backgrounded bash or
code task reports its verdict through a background_task_status attachment
rather than its output, and its dispatch step keeps a benign handle and usually
closes as completed, so the group reported no failed action beside a card
showing failure. Group metadata now derives that signal with the same helpers
the child card uses, correlated per part so a handoff response repeating a
provider tool-call id across agents still routes correctly.

Require a title on shopping results. A link-only card had no accessible name
and no visible identity: the anchor carried no label, its image is decorative
and the title span rendered empty.

* fix(client): compose shared buttons for two disclosures and seed re-enabled ask drafts

Compose the shared Button for the memory-artifacts and web-search disclosure
rows, keeping only their geometry local. The web-search row keeps its instant
hover by suppressing the recipe's color transition, since the chevron beside it
reveals without a fade and the existing test pins that.

Seed the ask-specific draft when Save drafts is re-enabled. An answer captured
while saving was off has no persisted ask draft for autosave to restore, so
expanding opened an empty composer and the next collapse read that empty value
back over the in-memory answer.

* fix(client): surface cancelled actions and protect the released composer

Track cancellation separately from failure in grouped tool metadata and name it
in the group header. A stopped step is settled and not an error, so the group
auto-collapsed under a success-sounding header with the only cancellation
notice hidden inside the panel it had just closed.

Preserve an ordinary unsent message across an ask handoff. With Save drafts
off, collapsing a question releases the composer; typing there and then moving
the question back overwrote that text with the answer and lost it. The released
text is now stashed per pending action and handed back on the next collapse.

Skip the named view transition for batched questions. Only the single-question
surfaces declare view-transition-name: ask-question, so wrapping a batch
handoff gave the browser nothing to pair and it cross-faded the whole root
instead of morphing the question.

* fix(client): restore the released composer text after answering

Submitting an answer from the expanded card never goes through collapse, so the
ordinary unsent message stashed when the question was moved back was never
handed over and its entry leaked. The resume success path now restores it in
place of the reset when one is stashed, and drops the entry either way. Read
through a ref because the callback settles long after the render that armed it.

* fix(client): stop stacking search snapshots and fix three identity defects

Upsert re-emitted web-search attachments instead of appending them. The search
helper rewrites and re-emits the entire payload after every highlight update,
keeping the same type, toolCallId and turn, and those attachments carry no
file_id so the existing upsert never saw them. Each snapshot was appended, so
collectSearchVerticals concatenated one copy of every image, product and place
per update. Keyed by type, tool call and turn, replacing wholesale since the
incoming payload is complete.

Preserve MCP identity in the tool-call label. function_name has already had the
MCP delimiter and server stripped, so re-parsing it classified an MCP function
sharing a built-in's name as that native tool and both showed and announced an
unrelated label.

Qualify grouped tool identity by agent. Provider tool-call ids repeat across
agents in a handoff transcript, and a lone tool with reasoning now groups too,
so two sibling groups shared one React key and one expansion override.

Compose the shared Button for the handoff disclosure, keeping only its geometry
local.

* refactor(client): build edit previews structurally instead of round-tripping diff text

The args preview was formatted into unified-diff text purely so parseUnifiedDiff
could read it back, and the separators that round trip needed between batched
edits are byte-identical to a real changed line: deleting a source line
`-- old_text` prefixes to exactly `--- old_text`. Neither the marker names nor
their position could tell the two apart, so the parser either inflated the
statistics or dropped a genuine edit.

buildEditPreviewDiff now constructs the preview's lines from the structured
edits, so edit boundaries are structural and no separator is ever emitted.
parseUnifiedDiff is back to parsing only real tool output, where `--- ` and
`+++ ` are genuine file headers, and it no longer special-cases synthetic
markers at all. Copy text comes from a one-way serializer.

Also include agent identity in the web-search snapshot key: handoff agents
repeat provider tool-call ids, so without it one agent's search results
replaced another's.

* fix(client): stop counting empty edit sides as changed lines

A pure deletion (`new_text: ""`) rendered a blank green row and reported one
addition, because `''.split('\n')` yields one element. A trailing newline did
the same at the end of either side. Both sides now go through one helper that
treats an empty side as no rows and a trailing newline as a terminator. The
defect predates the structural builder; it was carried over from the string
preview it replaced.

* fix(client): restore the released text for non-text answers too

Only the free-text path sets consumedComposerText, so gating the restore on it
discarded the stashed ordinary message whenever the user answered by clicking an
option or skipping, while the deletion below ran regardless. The stash is now
restored whenever one exists, independently of the reset condition, and it
outranks whatever the composer holds since that is the answer draft the
submission just consumed.

* fix(client): make group identities collision-proof and localize handoff labels

Disambiguate duplicate tool-group identities by how many earlier groups already
claimed the same provider tool-call id. Those ids repeat across agents in a
handoff transcript and restart per turn within one agent, and a lone tool with
reasoning now groups too, so sibling groups could share a React key and one
expansion override. Occurrence order is stable because groups are append-only,
and the first occurrence keeps the bare id, so an id that never repeats is
unchanged and still survives a content-index shift.

Restore a stashed ask message regardless of the current Save drafts setting. The
stash is only ever created while saving is off, so it never reached the
conversation draft; gating the restore on the setting dropped it whenever the
user enabled saving between expanding and collapsing.

Localize the handoff field labels the SDK defines (instruction, instructions,
context). Any other key is an admin-authored promptKey from the agent graph, so
it stays as written, like an agent's own name.

* fix(client): reject unsafe vertical links and scope group identity message-wide

Filter provider-supplied vertical URLs through the markdown allowlist. Search
result payloads are copied into the attachment without validation, so an image
or shopping target could be `javascript:` or another non-network scheme and the
card handed it straight to href. The same helpers are used when admitting a
result and when rendering it, so the two cannot disagree.

Derive tool-group identity from a message-wide occurrence ordinal instead of a
per-instance counter. Phase slices render through separate nested ContentParts
that share the expansion map, so a per-instance count restarted and let a later
slice inherit an earlier group's override. The ordinal is stable under
content-index shifts, and ordinal 0 keeps the bare id, so a group keeps its
identity when a completing phase moves it into a slice.

Build group attachments as a union. Repeated provider ids inside one group
resolve to the same bucket, so appending per part rendered each file and search
snapshot once per repeated call.

* fix(client): let an output diff supersede a create preview, and guard two edges

A diff in the output now supersedes everything, including a create's authored
content. `create_file` overwriting an existing file returns a summary plus a
unified diff, and preferring the authored content there rendered that diff
twice: once through DiffView and again verbatim as the raw output section. This
restores the precedence the string-preview version had.

Give image tiles distinct React keys. Syndicated images reuse one thumbnail
across results that link to different pages, and grouped searches merge
attachments, so the source alone collided.

Require the composer to still hold the submitted text before restoring the
stashed ask message. The textarea stays editable while the resume is in flight,
so anything typed since is newer than both the stash and the answer.

* fix: address remaining Codex findings after dev rebase

Stamp agent ownership on streaming and final search snapshots. Compose the shared floating thought button. Preserve rebased group behavior, correct stale test selectors, and keep expansion callbacks outside state updaters.

* Resolve follow-up review findings

* Scope grouped background verdicts by execution step

* Fix canonical artifact ownership and frontend review regressions

* Fix terminal answer handoff and encoded edit previews

* style(client): keep the search rows on dev's glyph rail

The rebase onto dev landed both halves of #15643's row rail except in
`WebSearch`, whose completed row had since moved its glyph into a shared
`Button` and whose streaming row kept a 10px gap. Restore the shared
`ROW_GLYPH_SLOT` around the completed row's favicon stack or globe and
drop the streaming row's gap to 8px, so both search rows put their glyph
on the avatar's axis and their text where the header's name starts.

* fix(client): pass the thinking preference into ReasoningCompact

`showThinking` is an app-global display preference the group only reads,
and `ToolCallGroup` already receives it from its host. `ReasoningCompact`
still reached for `showThinkingAtom` itself, coupling a grouped thought
to `~/store` and blocking the move to a feature-owned workspace. It
takes the preference as a required prop now, from the value the group
already holds.

* test(client): pin both web-search rows to the shared glyph rail

The rebase onto dev silently dropped `ROW_GLYPH_SLOT` from the completed
row — its glyph had since moved inside a shared `Button` — and left the
streaming row on a 10px gap, so nothing caught either. `Part.tsx` routes
web search past `ProgressText`, so this renderer owns its glyph and has
to take the slot itself; the test fails on both regressions.

* fix(client): keep grouped markers and folded search media visible

Two things a fold could swallow.

A detached-subagent projection is an empty THINK part flagged
`reasoning_unavailable`, which `Part` renders as a `ReasoningMarker`.
`ToolCallGroup` sent every THINK part to `ReasoningCompact`, which has
no text to show and returns null, so the marker disappeared the moment
its call joined a group. Those parts go back through the standalone
renderer.

A phase hoists its span's media out of the fold, and the nested segment
renders with `hideAttachments` so a `WebSearch` row inside it stands
down for that hoist. The hoist was `AttachmentGroup` alone, which drops
`web_search` attachments outright, so a folded search lost its images,
products and places instead of lifting them out. The card renders
`SearchVerticals` beside the attachment row, mirroring `ToolCallGroup`.

* fix(client): recover a released composer stash on revisit

Recovery of the ordinary unsent message stashed while the card owns the
composer was gated on `lastAnswerableRef` — what this hook happened to
observe. Navigating away mid-resume clears that ref while the stash
rightly stays put, so a resume that succeeded or expired while the user
was elsewhere left nobody to hand the message back: returning to the
conversation found `answerable === false` with no ref, exited early, and
the text sat unreachable in `releasedComposerTextAtom` for the session.

The stash carries its conversation now, and recovery keys on that, so a
revisit is a hand-back like any other exit. Only the exit this hook
armed knows what the composer held when the resume started; on a revisit
an empty composer is the only target that cannot overwrite something
newer, and either way the entry is dropped.
2026-09-06 14:56:50 -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
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
5e3cdd0b2c
📮 fix: Consume an Invite Only Once the Account Exists (#15611)
* 📮 fix: Consume an Invite Only Once the Account Exists

`checkInviteUser` deleted the invite token and then called `next()`, but everything
that can still reject a registration runs after it: the schema, the allowed-domain
check, and the email-already-in-use check. A mistyped password confirmation therefore
destroyed the invite — the invitee corrected it, resubmitted, and got "Invalid invite
token" with no way back short of an admin re-inviting them.

The deletion moves to `registrationController`, after `registerUser` reports success.

That report needed a new signal. `registerUser` returns the same 200 and the same
generic message whether it created an account or found the email already in use —
deliberately, so the response cannot be used to enumerate accounts — so the status
alone cannot say whether an account exists. It now also returns `userCreated` on the
creation path, which the controller reads and never forwards; the response body is
unchanged.

A failed deletion is logged rather than surfaced. By that point the account exists,
and leaving a usable invite behind is recoverable in a way that telling the user their
registration failed is not.

Fixes #15541

* ♻️ test: Fold the Invite Tests Into the Existing AuthController Spec

From Copilot's review. The new spec reached for `jest.requireActual` on
`@librechat/data-schemas`, `~/server/services/AuthService` and `~/models`, which pulls
real implementations into a unit test — `~/models` builds the data-schemas methods
against mongoose — and diverges from `AuthController.spec.js`, which stubs each module
outright.

Rather than restate that harness with narrower stubs, the tests move into the spec that
already has it. `deleteTokens` joins its `~/models` mock, which also keeps that shared
mock in step with the controller's imports: a stale mock there hands the controller an
undefined function, and the omission only surfaces when some later test happens to
exercise the path.
2026-09-04 15:38:32 -04:00
Danny Avila
fe30e3db32
🪫 fix: Back Off Failed Retention Deletes (#15608)
* 🧹 fix: Stop Undeletable Files Starving the Retention Sweep

`getExpiredFiles` returns the oldest `expiredAt` first, capped at `limit`,
and `processDeleteRequest` leaves the record in place when storage deletion
fails. Nothing records the failure, so the same files come back at the head
of the next batch an hour later, forever: no backoff, no cap, and — once
`limit` of them cannot be deleted — no file that expires afterwards is ever
swept again. On the deployment behind #15511 that is ~29k stranded objects
permanently occupying a 100-slot queue, which is why fixing the Code
Interpreter side alone (LibreChat-AI/code-interpreter#85) would not have
resumed deletion there.

Failures are now recorded on the file. `deletionRetryAt` holds it back with
a backoff doubling from one sweep interval to a day, and `deletionAttempts`
retires it from the query entirely once it reaches
`FILE_RETENTION_SWEEP_MAX_ATTEMPTS` (10, so roughly five days of retries).
Both fields are absent on existing records and absence means "never
attempted", so nothing already in the collection changes eligibility.

The record itself is kept rather than deleted — the reference is what an
operator needs to reconcile a bucket the sweep could not clear, and dropping
it would restore the silence that made this leak invisible. Raising
`FILE_RETENTION_SWEEP_MAX_ATTEMPTS` re-admits everything previously given up
on, which is the supported way to resume once the storage-side failure is
fixed; the give-up is logged with that instruction.

The counter is incremented server-side with `$inc` so concurrent sweeps on
separate nodes cannot overwrite each other's progress toward the cap, and a
failure to record a failure is logged and skipped rather than aborting the
rest of the batch.

* 🧭 fix: Delete Code Environment Files Through the Route That Exists

`deleteCodeEnvFile` tried `/sessions/:sid/objects/:fid` before falling back
to `/files/:sid/:fid`. Both have been there since #13424, but only the
second is mounted by any released codeapi — the first gained DELETE in
LibreChat-AI/code-interpreter#85 — so every deletion paid a guaranteed 404
and a wasted round trip, and #15511 read that 404 as the whole bug.

Call `/files/:sid/:fid` directly. It is the safe direction to collapse
toward: codeapi has mounted it since its first release, so this works
against older deployments as well as post-#85 ones, whereas keeping the
other path would not.

Collapsing the loop tightens two behaviours that only existed to serve it.
A 405 now surfaces instead of being swallowed on the way to a second
attempt; there is no second route to try, and a service that refuses the
method should say so. A 404 is still treated as "already gone" — that is
the only thing it can now mean — but it is logged rather than passed over
in silence, because a 404 caused by a misconfigured base URL looks
identical and this branch drops the file's metadata record either way.

* 🔒 fix: Settle sweep retry state from the write, not the read

Two findings from the review of 0e52924.

- `recordFailure` derived the attempt number by adding one to the count the
  batch had queried. Two nodes sweeping the same file read the same value,
  so both believed themselves to be the same attempt: each `$inc` landed,
  the stored count crossed `FILE_RETENTION_SWEEP_MAX_ATTEMPTS`, and neither
  caller ever saw the threshold. The file drops out of the query — the
  starvation guard still holds — but the give-up is never reported, and
  that log line is the operator's only notice, and carries the instruction
  for resuming. Return the count from the increment itself so every caller
  gets a distinct attempt number and exactly one observes the cap.

  The same staleness shortened the backoff, so `deferExpiredFile` now
  writes with `$max`: a deferral can only move later, and a node that
  computed a shorter delay cannot pull the file forward past one another
  node already committed.

- The backoff doubled from a hard-coded hour while claiming to start from
  one sweep interval. At the default they coincide; away from it the
  schedule stops meaning anything — on a six-hour sweep the first three
  attempts all land on consecutive passes, and on a five-minute one the
  first retry skips twelve. Derive the base from
  `FILE_RETENTION_SWEEP_INTERVAL_MS`, floored at a minute so a
  pathologically short interval cannot spend the whole give-up budget on a
  transient outage.

* 🧯 fix: Keep the give-up notice and the retry budget honest

Four findings from the review of ee1bac5.

- The give-up was reported after the deferral write, inside the same catch.
  Once the increment lands the counter is durable, so `getExpiredFiles`
  already excludes the file; a deferral that then failed left only the
  generic recording error and dropped the one line naming the file and
  saying how to resume. Report it as soon as the increment returns, and let
  the deferral fail on its own.

- Retry deadlines were measured from `Date.now()` after the deletion I/O,
  but `startExpiredFileSweep` arms its interval before the sweep runs. A
  one-interval delay therefore expired just *after* the next scheduled
  pass, which skipped the file and pushed its first retry out by a whole
  extra interval — and the same drift applied to every delay that is an
  exact multiple. Anchor deadlines to the sweep's start instead.

- The threshold used `>=`, so once two nodes pushed a file past the cap
  every one of them past it logged the give-up: one error per replica per
  exhausted file rather than one actionable notice. Only the attempt that
  lands exactly on the cap reports it.

- Retry state outlived the content it described. `processCodeOutput` reuses
  a record for a repeated `(filename, conversationId)` — new bytes, new
  storage key, and a fresh `expiredAt` from `getRetentionExpiry` — while
  `createFile` and `updateFile` set only supplied fields. A record carried
  to the cap by its previous content stayed excluded from the sweep
  forever, stranding the new object exactly as this PR set out to prevent;
  a partially failed one started the new object's budget already spent.
  Both write paths now clear the fields: the budget belongs to the storage
  a record currently points at.

The retry-state fixtures in `file.spec.ts` were seeded through `createFile`,
which now clears them, so they drive the real methods the sweep uses.

* 🎯 fix: Clear the retry budget only when a retention lifecycle starts

Both findings from the review of 8d26f9b, and both are consequences of that
commit's reset rather than of the original change.

- The reset was unconditional on every `createFile`/`updateFile`, on the
  reasoning that those paths write content. Two of them do not.
  `prepareImages{Local,Azure,Firebase}` call `updateFile({ file_id })` with
  nothing but the id — a TTL touch — every time an existing image is
  encoded for another chat, and the deferred preview uses the same method
  to transition `status`. Either handed a stranded record a fresh set of
  attempts and another give-up notice, on repeat, defeating the cap for the
  files most likely to be stranded.

  Gate it on the write actually setting `expiredAt`. The budget belongs to
  a retention lifecycle: that is the write which starts a new one, it is
  what `processCodeOutput` supplies when it repurposes a record, and a
  record with no retention deadline is never swept, so the fields are inert
  there anyway.

- Both retry writes bumped `updatedAt`. `processCodeOutput` falls back to
  `updatedAt` as the writer-order stamp for records that predate
  `metadata.sourceDispatchedAt`, so a failed sweep landing mid-harvest read
  as a newer content writer and the harvest dropped its attachment. Mark
  them `timestamps: false`, for the reason `claimCodeFile` already does:
  bookkeeping is not a content write.

* 🅿️ refactor: Park exhausted files instead of excluding them

Three review rounds in a row found defects in how the give-up cap interacts
with record reuse, each in the fix for the last. That is a design error, not
a bug list: a permanent exclusion has to be bound precisely to the content
lifecycle it was recorded against, and File records outlive their content.
`processCodeOutput` repurposes a row for a repeated
`(filename, conversationId)`, `createFile`/`updateFile` set only supplied
fields, and `getRetentionExpiry` returns `{}` on a lookup failure so the row
inherits its old deadline — three separate ways for bookkeeping to survive
into a lifecycle it does not describe, each needing its own guard, and the
two retry writes needing to be lifecycle-conditional on top.

Remove the category instead. `deletionRetryAt` becomes the sweep's only
hold, and reaching `FILE_RETENTION_SWEEP_MAX_ATTEMPTS` parks the file for a
month rather than excluding it. The bound on the batch is the same — a
stranded file costs one slot a month instead of one an hour — but a deadline
that outlives its content can only delay the next object, never lose it, so
nothing outside the sweep has to reason about this state at all.

That deletes more than it adds:

- `createFile` and `updateFile` go back to their original form. No reset, so
  no question of which writes install content, and `prepareImages*` and the
  deferred preview stop mattering here.
- `getExpiredFiles` loses `maxAttempts` and its `$and`; eligibility is one
  `$or` on the deadline.
- The interleaving race between the increment and the deferral degrades from
  a stranded object to a delayed one.

Working through a large backlog is throughput-bound either way: every
attempt costs a slot in the bounded batch, so N stranded files need N ×
`FILE_RETENTION_SWEEP_MAX_ATTEMPTS` passes to settle. Lower that limit when
recovering a deployment that has accumulated many.
2026-09-04 15:38:17 -04:00
Danny Avila
7d9db14ba5
📓 feat: Author Files in Attached Workspaces (#15548)
* feat(code): route file mutations to attached workspaces

* fix(code): Inspect attached edits before commit

* fix(api): Annotate workspace limits

* fix(code): Normalize workspace edit previews

* style(api): Sort workspace imports

* fix(api): Preserve Workspace List Validation
2026-09-04 12:16:23 -04:00
Danny Avila
770238f22b
🔣 refactor: Replace Mongo Filters With Typed Criteria (#15580)
* ♻️ refactor: Take Typed Criteria, Not Mongo Filters, for Actions and Assistants

`createActionMethods` and `createAssistantMethods` accepted `FilterQuery<T>`,
putting Mongoose's query language in the package's public surface: every caller
wrote raw Mongo, and no engine other than Mongo could satisfy the signatures.

Both now take domain criteria — `ActionQuery` and `AssistantQuery` — naming
concepts rather than stored fields, with an array meaning "any of these". A
per-domain field map translates them, so `FilterQuery` survives only inside the
two translators.

`buildFilter` throws on a criterion the field map does not cover. Callers in
`api/` are JavaScript and get no compile-time checking, and the server sets
`strictQuery`, under which a silently dropped criterion widens the filter
instead of narrowing it — an unscoped `findOne` returns an arbitrary document
rather than none. Failing closed is the only safe default.

`AssistantQuery.avatarFilepath` covers the avatar-authorization lookup, which
queries a nested path no plain field name reaches.

No behaviour change: every migrated call site produces the same filter as before.

* 🧪 test: Update Avatar Authorization Assertion to Typed Criteria

`validateImages.spec.js` asserted the old `{ 'avatar.filepath': { $in: [...] } }`
call shape for `getAssistant`. The earlier sweep matched on `action_id`,
`agent_id` and `assistant_id`, none of which appear in a dotted avatar path, so
this one assertion was missed.

The `getAgent` assertions in the same file keep the filter shape — that method
still takes a `FilterQuery`.

* 🔒 fix: Reject Unknown Criteria Before Omitting Undefined Ones

`buildFilter` skipped a criterion whose value was `undefined` before checking it
against the field map, so a JavaScript caller misspelling a key that happened to
carry `undefined` — `deleteActions({ agent_id: maybeId })` — produced `{}` and
reached `deleteMany` unscoped. That is the exact failure the guard exists to
prevent, defeated by the order of the two checks.

Validate every key first, then apply the undefined-omission rule. The lookup is
an own-property check so inherited names cannot resolve through the field map's
prototype: `{ toString: 'x' }` previously found `Function.prototype.toString`,
passed the truthiness test, and wrote a garbage filter key instead of throwing.

A recognized criterion left `undefined` is still omitted, so optional parameters
keep working.

* 🔒 fix: Reject Query Fragments as Criteria and Keep the Translator Internal

Three review findings on the criteria translator, all of the same shape: the
guard was narrower than the surface it protects.

Criterion values are now validated as scalars or lists of scalars. `matchAny`
copied any non-array through unchanged, so a JavaScript caller could pass
`{ agentId: { $ne: null } }` and have the operator land verbatim in the filter
— the fail-open case the field map exists to close. `null` is rejected too; it
would have matched documents missing the field entirely.

`matchAny`, `buildFilter` and `FieldMap` are no longer re-exported from the
package barrel. They emit Mongo, and shipping them as public API undercut the
point of the change; they were reachable as `require('@librechat/data-schemas')
.buildFilter`. `OneOrMany` moves to `types/query` because the domain query types
reference it and it names no engine — which also drops the `types` -> `utils`
import.

`loadActionSets` names `ActionQuery` through `import()` so the JSDoc type
actually resolves.
2026-09-04 11:54:21 -04:00
Danny Avila
a21c7944ee
🏷️ fix: Record the Media Type of Saved Images, Not the Declared One (#15606)
`saveBase64Image` recorded the media type declared in the incoming data URL
while persisting bytes that sharp had re-encoded. Those are not always the same
format: sharp rasterizes SVG to PNG, so an image arriving as `image/svg+xml` was
stored as `<id>-<name>.svg`, typed `image/svg+xml`, holding PNG bytes.

That type is not cosmetic. `encode.js` hands `file.type` to providers verbatim —
`media_type` for Anthropic, `inlineData.mimeType` for Google, and the `data:`
prefix for OpenAI — so re-attaching such an image sends PNG bytes under a media
type the provider does not accept for images, and the request fails.

A declared type is also only ever a claim. The same mismatch arises whenever a
producer mislabels what it sends, which for MCP tool output means any connected
server can decide what the record says about bytes it did not have to match.

`resizeImageBuffer` already read the encoded output back to measure it, so it now
resolves that metadata to a media type and returns it, and `saveBase64Image`
records it in place of the declared one, falling back to the declared type only
when sharp reports a format that has no media type of its own. The upload paths
were already correct — they record `image/${imageOutputType}`, the format they
convert to — so this brings the tool-output path in line with them.

`resolveImageMimeType` lives in `packages/api` because AVIF and HEIC share the
heif container and are told apart by its compression, which is worth stating once
with tests rather than open-coding at each call site.
2026-09-04 11:51:28 -04:00
Danny Avila
fddf7e31f2
🛻 feat: Route Bash Through Attached Workspaces (#15546)
* feat(code): Route bash through attached workspaces

* fix(code): harden attached bash schema adapter
2026-09-04 11:49:51 -04:00
Danny Avila
e23e13e712
🌳 feat: List Attached Workspace Trees (#15527)
* feat(code): list attached workspace files

* test(code): include workspace list history

* fix(code): validate workspace tool scope consistently

* fix(code): Harden workspace listings

* fix(code): Align workspace list validation

* fix(code): Continue complete workspace listings

* fix(code): preserve canonical path protections

* test(code): keep workspace list specs type-safe
2026-09-04 10:52:34 -04:00
Danny Avila
9a64e85b1e
🕯️ feat: Search Text in Attached Workspaces (#15525)
* feat(code): search attached workspaces

* fix(code): remove duplicate workspace type import

* fix(code): abort attached workspace searches

* docs(code): correct workspace abort parameters

* test(code): include workspace search history

* fix(code): keep workspace search foregrounded

* fix(code): preserve attached search routing

* fix(code): Bound workspace search output

* style(code): Format workspace search output
2026-09-04 10:07:06 -04:00
Danny Avila
b7fd4ba2a2
👓 feat: Read Attached Workspace Files (#15524)
* feat(code): read attached workspace files

* style(api): sort workspace handler import

* fix(code): abort attached workspace reads

* test(code): type workspace fetch doubles

* style(code): sort workspace bridge imports

* fix(code): Preserve workspace read boundaries

* style(code): Flatten workspace line validation

* fix(code): Bound attached workspace responses

* fix(code): Require canonical workspace paths

* fix(code): fence workspace bridge operations
2026-09-04 09:58:46 -04:00
Danny Avila
30cd1c2f2e
🔇 fix: Stop logging an absent create_file target as an error (#15591)
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
`create_file` reads its target before writing so it can tell a create from
an overwrite, so `readSandboxFile` missing the path is the ordinary case for
a create — yet it was reported at error level, through `logAxiosError`, which
rendered the sandbox's own `cat: …: No such file or directory` as "An error
occurred while setting up the request" plus a stack. Every ordinary file
creation emitted an error-level entry claiming a transport fault about a file
that was written half a second later, and an operator reading it concluded the
file had been lost.

Split the two failures `readSandboxFile` was conflating. A request that never
completed still goes to `logAxiosError`; a request the sandbox answered with
stderr is now classified on that stderr — an absent path logs at debug, and
anything else (permission, a broken interpreter) keeps error level with the
sandbox's own wording. The thrown error is unchanged, so the authoring flow's
create/overwrite decision still reads it the same way.

The classifier moves to `packages/api/src/files/code/errors.ts` and is now
shared: `handlers.ts` had two near-duplicate copies of it, the narrower of
which documented itself as narrower than the broader one without either
deriving from the other.

Also carry the cause on `[preflightCodeOutputBatch] … could not be inspected`.
That warning follows each `create_file` in the same request, and the batch
discarded the error entirely — so a Code API that refused the download, one
routed to the wrong execution profile, and a file too large to inspect all
arrived as the same sentence, and neither this warning nor the one above could
be told apart from the failures around them. It goes through
`getSafeErrorMetadata`, which is what keeps the upstream message and the
artifact's name — both of which can echo submitted content — out of the log.
2026-09-04 09:21:24 -04:00
Danny Avila
2d794a0439
💓 fix: Detect Silently Dead Redis Sockets With Deadline-Bounded Heartbeats (#15539)
* fix: Detect silently dead Redis sockets with deadline-bounded heartbeats

A Redis peer can vanish without a FIN or RST (a dropped NAT entry, a proxy
failover, a migrated VM). The socket stays `ready` from ioredis's point of
view, no error fires, and every command on it waits for the kernel's
retransmission timeout — about fifteen minutes at Linux defaults.

Observed in production: while the main client's socket was dead,
POST /api/agents hung until Cloudflare returned a 524; while the subscriber's
socket was dead, every attach failed with "Timed out synchronizing Redis
subscription" for sixteen minutes across five conversations.

`REDIS_PING_INTERVAL` could not help — it awaits ping() with no deadline, so
the probe just joins everything else that is waiting.

- Add startRedisHeartbeat: races a PING against a deadline and destroys the
  socket when the probe goes unanswered, so ioredis reconnects, replays queued
  commands, and re-subscribes a subscriber's channels.
- Destroy the stream rather than disconnect(true), whose FIN a vanished peer
  never acknowledges; cluster clients fall back to the regular reconnect.
- Add createIoRedisSubscriber so every dedicated subscriber gets an error
  listener and its own heartbeat, and route the generation-stream and both
  subagent-routing subscribers through it.
- Set ioredis keepAlive from REDIS_KEEP_ALIVE (default 10s).
- Add REDIS_PING_TIMEOUT, REDIS_SUBSCRIBER_PING_INTERVAL and REDIS_KEEP_ALIVE,
  documented in .env.example.
- Give the fake RESP server a silent mode and a SUBSCRIBE reply so the
  dead-peer case can be tested against real clients.

* fix: Probe each cluster node and refuse a non-positive heartbeat deadline

A cluster-level PING is routed to one node, so a healthy reply proved nothing
about the other sockets. The heartbeat now resolves its targets each tick —
every node for a cluster, the client itself otherwise — and races an
independent deadline per target, tearing down only the socket that stops
answering. A non-positive deadline disables the heartbeat with a warning
instead of forcing a reconnect on every tick, and the single-probe test no
longer depends on wall-clock slack.
2026-09-03 09:45:16 -04:00
Danny Avila
8a46ca070c
🧊 feat: Persist Context Fading Tier Across Agent Runs (#15505)
* 🧊 feat: Persist Context Fading Tier Across Agent Runs

Carry the pruner's latched context-fading tier from `@librechat/agents`
through `contextMeta` the way `calibrationRatio` already travels, so a
historical tool result keeps the same truncated bytes from one run to the
next and Anthropic's prefix-based prompt cache survives across turns
(danny-avila/agents#497).

- Add `IAgentFadingTier` to the message and conversation `contextMeta`
  types, schemas and the data-provider zod schema
- Add `isAgentFadingTier`, `resolvePersistableFadingTier` and
  `resolveRunContextMeta` in `packages/api`; a tier is persisted once it
  carries information (masking active or a budget below the window), and
  a latched tier is kept even at a neutral calibration ratio
- Seed `createRun({ fadingTier })` from the parent response's contextMeta
  regardless of encoding, since caps are character-based; the field rides
  the existing forward-compatible `runConfig` variable and older SDK
  versions ignore it, as does `Run.getFadingTier` being absent
- Accept a valid `fading` field in event actor context and reject a
  malformed one, matching the calibration checks
- Capture run context meta through one module-level helper for both
  completion paths instead of two inline copies

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Seed Fading Tier Across HITL Resume, Validate at the DB Layer

Address review findings on fading tier persistence:

- Leave the "is this tier worth persisting" decision to the SDK, whose
  `Run.getFadingTier()` now returns only informative tiers; comparing
  against the client's window misclassified untouched conversations
  whenever a reserve ratio was configured
- Carry `contextMeta` through the HITL pause projection (staged approval,
  job metadata, resume state) and seed the rebuilt client from it, so a
  paused turn resumes with the same tier and calibration instead of
  re-deriving a shallower one and rewriting the prefix
- Share run seeding between `chatCompletion` and `resumeCompletion`, and
  make capture tolerate partial runs and bare resume contexts
- Validate `contextMeta.fading` in `commitAgentEventActorState` with a
  shared `isAgentFadingTier` guard in data-schemas, so a malformed tier is
  rejected at the DB layer instead of silently cold-starting the actor
- Reuse `IAgentEventActorContextMeta` for `IMessage.contextMeta` instead of
  extending an inline duplicate

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 test: Cover Resume Seeding From Paused Context Meta

Assert the resume controller seeds the rebuilt client from the context meta
captured at the pause before rebuilding the run, and move the capture
helper's JSDoc back above its function.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧹 style: Sort Imports in Fading Tier Plumbing

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Persist Fading Tier Through Redis Jobs and Branches

Address Codex and Copilot review findings on fading tier persistence:

- Deserialize `contextMeta` in `RedisJobStore`, whose explicit read mapper
  rebuilt every other pause field but this one, so Redis-backed
  deployments resumed with no tier to seed after every HITL pause
- Copy server-private `contextMeta` onto the assistant message created by
  `POST /api/messages/branch`, so the next turn seeds its pruner the same
  way it would from the parallel source response
- Constrain the `fading` subdocuments in both Mongoose schemas (version
  enum, positive budget, required fields) to match the zod schema
- Throw a distinct "fading tier is invalid" error from both context meta
  validators instead of reporting a calibration failure

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Carry Context Meta Through Stop and Re-Pause

Addresses the second Codex pass on the fading-tier persistence:

- A re-pause whose resumed segment has nothing persistable now clears the
  job's `contextMeta` in the same transition instead of leaving the first
  pause's calibration and tier for the next resume to seed from.
- The client publishes the run's live calibration and fading tier onto the
  job after each pre-invoke context snapshot (deduplicated on value), and
  the Stop path copies `jobData.contextMeta` onto the stopped response, so
  a follow-up from a stopped turn keeps the prompt prefix stable.
- Mid-run captures (HITL pause, Stop) read the graph's live state; `Run`
  refreshes its own getters only after the stream settles, so the pause
  used to persist the seeds it started from.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Fence Context Meta Publication Ahead of Stop

A Stop handled by another replica reads the job before the owner is
signaled, so a fire-and-forget publish could lose the race, and a run
stopped before its first context snapshot never published the tier it
inherited. The client now publishes the inherited seed before the run
streams or resumes, and the context-usage handler awaits each
snapshot's publish so the write lands before the model call it
describes begins. Failures still only log; the pre-run publish is
optional-called so bare resume contexts keep working.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Read Back Context Meta After the Abort Claim

The run owner publishes its context meta ahead of each model call and
awaits the write, so a publish landing between the abort's initial job
read and its terminal claim describes the call whose partial output the
abort snapshot carries. abortJob now reads the same-epoch job back after
the claim and content re-read and takes its context meta, so the stopped
response persists the tier that produced its bytes. The refresh is
best-effort and logs on failure like the content refresh beside it.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Share In-Flight Context Meta Writes and Clear Neutral Live State

Two snapshot callbacks in a parallel-agent run can capture the same tier
before the first write settles; the publisher now caches the in-flight
promise per serialized value so the second caller awaits the same
durable write instead of treating an uncommitted value as published.
A live snapshot with nothing left to persist after an earlier publish now
writes a neutral record (ratio 1, no tier), since a running job's fields
cannot be deleted through the metadata writer, so a Stop no longer
persists an earlier non-neutral state onto the response. The pre-run
seed publish is unchanged and is told apart from live snapshots by a
flag from the context-usage sink.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 chore: Drop Unused Rest Parameter in Context Meta Test

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 feat: Persist Per-Agent Fading Tiers as Compact Context Meta

Reconciles the host with @librechat/agents at 9aacad1a, where graph
history stays canonical and each Run derives a provider-only projection
from the latched tiers. LibreChat therefore persists only compact state:
the calibration ratio, the default agent's tier, and now the per-agent
tiers from Run.getFadingTiers(), stored as validated entries so agent IDs
never become MongoDB field names and restored onto a null-prototype
record for RunConfig.fadingTiers. Truncated messages, canonical tool
content, projection provenance and per-message truncation state are
never persisted; the SDK's latched flag is stripped on capture.

The per-agent map travels everywhere the single tier already did:
response contextMeta, event-actor checkpoint state, HITL pause and
resume job metadata in both stores, the Stop path, and branch creation.
A shared Mongoose definition keeps the message and conversation schemas
identical, the zod schema validates entries, and commitAgentEventActorState
rejects malformed maps. Comments no longer describe graph messages as
truncated; only the provider projection is.

Tests cover: tool results and inputs reach the persisted content parts
at full size; contextMeta carries exactly the compact tier and
calibration fields; a persisted snapshot round-trips into the seeded
RunConfig unchanged and prototype-safe; per-agent tiers survive pause,
Redis job round-trips, and event-actor context validation.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Restore Event Actor Types and Publish Seeds Before Abortable Setup

The previous commit's doc rewording in the conversation types dropped the
event-actor interfaces that precede the fading tier, which broke the
declaration bundle in CI; they are restored unchanged.

The inherited context meta is now published onto the job as soon as the
parent's state is loaded and awaited at the top of chatCompletion and
resumeCompletion, ahead of run creation and the other abortable setup
stages, so a Stop during setup still persists the parent's tiers onto the
stopped response. The partial response saved when every subscriber
disconnects now carries the job's context meta like the Stop and pause
paths do. The branch route keeps the source's context meta in the saved
message but strips it from the HTTP response, matching client reads.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Serialize Distinct Context Meta Publications

Overlapping snapshots from parallel agents could carry different tiers;
each started its own write and both job stores keep whichever write
finished last, so an older snapshot could overwrite a newer one. Each
distinct publication is now issued after the previous one settles, in
call order, so the newest snapshot is what a Stop reads back. Equal
values still share one write.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Project Root Message Reads and Keep Inherited Meta on Setup Failure

The root GET /api/messages paths returned raw documents: the
single-message query read called getMessages without a projection and the
cursor page performed an unprojected find. Both now apply
CLIENT_MESSAGE_SELECT, with getMessagesByCursor accepting a select option,
so the persisted fading tiers never reach the client through the
paginated endpoint; the search-hydration read projects the same way.

A turn that loads a parent's context meta but fails before its run exists
used to replace the inherited meta with the capture of a missing run, so
the persisted error response lost the last valid tiers. Both finalizers
now keep the inherited meta when no run was created while a created run's
neutral state may still clear it.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 refactor: Move Context Meta Publication Into packages/api

The publication coordinator (ordering, deduplication, in-flight sharing,
failure state) now lives in packages/api as createContextMetaPublisher,
with selectRunContextMetaToPublish deciding what a publication carries;
the legacy client keeps only the wiring to the job metadata writer. A
transiently failing write is retried with a short backoff before the
publication is reported and forgotten, so a snapshot's model call no
longer proceeds on the first rejection while the job holds the previous
record. The SDK swallows handler errors, so retrying is the only way to
strengthen the fence without failing the turn.

A terminal save now supplies contextMeta: null when the finished run has
nothing to carry, and saveMessage unsets a previously stored value in
that case, so a disconnect snapshot's record cannot outlive the run that
completed neutrally.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Widen the saveMessage Method Type for the Null Unset

The exported method interface still typed contextMeta as the message
field, so the spec exercising the null unset failed typechecking; it now
mirrors the implementation signature.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Publish Every Agent's Snapshot, Sanitize Search Hits, Unbound Agent IDs

The context-usage handler now awaits the run's context meta publish for
every ON_CONTEXT_USAGE event, hidden sequential agents included, while
the visible-snapshot bookkeeping and the client emission keep their gate;
a Stop before the next visible snapshot finds the tier a hidden agent's
model call latched. Search results are built from the hydrated hit minus
its server-private contextMeta, since Meilisearch hydration projects every
schema field. The per-agent tier entry guard no longer caps the agent ID
length: ephemeral agent IDs encode endpoint, model and sender without a
bound, and dropping such an entry would let that agent alone re-derive
its tier on the next turn. Also formats the publication spec so the
static checks pass.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Unset Context Meta Atomically, Keep Publication State After a Failed Write

- `saveMessage` folds the `contextMeta` unset into the same update that
  persists the terminal response, on both the plain and the provenance-merge
  paths, so a failure between two writes can no longer leave a completed row
  carrying a disconnect snapshot's state.
- `createContextMetaPublisher` tracks whether any record has committed
  separately from the retryable latest publication, so `hasPublished` stays
  true after a later publication exhausts its retries and a following neutral
  snapshot still overwrites the earlier record on the job.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Carry Context Meta Through the Resumable Stop Route and Neutral Resumed Completions

- The resumable Stop route copies the refreshed `jobData.contextMeta` onto the
  terminal assistant message it persists, as the abort middleware already
  does, so a stopped resumable response keeps the tiers that produced its
  partial output.
- A resumed run that completes with neutral context state now saves
  `contextMeta: null`, which unsets the paused segment's record instead of
  letting an omitted field keep it on the completed row.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Trust Only Server-Authored Context Meta, Unset It on a Neutral Stop, Read It From the Job on Disconnect

- `POST /api/messages/:conversationId` and conversation imports drop any
  client-supplied `contextMeta`, and the run seeds its calibration and fading
  tiers only from a server-authored parent response, so a forged tier can never
  shape a provider projection.
- Both Stop paths save `contextMeta: jobData.contextMeta ?? null`, so a job that
  re-paused with neutral state unsets what an earlier pause stored on the row
  instead of leaving it behind.
- The disconnect partial save reads the same-epoch job record for the run's
  published context meta, since the client-facing resume snapshot never carries
  server-private state.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧪 test: Mock the Job Store in the Resume Metadata Spec

The disconnect partial save now reads the same-epoch job record through
`GenerationJobManager.getJobStore()`, so the resume metadata spec's manager
mock provides it as the other disconnect spec already does.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 🧊 fix: Keep Context Meta Off the POST Message Response

`POST /api/messages/:conversationId` returned the saved row verbatim, so a
client write against an existing server-authored message echoed the row's
stored context meta. The route now projects the row through the same client
projection the branch route uses, which drops the server-private field.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

* 📦 chore: Bump @librechat/agents to 3.7.16 for Persisted Fading Tiers

3.7.16 is the first release that carries danny-avila/agents#497, so
`Run.getFadingTier()` / `getFadingTiers()` now report the latched tier
this branch persists and `RunConfig.fadingTier` / `fadingTiers` seed the
next run from it. The dependency set is unchanged from 3.7.15; only the
package entry moves.

Adds an SDK-backed round trip to the fading spec: a first pruner latches
an informative tier, `resolveRunContextMeta` reduces it to the persisted
shape, and a second pruner seeded from that shape reproduces the first
run's projected tool-exchange bytes under instruction and calibration
drift, while an unseeded pruner under the same drift relaxes its tier and
rewrites them.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WNbgMeR2aZDyhsqAy4L8gF

---------

Co-authored-by: Claude <noreply@anthropic.com>
2026-09-02 20:29:43 -04:00
Danny Avila
0219513115
🛑 fix: Forward Run Abort Signal to Foreground Tool Calls (#15517)
* 🛑 fix: Forward Run Abort Signal to Foreground Tool Calls

Stopping a generation left in-flight tool calls running on the far side.
For MCP over streamable-http the external server kept burning CPU until
the tool finished or timed out, with no `notifications/cancelled` ever
sent.

The MCP layer was never the problem: `MCPManager.callTool` already
spreads `options.signal` into `client.request`, and the SDK's
`Protocol.request` sends `notifications/cancelled` the moment that signal
aborts. The signal simply never arrived.

Agent tools do not execute inside LangGraph's ToolNode — `eventDrivenMode`
dispatches `on_tool_execute` and `createToolExecuteHandler` invokes them.
The agents SDK puts the run's abort signal on the batch request
(`ToolExecuteBatchRequest.signal`, documented as one handlers should
forward), but the handler destructured every field except `signal` and
built its invoke config without it. `config.signal` was therefore
`undefined` in every foreground tool, so `createMCPTool` derived no
signal and the SDK registered no abort listener.

Forward it. This restores cancellation for all signal-aware foreground
tools, not just MCP. The detached background invoke keeps its own
controller, since that work deliberately outlives the turn.

An aborted call now rejects promptly and resolves an error result rather
than hanging, so the abort is logged at debug instead of as a tool error
— a user pressing Stop should not spray the error log.

* 🔇 fix: Keep Cancelled Tool Calls on the Filtered Path and Out of Error Logs

Addresses both findings from the Codex review of 48fded564a.

P1 — the cancellation branch returned before `filteredToolOutputResult`,
so with tool-output filtering configured an aborted call's error text
reached the turn uninspected. Worse, `runSignal.aborted` says the run is
over, not that this rejection was the cancellation: an unrelated failure
racing the Stop took the same unfiltered exit. Cancellation now only
selects the log level; filtering and the result shape are unchanged.

P2 — `createMCPTool` logs its own error before wrapping and rethrowing,
so every Stop during an MCP call still emitted an error-level MCP entry
and could pollute operational alerts. That catch now recognizes the abort
and logs at debug; the wrapped message still reaches the turn.

Both are covered by tests that fail against the previous commit.

* 🎯 fix: Require an Abort-Shaped Error and Spare Shared OAuth Flows

Addresses both findings from the Codex review of b3c67082cd.

An aborted run signal proves the turn is over, not that the rejection in
hand was the cancellation — a permission, OAuth, or upstream failure can
reject in the same tick a user presses Stop. Both quiet-log branches now
require the error to look like an abort as well, so a real failure racing
the Stop stays at error level and visible to operational alerts.
`isAbortError` moves to `@librechat/api` and the copy in
`abortMiddleware` is retired, keeping one implementation. Its spec pulls
the real function through its partial package mock rather than asserting
a stub.

Forwarding the run signal also reached ActionService, where two of the
three OAuth flows are keyed `userId:action_id` and so outlive the run
that opened them: a second run for the same action joins the very same
record, and the browser's OAuth callback reads its metadata to exchange
the code. `monitorFlow` deletes the key when a waiter's signal aborts, so
one Stop could strand a concurrent run and discard an authorization the
user had already granted. Those two flows no longer take the signal; the
run-scoped login flow, keyed by thread and run, still does.

Every guard is covered by a test that fails when the guard is removed.

* 🧷 fix: Detach the Stopped OAuth Waiter Instead of Dropping Its Signal

Addresses the Codex review of d53ff170ab.

Withholding the run signal from the shared Action OAuth flows protected
concurrent runs and the browser callback, but left this invocation
attached to the flow after the Stop. A callback landing late then resumed
`_call` straight into `preparedExecutor.execute` — a consequential API
request running after the user had already stopped the turn. Handing the
signal back is not the answer either, since `monitorFlow` deletes the
shared `userId:action_id` key on abort.

`detachOnAbort` separates the two: the caller stops waiting the moment
the signal aborts, while the shared flow runs on for whoever else needs
it — another run, or the callback exchanging its code. Late settlement of
the detached work is swallowed rather than surfacing as an unhandled
rejection.

Both shared flows now use it. The run-scoped login flow keeps the signal
directly, as nothing outside its run observes it.

* 🪪 fix: Keep an Action OAuth Abort an Abort

Addresses two of the three findings from the Codex review of 70fc0782b1.

Detaching the stopped waiter left the rejection to be misread on the way
out. The refresh catch treated it as a failed refresh and called
`requestLogin()`, emitting an OAuth prompt and opening pending
authorization state for a turn that had already ended. `requestLogin` and
the surrounding auth catch relabelled it `Failed to authenticate OAuth
tool` / `Authentication failed`, and the outer catch handed that to
`logAxiosError` — so an ordinary Stop produced two error-level entries
and returned failure text as the tool's result. Each boundary now lets an
abort through unchanged, and the outer catch logs it at debug and
rethrows so it stays a cancellation end to end.

Deferred: waiter-only cancellation in `FlowStateManager`, so a detached
`monitorFlow` stops polling instead of running out the flow's TTL. That
poller is bounded at three minutes and lives exactly as long as it does
for a waiter that never left, so it changes no lifetime this branch
introduced; giving `FlowStateManager` a non-destructive abort mode is its
own change, shared with MCP and indexSync.

* 🤫 fix: Classify the Abort at the MCPManager Boundary Too

Addresses the Codex review of 9bcf1aa57e.

`MCPManager.callTool` logs every rejection at error level before
rethrowing to `createMCPTool`, so a user Stop still emitted an MCP error
from the inner boundary even after the outer one learned to recognize it.
The same signal-plus-abort-shape classification now applies here.

The review also noted, correctly, that the existing test could not have
caught this: it replaces `getMCPManager` wholesale, so it only ever
exercised the outer logger. The new cases drive the real manager with a
connection whose `client.request` rejects the way the SDK does once a
request signal aborts, and cover both the cancellation and a genuine
failure racing the Stop.

* 🧹 chore: Restore Import Order in the Agents Tool Handler
2026-09-02 13:32:33 -04:00
Danny Avila
1c7a172321
🕹️ feat: Configure BYOM Tool Permissions (#15504)
* feat: add configurable BYOM permissions

* test: cover BYOM permission settings live

* fix: harden BYOM permission settings

* perf: reuse principals for environment discovery

* chore: normalize BYOM settings style

* fix: Align BYOM Discovery and Admission

* fix: Harden BYOM Permission Boundaries

* fix: Align BYOM Skill Approval Routing

* fix: Preserve Lazy BYOM Approval Capabilities

* perf: Parallelize Cached Environment Reads
2026-09-02 10:38:00 -04:00
Danny Avila
6372cb08ee
🧶 fix: Untangle Parallel Subagent Descriptor Paths (#15492)
* fix: isolate parallel subagent descriptor walks

* fix: preserve canonical subagent config identity

* fix: separate subagent graph assembly from path validation
2026-09-01 21:48:15 -04:00
Danny Avila
a3c2277c38
🚧 fix: Narrow Chat MCP Selections to What the Picker May Offer (#15478)
* 🚧 fix: Narrow Chat MCP Selections to What the Picker May Offer

The chat MCP picker offers a server only when the operator has not hidden it
(`chatMenu: false`) and the user reaches it directly rather than through an
agent (`consumeOnly`). Nothing enforced either condition on the request, so a
stale client, a replayed body, or a hand-written one could name any server and
get its tools — `consumeOnly` in particular is a permission the registry derives
per user, so ignoring it let a user reach in plain chat a server they had only
been granted through an agent.

Narrow the picker's selection in the two ephemeral loaders, against the same
predicate the client's `selectableServers` uses. The resolver is injected, so
`packages/api` stays free of the registry and the loaders keep their existing
`deps` shape.

Provenance is why this lives in the loaders rather than at the tool-resolution
chokepoint: only here is a server name known to have come from the picker.
Servers a model spec pins are the operator's own choice and are added after the
narrowing, so pinning a chat-hidden server to a spec keeps working.

`chatMenu` only ever comes from the config tier, so the overlay settles it with
no lookup at all. Lookups for the rest run concurrently and are cached per
server and user. A name the registry cannot resolve — request-tier, declared on
the body and never registered — is kept, as is one whose lookup fails: this
narrows an already-authenticated selection and is not the authorization
boundary.

* 📝 docs: Say What chatMenu Now Bounds

The schema still described `chatMenu` as dropdown visibility alone, which is
misleading now that it also bars a server from the chat selection a request
carries. Spell out both halves, and what it deliberately does not restrict:
agents, and servers a model spec pins.

* 🛡️ fix: Close Four Gaps in the Chat MCP Narrowing

Dedupe before resolving. The selection is request-supplied and the loaders used
to build a `Set` immediately; mapping the raw array first let one body repeat a
name enough times to launch a resolver — registry, cache, and ACL reads, with no
single-flight — per occurrence.

Resolve names through the tool-loading aliases. A configured `private server` is
addressable as `private_server` downstream, so an exact-key lookup let the
normalized spelling walk past a `chatMenu: false` check and still load tools.
Names are returned as sent; only the decision is canonicalized.

Let the request overlay settle `chatMenu`. The config tier is where the flag
lives and is what the picker itself was built from, so an overlay that re-enables
a server now outranks the durable config the registry returns, which was silently
dropping a valid selection. `consumeOnly` still comes from the registry.

Narrow the request body, not just the loader's copy. `applyContextToAgent` reads
`req.body.ephemeralAgent.mcp` directly and prefers it over the agent's tools, so
a hidden server's `serverInstructions` reached the model context even after its
tools were dropped.

* 🧭 fix: Ask the Registry Which MCP Servers the Picker May Offer

Three rounds of review found the same mistake in three disguises: this narrowing
re-derived the registry's tier precedence instead of asking for it. A config
entry was treated as authoritative over a user-DB or process-backed server of the
same name; the alias map was built from config names only, so a DB server named
`private_server` was misrouted to a config `private server`; and the overlay
short-circuit ran before any of that could be consulted.

Resolve the accessible set through `resolveAllMcpConfigs` instead — the same call
behind `GET /api/mcp/servers`, so what the request may use and what the picker
was offered come from one place and cannot drift. Precedence, cross-tier name
collisions, and the config overlay are the registry's job again, and the
selectability test is a single flag check over what it returns.

The result is smaller: no overlay branch, no per-server lookups, one catalog read
regardless of how long the selection is, and aliases built over the whole
accessible set so an exact name always outranks another server's normalized form.

Publish the servers actually in play back onto the request body, spec pins
included. `applyContextToAgent` prefers that list over the agent's tools, so
narrowing to the picker selection alone had started dropping a spec-pinned
server's `serverInstructions` while suppressing a hidden one's.
2026-09-01 18:47:36 -04:00
Danny Avila
51edb2a660
🛫 feat: Enable Self-Service Code Worker Pairing (#15367)
* feat: add self-service code worker pairing

* fix: enforce code worker principal boundaries

* fix: separate pairing control planes from execution

* fix: preserve principal worker routing

* fix: pin artifact traffic to code workers

* fix: authenticate code worker selection

* test: mock worker-aware artifact headers

* fix: harden self-service code worker routing

* fix: fail closed for principal code pairing

* style: format code execution headers

* fix: close self-service pairing gaps

* fix: satisfy code environment type checks
2026-09-01 18:46:33 -04:00
Oliver Faust
5e49868f0a
🔑 fix: Decode Action OAuth Credentials After Decryption (#14698)
* 🔑 fix: Decode Action OAuth Credentials After Decryption

Action client credentials are URL-encoded before encryption, but the OAuth
token exchange decrypted them with `decryptV2` alone, sending values like
`abc%40domain.com` to the provider and causing 401s for secrets containing
`@`, `+`, `=`, `/` or `:`. Decoding falls back to the raw decrypted value so
legacy secrets stored before 299cabd6e (which contain an unescaped `%`) still work.

Fixes #14636

* 🔑 fix: Share Action Credential Codec Between Read Paths

The encode-before-encrypt convention had two readers with different
behaviour. `ActionService.decryptSensitiveValue` decoded unconditionally,
so a credential stored before encoding was introduced (299cabd6e, March
2025) containing a stray `%` threw `URIError` and failed every read of the
action, not just the token exchange.

Extract the pair into `packages/api/src/actions/crypto.ts` as the single
implementation and consume it from both `ActionService` and the OAuth token
exchange, so the two paths cannot drift again.

Tests cover the codec against real crypto: reserved characters and literal
percent escapes round-trip, pre-encoding values are returned unchanged, and
a pre-encoding value with a stray `%` no longer throws.

A version marker on newly encrypted values was prototyped and dropped: it
changes the stored format for every action credential while fixing nothing,
since values written by the encoding convention already decode correctly
and pre-encoding values stay unmarked either way.

---------

Co-authored-by: Danny Avila <danny@librechat.ai>
2026-09-01 18:42:28 -04:00
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
dccd7ae496
🧶 fix: Preserve Multi-Agent Tool Call History (#15468)
* 🧭 fix: Preserve Multi-Agent Tool History

* 🧰 fix: Preserve Lazy Subagent Tool History

* 🧪 test: Enable Lazy Code History Fixture

* 🧭 fix: Derive Effective Lazy Tool History

* 🧪 fix: Tolerate Empty MCP Name Audits

* 🧪 test: Mock Lazy MCP Name Audit

* 🧭 fix: Expand Lazy Historical Tool Names

* 🧱 fix: Derive Durable Wildcard History

* 🧬 fix: Traverse Nested Wildcard History

* 🔭 fix: Resolve Wildcard Server Boundaries

* 🔐 fix: Persist MCP Call Identity

* 🪪 fix: Stamp Resolved MCP Execution Identity

* 🕸️ fix: Resolve Graph Member MCP Identity

* 🧭 fix: Fall Back From Empty Graph Member ID

* 🧭 fix: Preserve Aggregated MCP Identity

* 🧭 fix: Harden MCP Identity Resolution

* 🧭 test: Assert Concurrent Graph Limit Invariant

* 🧭 fix: Centralize Lazy History Policy

* 📦 chore: Bump `@librechat/agents` to v3.7.13

* 🧪 test: Align Content Filter Fixture With SDK History
2026-09-01 18:37:45 -04:00
Danny Avila
c8939c9730
🪄 fix: Reuse Downloaded Artifacts for Image Reads (#15454) 2026-09-01 10:55:53 -04:00
Danny Avila
8f6228095f
🖼️ fix: Stop Reporting a Failed Sandbox Image Read as an Unreadable Format (#15453)
* 🖼️ fix: Stop Reporting a Failed Sandbox Image Read as an Unreadable Format

`read_file` on a `/mnt/data` image collapsed every failure — rate limit,
missing path, truncated transfer, transport error — into the binary-file
hint, whose image branch reads as a permanent capability limit ("is an
image file and cannot be read as text"). The model stopped retrying, and
the operator had no way to tell which failure had actually occurred.

The failure that surfaced it: the reader windows the bytes through
`/exec` stdout, one call per window, against the Code API's per-user
execution limiter (20 requests / 30s in the reference deployment). At a
fixed 32KB window a 122KB chart cost 4 calls and a 1MB image cost 33 —
more than a whole limiter window, so anything over ~640KB could never
complete. A turn that generates several charts and reads them back
exhausts the budget partway through and reports the 429s as unreadable
images.

- `handlers.ts`: failed reads name the real cause and the affordance that
  matches it — rate-limited waits and retries, a missing path lists the
  directory, a truncated transfer retries. The "cannot be read as text"
  phrasing is now reserved for bytes that are not a supported image. An
  image past the inline budget says to downscale and read the smaller
  copy rather than pointing at `bash_tool`, which cannot show an image.
- `process.js`: the window is derived from the runner's stdout budget
  (48,960B vs 32,768B, so a 122KB chart is 3 calls and 1MB is 22), halves
  and re-reads the same offset when a runner truncates anyway (remembered
  per Code API, with the env remedy logged), waits out a rate limit it
  can afford instead of discarding the bytes already pulled, and stops at
  a 24-call ceiling so no read stalls a turn across limiter windows.
- `utils/code.ts`: shared `getCodeApiRetryAfterMs` (header, then the
  `retry_after_seconds` body); `skillFiles.ts` drops its private copy.

* ♻️ refactor: Move the Sandbox Image Reader Into TypeScript

Addresses the review round on #15453.

- P1: windowing, window sizing, the in-sandbox reader script, response
  parsing and byte assembly move to `packages/api/src/files/code/image.ts`,
  and the Code API rate-limit policy to `packages/api/src/utils/code.ts`.
  `process.js` keeps only the `/exec` transport it owns. Both new modules
  are tested against their real implementations (19 + 13 cases), including
  the round-trip ceiling, the changed-file guard and the shell-injection
  guard on the reader script.
- P2: a configured stdout budget below the old 8KB window floor produced a
  window the runner could never emit, and narrowing was pinned to the same
  floor, so no retry could succeed. The window now aligns to the base64
  group size instead, and the round-trip ceiling bounds what a small budget
  costs.
- P2: `.env.example` documented a 32,768-byte default that no longer holds
  and never mentioned the stdout budget. It now describes
  `LIBRECHAT_CODE_SANDBOX_OUTPUT_MAX_SIZE` as the setting to tune, with the
  chunk override as the escape hatch.
- P3: `process.spec.js` no longer reimplements the retry parser in its
  module mock. Behaviour is tested where it lives; the mock is a declared
  passthrough and the spec covers what the transport still owns — session
  forwarding, the profile header, parser hand-off and error logging.
- P3: `getCodeApiRetryAfterMs` reads the response body through `in`
  narrowing rather than an `as` assertion.
- P3: the window-sizing JSDoc no longer claims a multiple of 3 in the one
  case that is not — an explicit `LIBRECHAT_CODE_IMAGE_CHUNK_BYTES`.

* 🎯 fix: Size Sandbox Image Reads to What the Limiter Can Deliver

Addresses review round 2 on #15453.

- Narrowing now halves the caller's own failing window instead of the
  shared learned value, and the learned value only moves down to the
  narrowest size any read needed. Reads run concurrently (one per tool
  call), so two reads failing at the same size were compounding into a
  window two steps smaller than the runner would have accepted, and
  persisting it for every later read.
- Narrowing continues until a runner accepts a window or a floor below
  which no runner could serve an image at all. Two halvings stopped at
  12,240 bytes — still ~16KB of base64 — so an 8-10KB stdout cap that was
  not configured up front stayed unreachable.
- The round-trip ceiling is now one limiter window (20), and a read that
  cannot finish within it says so from the FIRST response rather than
  spending the whole budget rediscovering it: the first window reveals
  both the file size and what a call really delivers. A near-limit image
  no longer drains the turn's executions before failing, which is what
  made the wait budget look too small.
- The round-trip message names the size this deployment can actually
  inline, so the model has a downscale target instead of a guess.
- `python3: not found` from the sandbox reader was matching the generic
  "not found" in `isSandboxMissingFileError` and being reported as a
  missing image path, sending the model to `ls /mnt/data` while hiding a
  runner dependency. Image reads now classify only path-specific ENOENT
  forms as a missing file.

* 📐 fix: Size a Truncated Read's Retry From the Cap It Revealed

Addresses review round 3 on #15453.

A ~1KB sandbox stdout cap is a real runner default, and the 1,536-byte
narrowing floor sat above it: discovery stopped one step short and
declared every image unreadable, even though small ones fit comfortably
in the 576-byte window the configured path already derives.

Narrowing now runs until no smaller window exists. It also no longer has
to grope its way down: a truncated response emitted exactly as much as
the runner allows, so its length measures the cap this deployment never
declared, and the retry is derived from it directly — one round-trip
instead of the eight blind halvings it takes to fall from ~49KB to ~1KB.
Blind halving remains the fallback and bounds the result either way, so
every retry is strictly smaller than the window that just failed.
2026-09-01 09:08:01 -04:00
Danny Avila
224512851a
🗂️ fix: Resolve Code Interpreter Destination Collisions (#15449)
* 🗂️ fix: Resolve Code Interpreter Destination Collisions

Codeapi mounts each `/exec` input at a destination derived from its `name`
and rejects the whole request when two entries land on one. LibreChat never
enforced that: `primeFiles` sent `name: file.filename` verbatim, and the
graph seed deduped by `(storage_session_id, id)` — identity, not destination.

Two conversation files sharing a filename therefore killed code execution for
the rest of the conversation. The rejected request never reaches the sandbox,
so no result comes back to collapse the pair, and every later turn re-primes
both and fails identically. Only code-generated outputs are covered by the
`(filename, conversationId, context, tenantId)` partial unique index; uploads
carry `context: message_attachment`, so one conversation can hold several
records sharing a filename — two uploads of `image.png`, or an upload whose
name a later execution wrote back in place.

`packages/api/src/files/code/destinations.ts` now owns the name → destination
mapping for every contributor to an `/exec` file list, mirroring codeapi's
conflict rule including directory-prefix conflicts:

- `primeFiles` claims per pushed file, disambiguating with a `-2`, `-3`
  counter and echoing the resolved path into the tool context, so the model
  is told where each file actually mounts. Claim order comes from immutable
  `createdAt`, not `getFiles`'s `updatedAt` sort, which usage accounting and
  re-upload both bump — order-dependent destinations would silently repoint
  a path an earlier turn had already handed the model. Newest wins the bare
  name, matching `ToolNode.updateCodeSession` and giving an in-place rewrite
  precedence over the original it superseded.
- `seedCodeFilesIntoSessions` drops a later file whose destination is already
  claimed. Renaming there would invent a path nothing told the model about,
  and a conflict at that layer means one logical file arrived twice under
  different storage pointers.

Fixes #15443

* 🧹 chore: Tolerate Holes in the Destination Priority Sort

`primeFiles` sorts the raw `getFiles` result before its own per-entry guard
runs, so the comparator has to survive a null row rather than throw ahead of
the loop that was already defending against one.

Also tightens the tool-context assertion in the upload-vs-output case, which
was passing on a trailing space rather than on the path.

* 🔁 fix: Address Codex Review on Destination Assignment

Four findings from the review of `b0f61d0`.

**Identity-derived suffixes instead of a collision counter.** A counter is
assigned from whatever is in the set, so a third file arriving under the
literal name a displaced file had been given pushed that file along, and code
the model wrote in an earlier turn would silently start reading the newcomer.
The fallback now hashes the file's own `file_id`, so a displaced file keeps
its destination for the life of the conversation and only the winner of the
bare name can change. A trailing counter still guards a hash collision so the
search terminates on any input.

**Rank by last content write, not record creation.** `processCodeOutput`
claims one row per `(filename, conversationId)` and rewrites it in place, so
a repeatedly written output keeps its original `createdAt` while recording the
latest writer in `metadata.sourceDispatchedAt`. Ranking on `createdAt` alone
handed the bare path to an upload the output had since been rewritten over.
Recency is now `max(sourceDispatchedAt, createdAt)` — still never `updatedAt`,
which usage accounting bumps without any content changing.

**Advertise a displaced generated output.** Generated files are left out of
the tool context because the model already knows what it wrote — which stops
being true once a newer same-named file takes the bare path and the output
mounts somewhere the model has never seen. Displaced outputs are now listed
with the name they were written as, and displaced uploads name their original
too.

**Rank conversation files above an agent's private ones.** Each agent claims
destinations over its own candidate set — the conversation's files plus its
own — so a private file taking a bare name from a shared file in one agent but
not in another left two agents advertising different paths for the same file
into one shared mount namespace. Shared files now rank first, so the majority
that every agent sees resolves identically. The private-versus-private case
still drops at the merge and is documented there; closing it needs one
assignment across contributors, made before any agent renders its context.

* 📌 test: Pin Alias Ownership on an In-Place Rewrite

Codex round 2 flagged that the identity alias still falls back to a counter
when the alias itself is literally claimed, so a displaced file can move after
all. The mechanism is real and my JSDoc overstated the guarantee, which this
corrects.

The behaviour stays as it is. The realistic way a file arrives under another
file's alias is the model rewriting the displaced file in place: it was told
`<stem>-<hash>.ext`, wrote back to it, and `processCodeOutput` registered an
output under that literal name. Keeping the alias with the newest holder is
what makes the model's familiar path resolve to its own edit; reserving it
against the newcomer would hand that path back to the bytes it replaced. The
test pins that ordering so a later change cannot invert it silently.
2026-09-01 00:55:54 -04:00
Danny Avila
49e1e32f3a
🧺 fix: Funnel MCP Catalog Work Through a Shared Gate (#15429)
* 🧯 fix: Bound MCP Catalog Work Across Requests

* 🛑 fix: Reject Saturated MCP Discovery

* fix: fairly bound MCP catalog work

* test: type catalog snapshot mock arguments

* test: assert snapshot deadline forwarding

* fix: harden MCP catalog scheduling
2026-08-31 14:13:31 -04:00
Danny Avila
b7ae67307e
🕰️ fix: Anchor Temporal Variables to Turn Start (#15419) 2026-08-31 10:38:10 -04:00
Danny Avila
413906abc3
🖍️ fix: Honor Agent Image Detail Settings (#15407)
* 🖼️ fix: Apply an Agent's Configured Image Detail

The Agent Builder exposes an Image Detail slider and saves it to
`model_parameters.imageDetail`, but nothing ever read it back: `encodeAndFormat`
resolves the level from `req.body.imageDetail`, which the agents route never
populates, so every image was encoded as `detail: "auto"`. Providers that reject
`auto` — MiniMax's OpenAI-compatible API among them — fail every image upload
through an agent, and the setting that would fix it has no effect.

The root cause is a classification error rather than a missing argument.
`extractLibreChatParams` splits LibreChat's own settings out of an agent's model
parameters, and its key set is derived from the `librechat` group in
`parameterSettings`. `imageDetail` sat in `baseDefinitions` instead, so it was
neither extracted for LibreChat's use nor stripped from the options handed to the
provider.

It is now grouped with `resendFiles` and the other LibreChat-level parameters,
which gives it the same path those already take: extracted during agent
initialization, carried on the initialized agent, passed to the client, and read
by `encodeAndFormat`. `AgentClient.getSaveOptions` already expected
`options.imageDetail` — nothing had ever populated it.

The request body still wins where it is set, so the classic endpoints keep
resolving detail from conversation settings exactly as before.

Fixes #15359

* 🔤 fix: Type Image Detail as Its Enum, Not a Bare String

`TConversation.imageDetail` is the `ImageDetail` enum, so the extractor's return
type and the initialized agent's field are now declared as that enum rather than
widened to `string`, and the tests pass enum members instead of raw literals.

Caught by CI's type check, which I had run before adding the tests and not after —
jest compiles the specs through babel, so the suite stayed green over a type error.
2026-08-31 08:34:05 -04:00
Danny Avila
c7f0a8f247
📬 feat: Move Agent Queued Turns Server-Side (#15370)
* feat: move agent queued turns server-side

* test: update queued turn composition coverage

* fix: harden queued-turn lifecycle ownership

* fix: centralize queued-turn lifecycle ownership

* fix: close queued-turn lifecycle gaps

* fix: unify queued turn lifecycle transitions

* fix: fence queued-turn deletion admission

* fix: preserve queued-turn lifecycle ownership

* fix: close queued-turn reconciliation races

* fix: reconcile quarantined queued-turn admissions

* fix: annotate admission evidence default

* fix: deepen queued turn lifecycle admission

* fix: fence queued turn provider admission
2026-08-30 21:53:58 -04:00
Danny Avila
fcae1025c0
🧳 feat: Register Principal-Owned Code Environments (#15365)
* feat: add principal-owned code environments

* fix: address code environment CI coverage

* fix: expose principal code environments in endpoint config

* fix: reject missing code environment bodies

* fix: harden code environment lifecycle

* fix: revalidate principal code environments

* fix: synchronize code environment authorization

* fix: bind code environments to current principals

* fix: fail closed on code ACL cache errors

* fix: prevent code environment override shadowing

* fix: preserve principal environment defaults

* fix: suppress revoked code environment aliases

* fix: fail closed on environment augmentation

* fix: narrow code environment defaults

* fix: isolate code environment fallback

* style: sort code config imports
2026-08-30 19:33:19 -04:00
Danny Avila
30124f21b2
🎻 refactor: Orchestrate Agent Runs Through a Request-Free Host (#15366)
* refactor: decouple agent initialization from HTTP

* refactor: centralize remote agent execution lifecycle

* fix: preserve pre-settlement error rendering

* fix: preserve request-backed tool loading

* style: sort agent execution imports

* fix: adapt public agent tool loaders
2026-08-30 15:38:57 -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
2ae6c8aea9
🛎️ feat: Wake Agents on Background Tool Completion (#15350)
* feat: wake agents for background tool completion

* fix: isolate background completion contracts

* fix: anchor background completion identity

* fix: close background completion delivery gaps

* fix: preserve manual background polling

* fix: preserve legacy tool group identity

* fix: close background wakeup identity gaps

* test: type background wakeup enqueue mock

* fix: preserve background completion identity

* test: type activity phase fixture

* test: mock phase media query

* fix: arbitrate background result ownership

* fix: retain tool step routing helper

* test: expand phase groups before identity checks

* fix: preserve background completion ownership

* fix: bound durable background results

* fix: type wakeup input budget export

* chore: sort background handler imports

* fix: persist timed-out background completions

* fix: reconcile background completion ownership

* fix: type background completion capabilities

* fix: harden background completion terminalization

* test: type missing completion evidence

* fix: require evidence for completion retirement

* chore: satisfy background completion static checks

* test: cover automatic background completion wakeups

* test: assert background wakeup agent identity

* fix: wake capability-fenced trigger deliveries

* test: keep capability shield fixture public

* test: assert capability worker wakeup

* fix: close background completion ownership gaps

* test: satisfy completion lease static checks

* fix: preserve artifact completion wakeups

* fix: close background delivery recovery gaps

* fix: simplify background receipt guidance

* fix: recover dead background completion batches

* fix: fence background completion recovery

* fix: recheck background recovery ownership

* fix: fence unpublished background continuations

* test: await background message hydration
2026-08-30 08:56:41 -04:00