* Reduce server memory usage and add container memory best practices
Profiling a full-service boot showed ~300MB RSS per service process, much
of it client-only dependencies pulled into the server module graph, and no
heap limits applied in containers:
- Import date-fns locales individually rather than via the locale index,
which loaded all ~90 locales into every process
- Load @sentry/react dynamically in insertFiles so the browser Sentry SDK
stays out of the server-side editor graph
- Replace class-validator isHexColor with the existing validateColorHex
in the Highlight mark
- Extract icon names into a standalone IconNames module so server-side
schema validation no longer imports Font Awesome icon packs, with a
test to keep it in sync with IconLibrary
- Require S3Storage lazily so the AWS SDK and native CRT binding are only
loaded when S3 file storage is configured
- Default each forked service process to a V8 heap limit derived from the
cgroup memory constraint, preventing several processes from together
committing more memory than the container allows
- Cap glibc malloc arenas in the Docker image to reduce resident memory
fragmentation in multi-threaded processes
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016YMX5ReGHyxrR4ZE5ZbaXa
* Require both file storage backends lazily for consistency
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016YMX5ReGHyxrR4ZE5ZbaXa
* Load AWS SDK via dynamic imports compatible with test environment
Requiring the storage TypeScript modules lazily broke under Vitest, which
cannot resolve require() of source files and does not apply vi.mock to
bare requires. Move the laziness into S3Storage instead: the module is
imported statically, and the AWS SDK packages are loaded with dynamic
imports on first use, including deferred creation of the S3 client.
Also address review feedback: handle rejection of the dynamic Sentry
import, and warn when the minimum per-process heap limit exceeds the
memory budget of the container.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016YMX5ReGHyxrR4ZE5ZbaXa
* Consolidate S3 SDK loading into a single lazy accessor
The commands are loaded together with the client in getS3() rather than
as separate dynamic imports per method. The global S3Client test mock is
now a class, as a mock function is not constructable when reached through
a dynamic import.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016YMX5ReGHyxrR4ZE5ZbaXa
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix: OIDC 502 on sign-in from oversized id_token cookie (#12986)
Storing the full OIDC id_token in the `oidcIdToken` cookie inflated the
sign-in response headers enough to exceed reverse proxy buffers, causing a
502 Bad Gateway. Store the id_token server-side in Redis keyed by a short
session identifier and keep only that identifier in the cookie, preserving
spec-compliant RP-initiated logout via `id_token_hint`.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix: Make OIDC logout token persistence best-effort
A Redis failure when storing or reading the logout id_token_hint should not
block sign-in or logout; fall back to omitting the hint instead.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs
* refactor: Extract portable LogoutTokenStore for provider logout hints
Move the cookie-referenced, Redis-backed logout token persistence out of the
OIDC router into a provider-agnostic `LogoutTokenStore`. Any auth provider that
supports provider-initiated logout can now persist and consume a logout hint via
`new LogoutTokenStore(providerId)` without duplicating the cookie/Redis/error
handling. The OIDC-specific end-session URL construction stays in the router.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* Use full keyspace for email verification codes
Email verification codes were generated with randomInt(100000, 1000000),
which can never produce a value below 100000. That excluded every code
with a leading zero — shrinking the keyspace from 1,000,000 to 900,000
and letting an attacker assume the first digit is never zero. It also
made the existing padStart(6, "0") dead code.
Draw uniformly from the full 0–999999 range and rely on padStart to
render leading zeros, restoring the full six-digit keyspace. Add tests
covering the six-digit invariant across many samples and the presence of
leading-zero codes.
* Simplify verification code jsdoc and drop leading-zero test
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EWfLg5uQdYFuzj1UBEo1og
---------
Co-authored-by: Claude <noreply@anthropic.com>
* perf: Keep vendor-prosemirror out of the eager authenticated bundle
Split the data-only helpers `getEmpty` and `isEmpty` out of
`ProsemirrorHelper` into a new prosemirror-free `ProsemirrorDataHelper`
module, and point the sidebar, DocumentNew, and other lightweight
callers at it. This removes the static edge from the authenticated entry
graph to `prosemirror-model`, deferring the ~653 KB vendor-prosemirror
chunk until an editor surface is actually opened.
Fixes#13005
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* Use non-optional collection.data in defined branch
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
The test asserted the global rollupPeriod return value was 0 while passing
the full UUID range, so activity from other tests sharing the database
(especially near the UTC midnight boundary) leaked into the window. Scope
the partition to the test's own document id to make the count deterministic.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* fix: Consistent usage of configured proxy with auth providers
* lint
* fix: Harden proxy agent helper against missing OAuth2 client
Degrade to a logged warning instead of crashing auth at boot if a
strategy's internal _oauth2 client is ever absent.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* feat: section-aware unfurl previews for heading links
When a link targets a specific heading (URL hash anchor), the hover
preview now shows the content of that section rather than the top of
the document.
Adds DocumentHelper.getAnchorContent which extracts the Markdown for the
section beginning at the heading matching a given anchor, spanning until
the next heading of the same or higher level. The urls.unfurl endpoint
passes the URL hash through to the document presenter for both internal
and public share links.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0185Tt87p63vsnAC8oGSLLRs
* fix: retain heading anchor in internal document unfurl url
The internal-document unfurl branch never passed a url to the presenter,
so the response fell back to document.url without the hash. Clicking the
hover preview then navigated to the top of the document rather than the
summarized section. Re-append the anchor in the presenter fallback so the
unfurl points at the same heading it previews.
Also supports the mention anchorId path: document mentions serialize their
anchorId into the href hash, which flows through this same path.
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix(mcp): return text content for empty list tool results
When MCP list tools return zero items, success([]) previously produced
an empty content array. Anthropic's MCP connector rejects that format.
Return a single JSON text block for empty arrays so clients can
distinguish no results from a broken tool response.
Fixes#12858
* test(mcp): handle empty list sentinel in integration tests
Add parseMcpListContent helper for the [] sentinel block and update
list tool tests that can receive zero results.
* test(mcp): fix list_collections type assertion in integration test
---------
Co-authored-by: syf2211 <syf2211@users.noreply.github.com>
* Add Slab importer backed by the Markdown import pipeline
Slab exports a zip of Markdown documents that is structurally identical to
the Outline Markdown export, with one difference: images are referenced as
remote signed URLs rather than files inside the archive.
Rather than build a parallel importer, this reuses the Markdown task and
processor and teaches the shared import pipeline to download remote images:
- Enable per-page attachment upload for the Markdown task. The base
APIImportTask already downloads remote image/video/attachment URLs and
rewrites them to internal redirect URLs; the Markdown task previously
opted out because its attachments live in the zip.
- Make the base upload step skip URLs that are already internal, so local
zip attachments (resolved to redirect URLs during rewriteMarkdown and
uploaded from the archive in onAllTasksCompleted) pass through untouched
while remote URLs are fetched and re-hosted. Also guards the URL rewrite
against nodes not present in the download map.
- Introduce IntegrationService.Slab as a distinct, importable service that
routes through the Markdown task/processor, so imports are tracked and
labelled as "Slab" with no duplicated import logic.
- Wire the create API (schema + route), the import settings UI (Slab card
and dialog), and a placeholder logo asset.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AC5Ho2SHBjNpxEAMr5uL6o
* Add shell SlabAPIImportTask as a Slab-specific seam
Slab imports previously ran through MarkdownAPIImportTask directly. Introduce
a thin SlabAPIImportTask subclass so Slab gets its own task name (for
scheduling, tracing, and retries) and a dedicated place to override
behavior as Slab exports diverge from the generic Markdown shape.
The subclass inherits all conversion/attachment/persistence logic and only
overrides scheduleNextTask to keep the whole import chain on the Slab class.
MarkdownImportsProcessor now selects the task implementation by the import's
service when scheduling the first task.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AC5Ho2SHBjNpxEAMr5uL6o
* Unwrap single root "slab" directory in Slab imports
Slab exports wrap the entire workspace in a single top-level directory named
"slab". Without handling this, the import produces one "slab" collection
containing everything, instead of mapping each workspace area to its own
collection.
Add a `resolveCollectionRootNodes` seam on the Markdown bootstrap phase
(default: pass entries through unchanged) and override it in SlabAPIImportTask
to descend into a lone, case-insensitive "slab" root directory so its child
directories become collections. Paths are left intact, so the attachment
manifest, completion re-walk, and internal-link resolution stay consistent.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AC5Ho2SHBjNpxEAMr5uL6o
* Use the filename, not the first heading, as the title for Slab imports
In Slab a document's leading heading is real content, not its title — the
filename is authoritative. Add an `extractTitle` option to
DocumentConverter.convert (default true, preserving existing callers) that,
when false, skips lifting a leading H1 into the title and leaves it in the
body. Expose it through a `shouldExtractTitleFromHeading` seam on the
Markdown task and override it in SlabAPIImportTask. With no extracted title,
the page falls back to the filename-derived title.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AC5Ho2SHBjNpxEAMr5uL6o
* Update logo
* Narrow MarkdownAPIImportTask generic back to Markdown
Slab imports now run through SlabAPIImportTask, so the base Markdown task no
longer handles the Slab service directly and its generic can return to
Markdown-only. SlabAPIImportTask keeps a Markdown | Slab union on its
scheduleNextTask override, which is required so the override stays a valid
(contravariant) override of the base's Markdown-typed parameter.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AC5Ho2SHBjNpxEAMr5uL6o
* Optimised images with calibre/image-actions
* reorder
* Move Slab importer into its own plugin
Restructure the Slab importer to live in plugins/slab, mirroring the Notion
plugin, instead of being threaded through the core Markdown importer.
- plugins/slab/server registers SlabImportsProcessor (Hook.Processor) and
SlabAPIImportTask (Hook.Task). The processor extends MarkdownImportsProcessor
to claim the Slab service and schedule the Slab task; the task keeps its
Slab-specific overrides (remote image download, root "slab" dir unwrap,
filename-as-title).
- plugins/slab/client registers the Slab import card (Hook.Imports) and owns
the import dialog.
- Revert MarkdownImportsProcessor to Markdown-only and drop the hardcoded Slab
card from the Import settings screen.
The Slab service enum, import schema, and create route remain in core, as the
Notion plugin's equivalents do.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AC5Ho2SHBjNpxEAMr5uL6o
* Generalize DropToImport to accept an importable service
DropToImport previously branched on `format` to map MarkdownZip/JSON/Slab
to specific imports.create calls, with a dead collections.import fallback.
Replace this with a single `service` prop and one generic imports.create
call. Core no longer hardcodes any service name (notably the Slab plugin
service), and the Markdown, JSON, and Slab dialogs each pass their service.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AC5Ho2SHBjNpxEAMr5uL6o
---------
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
* fix: Hand desktop sign-in back to the app on self-hosted deployments
The desktop `/desktop-redirect` handoff — which bounces the transfer token
back into the app via the `outline://` protocol — was nested inside the
`env.isCloudHosted && team.subdomain` branch of `signIn`. On a self-hosted
deployment that branch is never taken, so a desktop OIDC (or any) sign-in fell
through to the single-domain path, which set an `accessToken` cookie in the
browser and redirected to the app's home. The desktop app never received the
token and appeared to hang on the SSO screen.
Move the desktop handoff out of the cloud-hosted guard so it runs for every
deployment. The cloud sessions cookie is still written for cloud-hosted teams
(web and desktop alike), and the web / single-domain redirects are unchanged.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Wv8LuN54DQoxofSoWmTPVK
* refactor: Encode transfer token in redirect URLs and clarify comment
Address PR review: URL-encode the transfer token before interpolating it into
the desktop-redirect and auth/redirect query strings so the value can't
truncate the query if the token format ever changes, and reword the comment
that inaccurately described a cookie being set in this branch.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Wv8LuN54DQoxofSoWmTPVK
* docs: Trim desktop redirect comment
Remove the trailing explanation added earlier; the comment reads cleaner
without it.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Wv8LuN54DQoxofSoWmTPVK
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix: Reject document updates that contain no updatable fields
A documents.update request carrying a valid id but no recognized body
field (e.g. text omitted, or an unmapped key stripped by validation)
previously returned 200 OK while writing nothing, so API and MCP
callers could believe content was persisted when nothing changed.
documentUpdater now throws a ValidationError when none of the
updatable properties are provided, making the no-op fail loud with a
400 for both the REST route and the MCP update_document tool. Requests
that only signal an editing session (done) or publish a draft continue
to work.
Closes#12875
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QCYPC1guJ85BezXLNw8Zjo
* fix: Error on MCP document update with no fields provided
Scope the no-op guard to the MCP surface only, and derive it from the
provided input rather than a hardcoded list of properties: any key
other than id counts, since the tool schema already strips
unrecognized fields.
Closes#12875
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QCYPC1guJ85BezXLNw8Zjo
* fix: Base MCP no-op update detection on whether the document changed
Compare the document's revisionCount before and after documentUpdater
runs instead of inspecting the request input. Every save increments
revisionCount, so an unchanged count means nothing was persisted —
this catches both requests with no recognized fields and requests
whose values are identical to the current document.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QCYPC1guJ85BezXLNw8Zjo
* fix: Extend no-op update detection to all MCP update tools
Apply the same fail-loud behavior to update_collection and
update_comment: check model.changed() before saving and return an
error when the request would persist nothing. Mark all three update
tools as non-idempotent since retrying an already-applied update now
returns an error rather than a second success.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QCYPC1guJ85BezXLNw8Zjo
* chore: Fix punctuation in no-op update error messages
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01QCYPC1guJ85BezXLNw8Zjo
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix: Validate all DNS records in validateUrlNotPrivate SSRF check
A hostname with multiple A/AAAA records would pass validation as long as
the first record returned was public, even if another record pointed at a
private or link-local address the connection could actually use. Resolve
with { all: true } and reject if any record is private and not allowlisted.
Also wrap DNS resolution failures in InvalidRequestError so an unresolvable
hostname returns a clean 400 instead of an unhandled ENOTFOUND.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UotVRi1xphwNwAchiZZNi5
* chore: Add explicit type for resolved DNS addresses
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UotVRi1xphwNwAchiZZNi5
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix: Math formulas not rendering correctly in HTML/PDF output
* Fix version exactly.
---------
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* perf: Use SCAN and batched UNLINK instead of KEYS in CacheHelper.clearData
The KEYS command blocks the Redis event loop while it walks the entire
keyspace, which stalls all other clients on large installations —
notably on startup where the whole unfurl cache prefix is cleared.
Replace it with incremental SCAN iteration and delete the matched keys
in batches with UNLINK, which reclaims memory asynchronously, so
neither discovery nor deletion blocks the server. Adds test coverage
for CacheHelper including multi-page prefix clears.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015K9AAwaGTLVQZnp7uv5Ebj
* fix: Address review comments on CacheHelper.clearData
Unlink each SCAN batch as the scan progresses instead of buffering all
matching keys in memory, repeating until a full pass matches nothing so
implementations with unstable scan cursors still clear every key. Wrap
the operation in try/catch with logging for consistency with the other
CacheHelper methods, and mark constants as readonly.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015K9AAwaGTLVQZnp7uv5Ebj
---------
Co-authored-by: Claude <noreply@anthropic.com>
Gate the workspace attachment-size query behind includeAttachments so it
no longer runs when attachments are excluded, and thread includeAttachments
through the single-document export path (previously hardcoded to true).
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* fix: Resolve _FILE env secrets lazily to avoid clobbering third-party variables
The Docker-style secrets support added in #11906 eagerly copied every
*_FILE environment variable into its base variable at boot. This broke
AWS SDK credential refresh on EKS Pod Identity: the rotating token in
AWS_CONTAINER_AUTHORIZATION_TOKEN_FILE was frozen into the static
AWS_CONTAINER_AUTHORIZATION_TOKEN variable, which the SDK prefers and
never re-reads, so S3 access failed once the boot-time token expired.
It similarly resolved the standard OpenSSL SSL_CERT_FILE CA bundle into
Outline's SSL_CERT setting, failing validation at startup.
File secrets are now resolved lazily through a proxy when a variable is
read off the environment export, which limits resolution to variables
declared on the Environment classes, and SSL_CERT_FILE is explicitly
reserved for OpenSSL.
Fixes#12885
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MqP3Gb29iG5bJmhYuEAAjF
* refactor: Generalize reserved file variable documentation
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MqP3Gb29iG5bJmhYuEAAjF
* refactor: Clear SSL_CERT_FILE in test environment instead of reserving it
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MqP3Gb29iG5bJmhYuEAAjF
* docs: Trim withFileSecrets JSDoc
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MqP3Gb29iG5bJmhYuEAAjF
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix: Add WWW-Authenticate header to /mcp 401 responses
RFC 9728 and the MCP authorization spec require protected resources to
include a WWW-Authenticate header on 401 responses so clients can
discover the authorization server via the resource metadata document.
* fix: Preserve existing WWW-Authenticate and guard against non-object errors
Avoid overwriting an upstream WWW-Authenticate header if one is already
present on the error (case-insensitive lookup), and only attempt to
attach headers when the thrown value is a non-null object.
RFC 7591 §2 marks every client metadata field as OPTIONAL. Outline's
schema currently treats client_name as required, which rejects MCP
clients that omit it during dynamic registration. The field becomes
optional and a generic 'Untitled application' label is used when the
caller doesn't supply one.
Drop rejectOnEmpty so a deleted team yields null and the existing optional
chaining guard short-circuits the no-op transaction.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
The team or user can be deleted before the async processor runs, so
gracefully skip the avatar upload instead of throwing.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
* feat: support presigned PUT uploads for S3-compatible storage
Add `AWS_S3_UPLOAD_METHOD` config ("post" default / "put") so providers
that do not implement presigned POST (e.g. Cloudflare R2) can use
presigned PUT instead.
- BaseStorage.getPresignedPut() returns undefined by default; S3Storage
overrides with PutObjectCommand + getSignedUrl.
- Content-Length is signed into the presigned PUT URL so S3 rejects
uploads that do not match the declared size.
- API route and MCP tool branch on the config: only one upload method
is returned per request.
- Frontend uses existing presignedPutUrl presence check — unchanged.
- When put is configured but unsupported (e.g. LocalStorage), a clear
InvalidRequestError is returned.
Defaults to "post" — zero behavior change for existing deployments.
* refactor: return explicit upload mode instead of inferring from fields
Address review feedback from @tommoor: the client should not need to
infer the upload method from the presence of presigned-PUT-specific
fields. The API now returns an explicit `mode` ("put" | "post"), and the
PUT branch uses generic `url` / `headers` fields so the client has no
special knowledge of "presigned puts".
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* refactor: use this.getBucket() in presigned post/put methods
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>