Review Flow PR 3. Stacked on #7296. A recorded failure becomes readable
by the person who caused it.
## What changes
Before this, reading or triaging a failure required leader permissions:
`FileRunEventController.requireFailureReviewAllowed()` returned 403 to
anyone who could not edit policies. #7296 lets any user report a
failure, so they could file into a queue they could never read.
That gate is removed from the endpoints and the decision moves into
`FileRunEventService`:
| Caller | Reads and closes |
|---|---|
| Team leader or admin | the whole team's failures (unchanged) |
| Anyone else | only failures where `actor` is them |
| Team unresolvable | nothing |
| Name unresolvable | nothing |
`GET /kinds` is also opened. It returns static enum metadata, and a
member needs it to render failures they can already see.
## Additions
- An `actor` predicate on both list queries in `FileRunEventRepository`,
threaded through `FileRunEventStore.list`.
- `ReadScope` (permitted, teamId, actor) replacing `TeamScope`, with
`wholeTeam` / `mine` / `denied` factories.
- An actor filter on `dispatch`, so acting on another person's row
answers **404, not 403** — the same response as an id that does not
exist.
## Fixes
- **`report()` filed rows under the wrong team.** It took the team from
the read scope, which returns null for a caller who cannot be named, so
such a report landed unteamed in the bucket every team shares. It now
uses a dedicated `currentTeamId()`.
- **`forgetFiles` narrows to the caller even for a leader.** File ids
are minted by each client, so scoping on team alone would let one caller
close a colleague's incidents by naming ids.
- The controller no longer injects `PolicyManagementAuthority` or
`ApplicationProperties`; with the gate gone it decides nothing.
## Team isolation
Unchanged and covered by database-backed tests rather than mocks.
`FileRunEventStoreDbTest` asserts that a caller with a team sees only
their own team's rows and never the unteamed ones, and that the actor
predicate narrows within a team without ever widening across one. Delete
either clause from the JPQL and one of those tests fails.
No endpoint accepts a team parameter; the team always comes from the
authenticated principal.
**Attribution is fixed here too, because this PR depends on it.** A
failure's actor was read from the MDC audit principal, which carries the
BILLING identity — for a stored policy, always its owner. Since reads
are now narrowed to the rows you are the actor on, a wrong actor means
the member who caused a failure and holds the document reads nothing,
while the policy owner is handed incidents from runs they never
triggered. The triggering user is now carried on the run, separate from
the billing principal and the output owner, and is null for a
trigger-fired sweep so an unattended failure stays ownerless.
`PolicyFailureAttributionTest` runs the real engine, recorder, store and
service together. The two sides used to assert independently — the
engine's test matched the actor with `any()`, which is how this went
unnoticed.
## How to test
Needs a proprietary or SaaS build with login enabled and two accounts in
the same team, one a leader and one not. `task dev:all` gives you the
stack.
1. **As the member**, fail a tool: open a PDF and run **Remove
Password** with a wrong password.
2. **Still as the member**, go to `/processor/documents` → **Failures**.
Before this PR you got nothing here. Now you see your own row, and only
yours.
3. **As the leader**, open the same view. You see the whole team's rows,
including the member's.
4. **Member cannot reach a colleague's row.** As the leader, copy a
row's id from **Show raw JSON**. As the member, `POST
/api/v1/file-run-events/{thatId}/actions/DISMISS`. It answers **404**,
and the row is untouched — it must not answer 403, which would confirm
the row exists.
5. **Member can close their own.** Dismiss your own row as the member.
It leaves the default view.
6. **Deleting a file only closes your own rows.** As the leader, delete
a file in your editor. The member's incidents are untouched even if the
leader's client happened to name the same ids.
## Migration
None. `actor` is an existing column; this only adds predicates to
existing queries.
Review Flow PR 2 of 5. Editor tool failures now reach the same durable
queue as failures from folders, buckets and webhooks.
## What's added
**A report endpoint** — `POST /api/v1/file-run-events/reports`, open to
any authenticated user. Takes four fields: `operation`, `errorCode`,
`fileIds`, `detail`. No team, no actor, no filename: the first two come
from the session, the third is never a field. Refused with 400 above 200
file ids, and nothing is written when refused.
**Automatic reporting from every tool** — wired into `useToolOperation`,
so no per-tool work is needed. Client-side refusals (an unsupported
format that never reaches the server) are reported too. User
cancellations are not.
**Error codes parsed from Blob bodies as well as JSON** — a
download-typed tool call fails with a Blob, so `errorCodeOf` handles
both shapes.
**Source attribution for unattended runs** — `sourceId` is threaded from
`PolicyRunner` through `PolicyRun` to the recorded row and out to the
wire, so a folder, bucket or webhook failure names what fed it.
Previously it had none.
**Deleting a file closes its failures** — `FileContext.removeFiles`
notifies `POST /removed-files`, which transitions those incidents to
`FILE_REMOVED`. Terminal, so they leave every reviewer's queue. The rows
stay for audit.
**The queue can be emptied** — reads now default to open statuses only;
ask for a status explicitly to see closed rows.
## Behaviour changes
- **Editor failures dedup per person.** `RecordFailure.scopeRef()`
includes the actor for TOOL-origin rows, so two people hitting the same
failure on the same file are two incidents rather than one. Processor
rows are unaffected and their dedup key is byte-identical to before.
- **`UNKNOWN` offers only Dismiss.** Acknowledge is no longer offered on
it.
- **Background reports no longer raise a toast.** Both calls pass
`suppressErrorToast`, so a failed report is silent as intended;
previously a core build showed the user a "Not Found" toast on every
tool failure.
## What is stored
File ids only, never names. The request type has no filename field, and
a `fileNames` value handed to the client reporter is accepted and
ignored.
One caveat to review deliberately: the free-text `detail` is stored
**verbatim**. `RecordFailure` truncates it at 2000 characters and
nothing else; the redaction that used to strip name-shaped text was
reverted in `024899f3f6` because it made an unclassified failure
impossible to act on. A backend message that embeds a filename
(LibreOffice conversion errors, IO errors) will therefore persist that
text and show it to a team leader.
## How to test
Needs a proprietary or SaaS build with login enabled. `task dev:all`
gives you one.
1. **Report a failure from a tool.** Open a PDF, run **Remove Password**
on it with a wrong password. Nothing visible changes for you: reporting
is silent by design.
2. **See it recorded.** Go to `/processor/documents` and scroll to
**Failures** (dev builds only). A row appears titled "Password-protected
document", with `Hit by <your user>`. Press **Show raw JSON** to see
exactly what was stored.
3. **Confirm no filename is stored as data.** In that JSON, `fileId` is
an opaque uuid and there is no name field. Note the `detail` string may
contain a filename if the backend put one in its message, per the caveat
above.
4. **Confirm the request is capped.** In DevTools, POST to
`/api/v1/file-run-events/reports` with 201 entries in `fileIds`. It
returns 400 naming the limit, and no rows are added.
5. **Deleting a file clears its failure.** Back in the editor, delete
the file you just failed on. Refresh the failures list: its row is gone
from the default view. Filter by `FILE_REMOVED` to see it still exists.
6. **Two people, two incidents.** Have a colleague fail the same tool on
their own copy of the same file. Two rows, not one occurrence count.
## Migration
`source_id` is a new column and `FILE_REMOVED` a new status value. Both
are already in the SaaS migration ([Stirling-PDF-SaaS
#322](https://github.com/Stirling-Tools/Stirling-PDF-SaaS/pull/322));
self-hosted picks them up from `ddl-auto`.
Five unrelated snags in the processor (portal) UI, plus fixes they
turned up. No backend changes.
`84 files changed, +892 / −3364`
## Fat CTA buttons
- New `fat` prop on the SUI `Button`: 2.75rem tall, 1.25rem side
padding, 0.75rem corners, semibold. Composes with all four
variants/accents.
- Applied to the page-header CTA on Sources, Documents, Pipelines, Users
(both), Usage, Integrations, Infrastructure — 8 buttons, all in line
with a page title. Nothing else.
- `LandingActions` migrated onto the prop; `.landing-btn-primary` /
`.landing-btn-secondary` and their four `!important`s deleted. The
editor landing CTAs come down 4px with everything else.
- Infrastructure's header CTA is now primary; its "Create key" dropped
to secondary so they stop competing.
<!--IMG:buttons-->
## Documents empty state
- "Connect a source" opened the Sources *page*; it now opens the
`SourceModal` connect flow in place, no route change.
- No extra cache wiring: `SourceModal` already invalidates the sources
query.
<!--IMG:documents-->
## Infrastructure tabs
- Only API Keys and Audit Logs hit real endpoints. Deployments,
Security, Models and Storage read mock-only `/v1/infrastructure/*` that
no backend serves.
- Those four are now disabled: native `disabled`, out of the keyboard
tab order, `aria-disabled`, with the view refusing non-enabled keys as a
second guard.
- Real tabs moved leftmost; API Keys is the default; `?tab=` deep links
validated against the enabled set (the home flow's audit link still
works).
- Deleted: 4 tab components, their fetch fns and ~25 dead types, MSW
handlers, fixtures (908 → 253 lines), dead CSS, unused formatters, 240
lines of `en-US` strings. Most of the −3364.
- Page subtitle no longer advertises the disabled tabs.
<!--IMG:infrastructure-->
## Surface consolidation
- New `Surface` primitive (`sui-surface`): fill, hairline, radius, no
shadow. Kept separate from `sui-nav-surface` so nav chrome can diverge
later.
- `Card` composes it and no longer draws its own shadow — this changes
editor Card usages too, by design.
- SUI primitives that are surfaces adopt it: `MetricCard`,
`MetricStrip`, `NodeCard`, `Table`, `Collapsible`, `CodeBlock`.
- The portal gets its own `.portal-surface` with the same three
declarations, applied to 19 elements. A `sui-` class belongs to the
component that emits it, so feature markup doesn't wear one.
- `raised` variant = one subtle shadow for a surface in front of another
surface (the flow diagram's tiles). Same fill as its parent, so nesting
never shifts a region's colour. Dark has its own value.
- Floating chrome (modals, drawers, dropdowns, assistant, sidebar) keeps
its elevation; sunken wells stay sunken.
<!--IMG:surfaces-->
## Sources list
- Centred "No sources connected yet" empty state removed — it duplicated
the header CTA and pushed the table down the page. The header's "Connect
source" is the single way in.
## Drive-by fixes
- The connect flow rendered unstyled outside the Sources view:
`.portal-conn-picker__*` / `.portal-sources__connection-*` lived in
`views/Sources.css`, which none of the five components rendering them
imported. Moved to `components/sources/connections.css`.
- Three inert custom properties (`--surface-input`, `--color-border-2`,
`--text-default`) are defined nowhere in the codebase —
`.portal-conn-picker__card` had no fill at all as a result.
- Dead CSS removed from `Sources.css` (grep-verified unused): old
expanded-row panel + its keyframes, type-card block.
## Testing
- `task frontend:check` — typecheck, lint (oxlint + 4 theme-lint passes
+ stylelint), format, 238 files / 2063 tests.
- `frontend:typecheck:all` across all 9 tsconfigs.
- `frontend:storybook:a11y:changed` — 119 stories, light and dark, zero
violations, no regressions vs baseline.
- New tests: `Infrastructure.test.tsx` (tab order, default, disabled
behaviour, deep-link filtering) and a Documents test that the
empty-state CTA opens the modal without navigating.
- Merged `origin/main` (#7438 replaced `PipelineHeader` with the new
Create/Edit headers); full suite green at 240 files / 2072 tests after
the merge.
# Description of Changes
Follow-up to #7314, which fixed the IndexedDB blob rejection itself.
This one fixes the remaining WebKit engine gaps, fixes the ways that
class of failure surfaced to the user, and adds the cross-browser signal
that would have caught them on the PR instead of six weeks later.
## Why this exists
Two total WebKit outages sat on `main` for weeks:
1. pdf.js reads its text stream with `for await (… of readableStream)`,
and WebKit has no `ReadableStream[Symbol.asyncIterator]`. **All** pdf.js
text extraction threw `TypeError: undefined is not a function` —
Compare, read-aloud and the PDF text editor were dead on Safari.
2. IndexedDB in WebKit rejects Blob/File values with `UnknownError:
Error preparing Blob/File data to be stored in object store`, so nothing
persisted and every reload came back empty.
Neither was caught, because the existing specs never did the work. The
Compare specs filled both slots and asserted the button was enabled;
none of them clicked it. The persistence specs asserted a *filename*
reappeared after a reload, which only needs the metadata record, not the
bytes.
Every failure here **looked like success** — empty panes, blank
thumbnails, a `src` that was set but empty. That shapes the tests more
than the fixes.
## WebKit engine gaps
- **`ReadableStream[Symbol.asyncIterator]`**, installed at the entry
point before any PDF work starts. The lock discipline is the subtle
part: releasing is idempotent, is *not* done after a successful read,
and *is* done in the read's error steps — `for await` never calls
`return()` when `next()` rejects, so nothing else would ever unlock an
errored stream.
- **`requestIdleCallback`**, installed once instead of guarded at each
call site. This one wasn't broken, it was mistimed: the local fallbacks
fired at 200ms and 1000ms, landing the pdfium WASM compile on top of the
app's first renders. The shim honours the caller's full timeout, so
`{timeout: 2000}` means 2000ms.
- **`convertToBlob()` does not fail on a format it can't encode.** Per
spec it silently serialises to PNG, so asking for WebP and getting PNG
back looks like success. Canvas output now probes what the engine really
produced (once per realm) and uses the best lossy format it honours. PNG
of a rendered page is several times the size of the equivalent WebP or
JPEG, held as object URLs for every page on screen, on the engine with
the tightest renderer memory budget.
## WebKit storage failures
These read as generic transaction hygiene. They aren't — a refused blob
write **aborts its transaction**, which is the mechanism that turned a
WebKit rejection into a hang.
- **Blob refusal is remembered from any write**, not just the initial
`add`. WebKit reports it when it can't write the blob's *backing file*,
which is per-operation — an engine that accepted the add can still
refuse the rewrite, and every read-modify-write rewrites the record with
its body attached.
- **Aborted transactions no longer hang.** Read-modify-write moves to a
single `updateRecord` helper that owns its transaction, guards it once,
and resolves on **commit** rather than on the put's `onsuccess`. The
previous shape — two promises over one shared transaction, with an
`await` between the get and the put — put the abort guard on the read,
leaving the write with no handler at all. `persistVersionedOutputs`
awaits that, and `.catch` can't rescue a promise that never settles, so
tool outputs could silently stop persisting.
- **Stored blobs are no longer re-wrapped on read.** Since #7175 the
record holds the `File` itself; wrapping it in `new Blob([record.data])`
can cost WebKit the backing handle, giving you an object that looks
valid and reads as empty.
- **The file sidebar reaches a resting state** when the library can't be
read, instead of spinning forever on a rejection nobody observes. It
carries on with the in-memory workbench files: an unreadable library
should cost the user their history, not the file they're working on.
- **Thumbnail failures are logged.** Three `catch {}` blocks returned
`""`, and an empty thumbnail is indistinguishable from "this file has no
preview" — which is how outage #1 hid as a cosmetic nicety.
## CI
`main` now runs the whole stubbed suite once per engine (#7304), so the
new `@engine-capability` specs get chromium, firefox and webkit for
free. They assert the primitives actually work — a **counted**
comparison, a raster thumbnail data URL with real payload, and a page
rendered from a file restored by a reload — rather than that the UI
rendered. Deliberately small: anything added there is paid for three
times per PR, so add depth, not breadth. Run them alone with `task
e2e:cross-browser -- --grep @engine-capability`.
The cross-browser projects now share the stubbed project's viewport. At
the device presets' default 1280x720 a layout difference would fail
these specs on Firefox/WebKit only, which reads as an engine outage.
`vite.config.ts` gains a `worker.plugins` entry so `@app/*` resolves
inside worker bundles. Worker bundles are a separate Rollup pass and
don't inherit `plugins`, so the alias worked in the app and failed in a
worker — previously worked around with a relative import plus a lint
exemption, which silently bypasses the layer cascade.
## Verification
- `task frontend:check` green: typecheck, oxlint, theme lint, stylelint,
prettier, 215 test files / 1841 tests.
- The `@engine-capability` suite passes on Chromium and WebKit locally.
- **Negative control:** with the `ReadableStream` shim removed, the
WebKit comparison spec fails at the Deletions/Additions assertion — the
exact reported Safari symptom. Restored, and it passes. Both the fix and
the test that guards it are load-bearing.
- The worker alias change verified both ways: the build inlines the
encoding probe into the worker chunk, and removing `worker.plugins`
fails with `Rollup failed to resolve import
"@app/utils/canvasImageEncoding"`.
- The abort regression test aborts the transaction mid-write and asserts
`markFileAsProcessed` settles. Before the fix it never settles and the
test times out.
## Split out of this PR
Two things in earlier revisions of this branch were engine-agnostic —
found via the same symptom, not the same cause — and now have their own
PRs:
- **#7416** — blocked IndexedDB upgrades hanging the file library
(multi-tab lifecycle, the concurrent-open race, `onversionchange`).
- **#7417** — the thumbnail TTL rewriting the whole library on every
listing.
`FileSidebar`'s try/catch appears in both this PR and #7416,
identically: a WebKit rejection and a blocked-open rejection both have
to stop stranding the spinner. Whichever merges second is a no-op for
that file.
## Known gaps
- The blob-refused **rewrite** recovery in `updateRecord` isn't
unit-tested. `fake-indexeddb` never returns Blob values from a read, so
the branch that converts to a copy can't be reached there. Noted in the
test file.
- For the same reason, `fileFromRecord`'s "hand the stored File back
untouched" path is only covered on a real engine, by the reload spec.
- Nothing asserts that `src/index.tsx` imports the shims. The unit suite
installs the same module via `setupTests.ts` (jsdom has the same gaps
WebKit does), so a future regression where the entry point drops the
import would still be green under vitest.
- `FileSidebar`'s resting-state fix loses its E2E coverage until #7416
lands — forcing WebKit's blob refusal from a spec isn't practical, which
is why that spec blocks the database instead.
---
## Checklist
### General
- [x] I have read the [Contribution
Guidelines](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/CONTRIBUTING.md)
- [x] I have read the [Stirling-PDF Developer
Guide](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/DeveloperGuide.md)
(if applicable)
- [x] My changes generate no new warnings
### Documentation
- [x] I have updated relevant docs on [Stirling-PDF's doc
repo](https://github.com/Stirling-Tools/Stirling-Tools.github.io/blob/main/docs/)
(if functionality has heavily changed)
# Description of Changes
PR 1 of the failure-notification work: a durable, team-scoped record of
**why a policy run failed**, surfaced in the portal with the triage
actions each failure allows.
Today a failed policy run is not quite invisible, but it is unusable:
the ledger marks the file `ERROR`, and the audit aspect keeps the
exception message and status code. Nothing classifies either one,
nothing surfaces them, and neither offers a next step. If the file came
from a folder, bucket or webhook there is also no user watching, so
nobody learns it never made it through. This adds the record and the
read surface; the remediation that acts on documents comes later (see
below).
## What this does
**A failure kind registry as data.** `FailureKind` describes what can go
wrong: a stable wire id, i18n keys, an English fallback, and four facets
the review surface needs (`Stage`, `Severity`, `Remedy`, `Scope`). It is
shaped like the existing `ExceptionUtils.ErrorCode` and *links* to that
vocabulary rather than replacing it.
**Classification off structured codes, not message matching.** Policy
steps dispatch over loopback HTTP, so a tool's 4xx arrives as a
`RestClientResponseException` whose body is the Problem Details document
carrying `errorCode`. `FailureClassifier` reads that. Anything
unrecognised becomes `UNKNOWN`, which is the point: every failed run
gets an addressable record from day one, and which kinds to promote next
is answered by production frequency rather than guesswork.
**Actions declared by a kind, implemented as beans.** A kind lists the
`FailureActionId`s it offers; behaviour lives in `FailureAction` beans
resolved by id — the idiom this codebase already uses for `InputSource`,
`PolicyOutputSink` and `PolicyTrigger`. A kind cannot be sent an action
it never declared (400), so an incoherent pairing is unreachable rather
than merely unrendered. A new kind ships as a registry entry plus copy:
no new endpoint, no UI change.
**Repeat folding.** Recording folds a genuine repeat into the existing
incident instead of inserting again, keyed on `(team_id, dedup_key)`.
That matters for a snapshot-mode source that re-lists every file on each
sweep: the same broken file is one incident, not one per sweep. Distinct
files keep distinct rows. The unique constraint is enforced by the
database, and a writer that loses the insert race folds into the
winner's row.
One granularity caveat worth naming: nothing populates `file_id` in this
PR, so every row has it NULL. A FILE-scoped kind therefore dedups on
`policy + run` rather than `policy + file`. That still yields one row
per document for the sources shipped here, because the folder, S3 and
webhook sources each start one run per file; it stops holding as soon as
a single run carries several documents, which is why editor-origin
reporting (item 3 below) populates `file_id`.
**No document identity is stored.** No file name, no content. `fileId`
is an opaque reference only the owner's own client can resolve locally.
`detail` keeps the raw message (the only diagnostic an `UNKNOWN` failure
has) with anything path- or filename-shaped stripped on the way in,
capped at 2,000 characters. `PolicyExecutor`'s type-mismatch message now
reports the *extension* rather than the filename, since that message
becomes the stored `detail`.
**Access.** Reads and triage are leader-only, gated exactly the way
`PolicyController` gates policy editing, with the single-user carve-out
when login is disabled. Every read and write is scoped to the caller's
own team from the authenticated principal — there is no team parameter
on the API.
Self-hosted needs no migration: the table is created from the entity by
`ddl-auto=update`, as with every other table.
## What this does not do yet
- **Actions are incident dispositions, not document dispositions.**
Acknowledge and Dismiss change how a failure is displayed and touch
nothing else — not the document, not the processed-file ledger, not the
run, not any output destination. That is what makes them safe to offer
against `UNKNOWN`, and why there is no Approve/Release yet.
- **Two kinds only.** `INPUT_PASSWORD_PROTECTED` and `UNKNOWN`.
Everything else classifies as `UNKNOWN` and shows its raw message.
- **Editor-origin failures are not reported.** Every row is `PROCESSOR`.
`FailureOrigin.EDITOR` and `API` exist in the enum but nothing writes
them.
- **The list is dev-only for now.** The section renders behind
`import.meta.env.DEV`, so it ships in no production bundle. The
endpoints are live and gated.
- **No retention or per-team cap** on `file_run_events`. Tracked
separately.
- **No suspend-and-prompt.** `PolicyInputRequiredException` and the
engine's `suspend()` exist but nothing throws it, so a run cannot pause
to ask for a password today.
- **SaaS needs a migration** in `Stirling-PDF-SaaS` (`CREATE TABLE IF
NOT EXISTS stirling_pdf.file_run_events`), per the convention documented
at `app/saas/src/main/resources/application-saas.properties:21`.
## What follows in later PRs
1. **Map the remaining error codes to specific kinds** — corrupted file,
OCR unavailable, output destination unreachable, entitlement refusals,
and so on — each with its own copy and its own action set, replacing
today's `UNKNOWN` catch-all with a named notification in the review UI.
2. **Real remediation actions** attached to those kinds: fix (supply a
password and resume), skip (drop this file, continue the batch), and
decline (reject an incoming file outright), acting on the held document
rather than only on the incident row. This is where the
suspend-and-prompt path gets wired.
3. **Editor-origin reporting**, so a failure a user hits in the editor
lands in the same queue as one from a bucket.
4. **The user-facing review surface**: notifications with a sticky
review section, per-file badges, and an export gate, with the dev-only
list here replaced by the real thing.
## How to test
Needs a SaaS or proprietary build with login enabled, and an account
that leads a team.
1. Create a policy in the Processor with any step (Auto-redact is fine)
and a source you can drop files into.
2. Upload two files that will fail it: **a password-protected PDF**, and
**a corrupted PDF** (truncate a valid one, or rename a `.csv` to
`.pdf`).
3. Let the policy run and fail on both.
4. Go to the portal's **Documents** view and scroll to **Failures** (dev
builds only).
Expect two rows:
- **Password-protected document** — classified from `E004`, with the
kind's own labels **"I'll unlock this"** and **"Skip this file"** rather
than generic wording.
- **Unrecognised failure** — the corrupted file, classified `UNKNOWN`
(`E001` is not claimed by a kind yet), showing its raw message with
generic **Acknowledge** / **Dismiss**.
Neither row contains a file name anywhere, including in the raw message.
Press **Show raw JSON** to read exactly what the server returned. Acting
on a row transitions it and comes back with both buttons disabled and a
reason.
Re-running the same batch increments the occurrence count on the
existing rows rather than adding new ones; two *different*
password-protected files produce two separate rows.
---
## Checklist
### General
- [ ] I have read the [Contribution
Guidelines](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/CONTRIBUTING.md)
- [ ] I have read the [Stirling-PDF Developer
Guide](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/DeveloperGuide.md)
(if applicable)
- [ ] I have read the [How to add new languages to
Stirling-PDF](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/devGuide/HowToAddNewLanguage.md)
(if applicable)
- [ ] I have performed a self-review of my own code
- [ ] My changes generate no new warnings
### Documentation
- [ ] I have updated relevant docs on [Stirling-PDF's doc
repo](https://github.com/Stirling-Tools/Stirling-Tools.github.io/blob/main/docs/)
(if functionality has heavily changed)
- [ ] I have read the section [Add New Translation
Tags](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/devGuide/HowToAddNewLanguage.md#add-new-translation-tags)
(for new translation tags only)
### Translations (if applicable)
- [ ] I ran
[`scripts/counter_translation.py`](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/docs/counter_translation.md)
### UI Changes (if applicable)
- [ ] Screenshots or videos demonstrating the UI changes are attached
(e.g., as comments or direct attachments in the PR)
### Testing (if applicable)
- [ ] I have run `task check` to verify linters, typechecks, and tests
pass
# Description of Changes
Fixes the WebKit nightly failures ([run
31067620195](https://github.com/Stirling-Tools/Stirling-PDF/actions/runs/31067620195/attempts/1)):
8 tests failed on `stubbed-webkit` only, and every one of them logs the
same thing in its trace:
```
IndexedDB add error: UnknownError: Error preparing Blob/File data to be stored in object store
```
## What broke
`storeStirlingFile` stores the `File` itself in IndexedDB, so multi-GB
uploads are persisted by reference and never materialize in JS memory.
That came in with #7175 (`data: stirlingFile` replacing `data: await
stirlingFile.arrayBuffer()`), which is a real memory win and worth
keeping.
WebKit refuses blob values whenever it can't write the blob's backing
file, and rejects the request with the error above. The rejection was
only `console.error`d, so on WebKit **no upload ever persisted**, and
everything that reads the bytes back behaved as if the upload never
happened:
- `file-state-across-tools` — file gone after navigating; the sidebar
shows "No files yet"
- `compare` — `FileSelectorPicker: upload failed`, so the slot stays
`data-slot-state="empty"`
- `classification-grouping` / `classification-heuristic-upload` — the
label backfill and thumbnails read from IDB (`not in IndexedDB (likely
remote-only stub)`), so files land in "Recent" with no category headers
Chromium and Firefox store blobs fine, and PR CI only runs the `stubbed`
(chromium) project, so nightly was the only gate that could catch it.
## The fix
Try the blob first, keep a fallback:
- `storeStirlingFile`'s `add` is extracted into `addFileRecord` so it
can run twice
- if the value was a Blob and the failure is `UnknownError` /
`DataCloneError`, re-add the record with an `ArrayBuffer` copy and set
`blobValuesSupported = false`, so later files in that session go
straight to the copy path instead of losing the blob attempt every time
- deliberately narrow: `QuotaExceededError` and `ConstraintError` still
propagate, because a copy would fail the same way and retrying would
hide the real cause
- dropped two internal `console.error`s: every caller already reports
(`addFiles`, `FileSelectorPicker`, `zipFileService` collects into
`result.errors`), so they were duplicate noise
Every writer goes through `storeStirlingFile` (uploads, the file picker,
zip extraction, folder automation, `IndexedDBContext`), so this one seam
covers all of them. The read paths already accept either shape (`new
Blob([record.data], ...)`).
Net effect: Chromium and Firefox keep the no-copy path; engines that
refuse blobs degrade to the pre-#7175 behaviour instead of silently
losing files. On such an engine a very large file can still exhaust
renderer memory — the fallback warns about exactly that. Fixing that
properly means chunked storage, which is out of scope here.
## Verification
Reproduced and confirmed the cause by A/B on a branch that predates
#7175: as-is 8/8 pass on WebKit, and applying only #7175's `data:
stirlingFile` line reproduces the exact CI failure set.
| Check | Result |
|---|---|
| `stubbed-webkit`: the 8 nightly failures +
`classification-heuristic-upload` | 9 passed |
| `stubbed-webkit`: `files-page`, `page-editor-rotation`,
`encrypted-pdf-unlock` | 32 passed, 1 skipped |
| `stubbed` (chromium): the same specs + `files-page` | 35 passed, 1
skipped |
| Frontend unit suite | 210 files, 1797 passed |
| `typecheck:core`, `typecheck:proprietary`, eslint, prettier | clean |
New unit coverage in `fileStorage.blobFallback.test.ts` pins the
contract over `fake-indexeddb` with `add` instrumented to count blob vs
copy attempts: blob path when accepted, blob-then-copy when refused (and
readable back), one attempt only for later files, and quota not retried.
---
## Checklist
### General
- [x] I have read the [Contribution
Guidelines](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/CONTRIBUTING.md)
- [x] I have read the [Stirling-PDF Developer
Guide](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/DeveloperGuide.md)
(if applicable)
- [x] I have performed a self-review of my own code
- [x] My changes generate no new warnings
### Testing (if applicable)
- [x] Frontend typecheck (core + proprietary), eslint, prettier, the
unit suite, and the affected Playwright specs on chromium and webkit all
pass
Co-authored-by: Anthony Stirling <77850077+Frooodle@users.noreply.github.com>
## Goal
Three related improvements to how policies and file state behave in the
editor: classification no longer blocks the user, policy enforcement
pipelines across a batch upload instead of waiting for the whole drop,
and file-state changes no longer re-render the entire UI.
## 1. Classification never blocks (and never versions)
Classification is metadata-only — it reads a document and records
labels; it never rewrites the file. Previously it ran like an
enforcement policy: it blocked viewing/editing behind the "Enforcing
policy…" overlay, forked a new versioned child (an `automate` entry in
version history), and could run before other policies — letting the user
in, then a later enforcement policy would fork a version and drop their
edits.
Now classification:
- **Never blocks.** A classification run never marks a file `enforcing`
(badge map + viewer overlay both skip it), so the file stays fully
viewable/editable while it runs.
- **No version bump, no history entry.** Its result is stamped onto the
file's existing stub in place (workspace + IndexedDB) — the labels just
appear as tags. It targets the document's *current leaf*, so an edit
made during the async run still gets the tags; a run that completes with
no outputs settles cleanly instead of pinning in-flight.
- **Always runs last** in an enforcement chain (regardless of configured
order, pinned at persist-time too), so every enforcement policy finishes
forking versions before the user is let in.
## 2. Pipeline policy enforcement across a batch upload
Dropping ~50 files enforced policies only *after the whole drop finished
scanning* — every file got the "Enforcing policy" overlay together, then
processing began. Root cause: the chunked `ADD_FILES` dispatches in
`addFiles` were never separated by an event-loop yield, so React batched
them into a single commit and the enforcement effect fired once over the
full list.
**Fix** (`core/contexts/file/fileActions.ts`): after each chunk, `await`
that chunk's IndexedDB writes, then yield a macrotask so React commits
the rows and runs the enforcement dispatch *before* the next chunk
scans. Files start enforcing as their rows land, overlapping with the
rest of the drop. Persistence is streamed per chunk (the policy auto-run
reads bytes from IndexedDB with no in-memory fallback).
**Second fix — bounded dispatch window**
(`proprietary/components/policies/usePolicyAutoRun.ts`): even with
streamed dispatch, the drop still *looked* serial — each dispatch POSTs
the file's bytes, and firing them all at once saturates the browser's
per-origin connection pool, so the status polls and output downloads of
already-running files queued behind the pending uploads; nothing visibly
progressed until the last upload drained. Dispatch is now gated behind a
small concurrency window (4), keeping connections free so early files
run, poll, and complete while later ones are still dispatching. The
first status poll also fires at 500ms (then the normal 2s cadence) so
fresh runs show real progress immediately. The batch test asserts the
window (dispatches overlap but never exceed 4).
## 3. Selector subscriptions for file state (no more whole-UI
re-renders)
`FileContext` published `{state, selectors}` through a plain React
context, so **every** consumer re-rendered on **every** state change —
one file's new version re-rendered the entire workspace.
**Phase 1 — infra** (`file/contexts.ts`, `file/fileHooks.ts`,
`FileContext.tsx`): the state context is replaced by a stable
subscription store (`FileStoreContext`); hooks are rebuilt on
`useSyncExternalStoreWithSelector` (the `use-sync-external-store` shim
react-redux uses — new direct dep, React 19 compatible). Each consumer
now re-renders only when its selected slice changes:
- `useStirlingFileStub(id)` → only that file's record
- `useAllFiles` → file-list changes only (immune to selection/UI churn)
- `useFileSelection`/`useSelectedFiles` → selection + the *selected*
files' records only
- `useFileUI` → its three UI scalars; `useFileContext` → files + pinned
slices
- `useFileState` keeps its whole-state contract for existing broad
consumers
A render-count test (`fileHooks.selector.test.tsx`) locks the bail-out
contract.
**Phase 2 — hot-path rows**: sidebar `FileItem` is memoized (with stable
empty-array props), so one file's change re-renders one row, not the
list. Active Files thumbnails were already memoized.
**Phase 3 — narrow the hottest consumers**: always-mounted whole-state
consumers migrated to slices — `Workbench`, `EmbedPdfViewer`, `Viewer`,
`NonPdfViewer`, `WorkbenchBar`, `ViewerContext`, `ViewerShareButton`,
`ZoomAPIBridge`, `ViewerAnnotationControls`, `ConvertSettings`,
`DismissAllErrorsButton`, `FileEditorThumbnail`,
`usePageEditorDropdownState`, `useSaveShortcut`, plus a new
non-subscribing `useFileSelectors()` for event-time reads
(`ReviewToolStep`, `useViewerReadAloud`, `useExitWarning`). Net effect:
selection/UI churn no longer re-renders the viewer/workbench, and a
version landing touches only components observing the files slice.
Broad readers (`FileSidebar`, `PageEditor`, `FileEditor`, `Redact`,
`FormFill`) deliberately stay on `useFileState` — they read most of the
state anyway.
**Hardening**: store notifications run in a layout effect (subscribers
re-render before paint — no stale frames), and outside production
`useFileSelectors()` wraps its selectors to `console.error` if one is
invoked during render (those reads don't subscribe, so render-time use
would silently go stale — not statically lintable, so it's guarded at
runtime; the full test suite passes under the guard).
## 4. Policy indicators: shared icons, non-blocking run chip, no pulse
- Badges and enforcement overlays now take their glyph from the shared
`policyCategoryIcon` map (the same source the processor's catalogue
uses) — label icon for classification, shield for security — instead of
a hardcoded shield everywhere.
- A non-blocking run (classification) shows a small accent-tinted pill
in the top-right of the Active Files card (category icon + loader) and
the normal spinning badge in the sidebar, via a new `background` badge
flag that nothing gates on. When the run finishes, the tagged files keep
a plain category badge.
- The post-run pulse/glow on sidebar badges is gone (with its `recent`
plumbing): spinner while running, static category icon when done.
## Verification
Full CI gate locally: `og:check`, `typecheck:all` (all variants),
`lint`, `format:check`, `build`, `test` (1366 — incl. the render-count
contract test, the classification-order/import unit tests, and the
61-file batch integration test driving the real dispatch → poll → import
→ chain effects), `storybook:build` — all green.
## Held for follow-up (not in this PR)
- **Reuse one PDFium engine across viewer file switches** (kills the
per-open "Loading PDF Engine" rebuild). Implemented on branch
`viewer/reuse-pdfium-engine`, but review found a confirmed leak
(orphaned PDFium handles when switching files mid-load); needs an
in-flight-load teardown before shipping.
Fixes the two nightly frontend jobs that started failing after #7163
(new design, part one). Two unrelated causes, one small fix each.
## Accessibility scan (`--c-primary-hover`)
The full a11y scan flagged a colour-contrast violation on the portal
pipelines ToolPicker story. #7163 moved the light canvas from
`--p-gray-50` (#f9fafb) to the slightly darker `--p-paper` (#f5f4f1),
but the accent text colour stayed put. That token doubles as the label
colour for quiet and tertiary buttons, so the pairing slipped from
4.64:1 to 4.41:1 purely from the background change.
Darkening the custom-theme mix from 85% to 80% primary puts it back at
4.89:1. It is also the hover fill for primary buttons, where a
marginally deeper blue is if anything more correct.
Only the nightly caught this because PR runs scan just the stories whose
files changed, and #7163 did not touch that story file.
## Cross-browser Playwright (right-click Copy menu)
Failed in Firefox only. The feature itself is fine in every browser. The
test hit-tested a word using a fixed fraction of the page box, and the
page is auto-fit to the viewer, so the rendered text scales with the
viewport. The Firefox and WebKit projects run at 1280x720, where the
page renders about 375px wide and the first line of text is only a few
pixels tall. #7163 shrank the viewer area slightly (the rails now float
with a gutter), which shrank the auto-fit page just enough to tip that
fraction to landing below the glyphs. Nothing was selected, so no menu
appeared.
Pinning 1920x1080 for that one test makes the glyphs comfortably larger
than the click tolerance everywhere, rather than re-tuning a fraction
that was only ever a couple of pixels from failing.
A Firefox skip was considered and rejected: the sibling clipboard test
is already Chromium-only and its comment states the Copy menu is covered
cross-browser by this test, so skipping would leave the menu with no
Firefox coverage at all.
## Verification
Test pinned run is green across chromium, firefox and webkit. Contrast
checked with the theme linter's contrast report.
## Overview
First part of the move over to the new designs. This lays the groundwork
(shared brand components, button/nav styling, theme tokens) and applies
it across the editor and the processor. Later parts will build on top of
it.
## What's changed
**Branding**
- Shared `Logo` and `BrandMark` components used everywhere, so the mark
and wordmark are identical across the editor, processor, auth pages and
the chat FAB.
- The sidebar logo doubles as the editor to processor switcher, morphing
into a chevron on hover. It only appears for users who can actually
reach the processor.
**Navigation and layout**
- Both sidebars restructured onto the floating nav surface treatment,
with rounded panels sitting on the app canvas.
- The editor file sidebar is now three sections (controls, PDF Library,
settings) and the workbench top bar and tools panel match.
- Added a collapse toggle to both sidebars, with an animated expand and
collapse and a tidy icon rail when collapsed. The processor did not have
a desktop collapse before.
**Components**
- Buttons and action icons now share one styling system, so both react
to the same tokens.
- Secondary buttons in dark mode use a neutral fill and border instead
of inheriting the primary colour.
- Status badges default to a clean dot with no background, with a filled
pill as the alternative.
- Metric strips gained a row layout with an optional leading icon.
**Theme**
- Colour tokens consolidated. Literal colours live only in the palette
file, everything else references the semantic `--c-*` tokens.
- `saas-theme.css` removed and the parts that were genuinely needed
moved into the shared theme, so all builds get them.
- The colour linter enforces this across the app and runs in CI.
## Notes
- Nothing functional should change here, it is styling plus the sidebar
collapse feature.
- Main has been merged in. The Sources and billing pages picked up
changes from main during that merge and are worth a look alongside the
new styling.
---------
Co-authored-by: Reece Browne <74901996+reecebrowne@users.noreply.github.com>
Fixes#7189
Consolidates all 29 `no-duplicate-selectors` violations across 18
stylesheets by merging later duplicate rule blocks into the first
occurrence. Where duplicates had conflicting values, the cascade-winning
(later) value was kept, so computed styles are unchanged.
Also:
- Adds `frontend/stylelint.config.mjs` with only
`no-duplicate-selectors` enabled (Prettier and the theme linter own
everything else).
- Adds a `frontend:lint:css` task, wired into `task frontend:lint` as a
blocking check so regressions can't creep back in.
- Lints all first-party CSS (`editor/**/*.css`, so `public/css` and any
future non-`src` stylesheets are covered too), excluding only the
vendored `cookieconsent.css` and build output via `ignoreFiles`. This
surfaced and fixed 5 additional duplicate selectors in
`cookieconsentCustomisation.css` that weren't in the original issue
report.
Note: `portal/views/Sources.css` goes beyond dedup — the whole
`.portal-sources__connections*` block is deleted as dead code
(unreferenced since the S3 connections redesign in #6965; only
`-actions` in it was an actual duplicate).
🤖 Generated with [Claude Code](https://claude.ai/code)
Follow-up to #7009, which built the theme token layer (`primitives` →
`colors` → `compat`). This PR moves the whole app onto it, removes
hardcoded colours, turns on enforcement so they can't come back, and
adds a user-selectable accent.
## What this does
- **Semantic tokens everywhere** — legacy colour aliases and raw hex are
rewritten to `--c-*` tokens (`--c-surface*`, `--c-text*`, `--c-primary`,
…). Straight rename, no visual change. Genuine literals (brand/OAuth,
colour pickers, data-viz) are left as-is.
- **Fixes missing colours** — some tokens the migration referenced were
never defined, so a few surfaces (login button, auth banners, badges,
procurement view) silently lost their colour. All defined now, and they
adapt to light/dark and the accent automatically.
- **Blocking colour lint** — CI now fails on hardcoded colours,
undefined tokens, or unreadable low-contrast status colours.
- **User-selectable accent** — light and dark each get their own accent
from Settings → Appearance, contrast-clamped so text stays legible.
"Default" keeps the standard blue.
## Still to come
Remaining inline-style hex, the legacy token-definition files
(`theme.css`, `tokens.css`), and folding `zIndex.ts` onto the dimension
tokens.
## Testing
`task frontend:check:all` green; light/dark and accent switching
spot-checked.
Simplifies the processor home page:
- Removes everything below the processor flow (processing status strip,
recent activity, quick actions, policy summary), leaving just the
onboarding hero and processor flow.
- Changes the policy "Set up" buttons (Security/Classification) from
primary to secondary variant.
<img width="2056" height="1047" alt="Screenshot 2026-07-20 at 7 27
56 PM"
src="https://github.com/user-attachments/assets/8625b274-b52a-47be-9052-80ac3d32dd93"
/>
## Summary
Follow-up polish to the home **PDF Processor** flow visualiser (base
landed in #7014). Tunes the particle animation to react to real volume
and adds a Storybook playground to tune it live.
## Changes
### Particle emission — `useFlowParticles.ts`, `flowTypes.ts`
- Emission rate now scales ~linearly with a source's 24h volume (**2×
volume ≈ 2× dots**) instead of the flat `rate / 86400 × SPEED`, capped
at **one dot / 250ms** (`MAX_EMIT_PER_SEC`) for busy sources (~≥
800/24h).
- Bounded the spread so the busiest source emits at most **5×** the
quietest (`EMIT_SPREAD_CAP`) — a dominant source can't starve the
others.
- Wider departure jitter (`[0.4×–1.7×]` the mean, still floored at the
per-source min-gap) and ~**2× faster travel** so the flow reads
livelier.
- Replaced the single `SPEED` constant with `EMIT_DIVISOR` /
`MAX_EMIT_PER_SEC` / `EMIT_SPREAD_CAP`. Weighted round-robin outcome
split is unchanged (e.g. 3 failed / 30 delivered → ~1 red dot in 11).
### Storybook Playground — `ProcessorFlow.stories.tsx`,
`ProcessorFlow.tsx`
- New **Playground** story with live controls: per-input rate sliders,
the delivered/failed split (drives the red-dot ratio), and a
Classification-active toggle.
- Added an optional `dataOverride` prop (prod-inert testing seam) so the
story renders a supplied flow model instead of fetching — changes apply
instantly.
### Housekeeping
- Condensed authored comments across the feature to ≤ 2 lines.
## Testing
- `task frontend:check` green — lint, typecheck, 1353 tests.
- Verified emission numerically (proportionality, 250ms ceiling, 5×
spread cap) and confirmed live animation in a focused Storybook tab.
## What this does
Consolidates the frontend's colour/theme system into a small,
well-defined token layer and reworks the theme picker. The goal was a
minimal, scalable set of semantic tokens that the editor **and** the
Processor/portal (and Storybook) all share, plus a theme model that's
easy to reason about.
## Token architecture (`core/theme/`)
A four-file layer, imported once via `index.css`:
| File | Role |
|---|---|
| `primitives.css` | The raw palette — the **only** place literal
colours live (neutral ramps `--p-gray-*`/`--p-zinc-*` + status hues). |
| `colors.css` | ~21 semantic `--c-*` tokens (surfaces, text, borders,
primary, status) mapped from primitives per theme. **Reference these.**
|
| `compat.css` | Legacy names (`--bg-*`, `--text-*`, `--color-*`)
aliased onto `--c-*` via `:root:root` so ~200 existing files keep
working. |
| `dimensions.css` | All non-colour tokens (spacing, radius, z-index,
type, motion) — single source, resolving prior collisions. |
A blocking linter (`scripts/lint/theme-lint.mjs`, run in
`frontend:lint`) enforces "literals only in `primitives.css`" within
`core/theme/`, and has a non-blocking WCAG contrast report. See
`core/theme/README.md`.
## Theme model
- **Mode** (`light` / `dark` / `system`) and **accent** are independent.
Each mode has its own accent (`lightPrimary` / `darkPrimary`).
- The editor is always `data-app-theme="custom"`; `ThemeProvider`
injects the accent as `--user-primary` and sets `data-accent`.
- **Two accent states:**
- A **colour** (preset or custom hex) → tints every surface that hue
(whole-app theming).
- The **`default`** sentinel → neutral surfaces (white/grey light, zinc
black/grey dark) with blue buttons, no tint. (`data-accent="default"`
opts surfaces out of the tint.)
- Accent contrast guardrails (`utils/customPrimary.ts`): lightness
clamps so an accent can't collapse into the base, a contrast-picked
on-primary foreground, and an accent-as-foreground variant so accent
text is never dark-on-dark.
## Theme picker (Settings → General)
- 3×5 grid: a distinct **Default** icon chip (not a colour) + 14 curated
accents, in a dropdown per mode.
- **Custom** colour via the shared `ColorInput`, with a live gamut clamp
(`clampValue`) that refuses white/grey/black — the picker handle sticks
at the boundary and preserves the working hue at achromatic extremes.
- "Restore theme to default" resets both modes.
## Other
- Dark mode is a true neutral zinc (no navy "midnight" tint); the
Mantine dark ramp and Tailwind dark channels were neutralised to match.
- Pre-paint inline script in `index.html` applies theme + accent before
first paint (no FOUC); portal and editor now share the same
`preferences.theme` source of truth.
- High-visibility surfaces migrated to tokens (FAB, landing upload
buttons, portal hero banners); scattered per-component colour swaps were
intentionally **left for a follow-up** to keep this PR focused.
## Testing
- `task frontend:check:all` (typecheck all variants + eslint + prettier
+ colour-lint) green.
- Verified light/dark, default vs tinted accents, and the custom clamp
via computed styles in the dev preview.
> Note: the `prerender-og` build step failing in the e2e/deploy jobs is
unrelated to this diff — it's in `vite.config.ts` (untouched here) and
builds cleanly locally.
## Overview
Adds a **Classification policy** to the processor's policy catalogue,
set up the same way as the Security policy. This moves classifier
configuration out of the editor (where the labels UI landed in #6898 and
was then removed with the rest of the editor's policy-management surface
in #6932) and into the processor, which is now the single place policies
are configured.
## What it does
- **Classification card** in the processor policy catalogue. Always
shown, but **setup is locked until the backend reports the AI engine is
on** — so admins can see the capability they're missing rather than it
being hidden entirely.
- **Setup wizard** mirrors Security: the workflow step shows the team's
**classification label editor** (reused
`LabelsEditor`/`LabelsEditorModal` — add box, chip grid, per-label icon
picker, import/export, reset) instead of tool toggles, since classify is
a single non-configurable step.
- On enable, the team's label vocabulary is **seeded with the 268
built-in defaults** (clobber-safe: only when the team has none). On
upload the document is classified against the team's labels and tagged;
on SaaS with the engine on, files group by category in the editor
sidebar.
## Reuse & consolidation
- Reuses the existing labels table, `labelsFile` helpers, and default
vocabulary. Labels read/write through the processor's own
`apiClient.local` (not the editor's axios client) so auth/base routing
stays explicit; the wire shape is shared.
- Consolidates policy-category icons into a shared, **id-keyed**
`policyCategoryIcon` util (outline glyphs) used by both the editor and
the processor, replacing the processor's emoji-glyph map (and the stray
`schedule` key that rendered a bare dot).
## Testing
- `task frontend:typecheck:{core,proprietary,portal}`,
`frontend:lint:eslint`, `frontend:test` (156 files / 1305 tests) — all
green.
- Verified in Storybook: the Classification card renders, the setup
wizard shows the label editor (268 defaults), and the full labels editor
opens with icons/import/export/reset. Added an MSW handler for the
app-config + labels endpoints and a `Classification` wizard story.
## Notes for reviewers
- The AI-engine gate reads the public `/api/v1/config/app-config`;
classification labels use `/api/v1/classification/labels` (team-scoped,
team-lead/admin-gated, `policies.enabled`); the classify step hits
`/api/v1/ai/tools/classify-and-label` — all pre-existing backend from
#6898.
- Known parity behavior (matches the editor hook): a transient failure
loading team labels falls back to showing the defaults; not changed here
to avoid diverging the two hooks.
The external-link "Developer Tools" buttons (API, Automated Folder
Scanning, SSO Guide, Air-gapped Setup) used `p="sm"` while normal tool
buttons use `p="none"`, making them render larger; this aligns their
padding so they match the size of every other tool button.
<img width="308" height="196" alt="Screenshot 2026-07-10 at 5 01 40 PM"
src="https://github.com/user-attachments/assets/fb125500-28fb-4b83-85ed-2edc12e66fc0"
/>
## What this does
Adds one test to `ResourceAccessServiceTest`: a foreign team's lead is
**denied** on a team-owned resource under the `ADMINS_AND_TEAM_LEADS`
default policy, even when an unscoped `isAnyTeamLeader` check would
admit them (stubbed `lenient()` to `true` precisely so the test fails if
the scoped path ever consults it again).
## Why
Main is already correct here — no behaviour changes in this PR. #6913
landed the scoped implementation (`matchesTeamLeadDefault`: ownerless
portal → `isAnyTeamLeader`, team-owned → `isLeaderOfTeam`), which
superseded #6893. The only piece not carried over was #6893's boundary
test, so the cross-team scoping isn't currently pinned by any test. This
adds that pin as cheap insurance for future refactors.
Verified the test does its job: it passes on main as-is, and fails if
the scoped check is swapped back to the unscoped one.
## Test plan
- `:proprietary:test --tests
"stirling.software.proprietary.access.service.ResourceAccessServiceTest"`
— green
- Spotless applied
Closes the loop on #6893.
## Summary
A batch of shared **design-system** fixes (Button, SegmentedControl,
Chip, a new CarouselDots) and the consumer/call-site cleanups they
unlock, following the button consolidation (#6787). Also includes
dark-theme token alignment and some portal/auth polish that rides on the
same components.
The shared Button now sizes to its content instead of clipping it, gains
per-axis padding controls, and no longer misbehaves while loading or
disabled; several call sites are then migrated onto the proper component
APIs.
## Shared components (`core/ui`)
### Button
- **Content-driven height.** `--button-height` is now a `min-height`,
not a fixed cap. Single-line buttons still land exactly on the shared
control-height scale (pixel-aligned with `ActionIcon` /
`SegmentedControl`), while taller content — wrapped labels, stacked
title + subtitle rows — grows the button instead of being clipped
mid-glyph. Short content is re-centered with `align-content`,
**without** overriding the root `display`, so a consumer's own layout
(e.g. a full-width list row) isn't disturbed.
- **Padding props.** New `p` / `px` / `py` props
(`none`/`xs`/`sm`/`md`/`lg`/`xl`) override the size-based padding per
axis. Vertical padding is applied through a `--sui-btn-py` CSS variable,
so consumers can also set it from their own class.
- **Loading no longer collapses.** A `fullWidth` button is never treated
as icon-only, so an execute button whose label is momentarily absent
while files hydrate (e.g. `ScopedOperationButton`) keeps its full width
with a centered spinner instead of shrinking to an icon-sized square for
a split second.
- **Disabled in dark mode.** A disabled *primary* button keeps a muted
version of its own accent fill (`opacity: 0.55`) instead of Mantine's
near-black `--mantine-color-disabled`, which blended into dark surfaces
and made the button all but disappear. Loading spinners are excluded so
they stay full-strength.
No breaking API changes — buttons that don't opt in render exactly as
before.
### SegmentedControl
- Fixed a bug where a segment marked `disabled` that also happened to be
the currently-selected value was rendered disabled, leaving the active
segment un-selectable/greyed. A disabled option is now only disabled
when it isn't the current value.
### CarouselDots (new)
- New shared dots indicator component (with Storybook story), used by
the login carousel.
### Chip / theme
- Dark-theme tokens in `theme.css` aligned to the portal's `tokens.css`
so the editor and portal (Processor) dark modes stop drifting (chrome
surfaces lift off the darker canvas); plus a Chip dark-mode styling fix
and a small `mantineTheme` cleanup.
## Consumer / call-site cleanups
- **Compare** tool: the swap control is now a regular shared Button
placed **between** the Original and Edited file cards (the bespoke
full-height vertical swap button and its CSS were removed), and the file
cards fill the full available width.
- **Certificate format**: replaced the inline-styled buttons with clean
two-state (primary / secondary) buttons.
- **ToolPicker**: restored the label selectors that #6787 renamed to the
never-emitted `.sui-btn__label`, and fixed the sidebar-search row
clipping.
- **File sidebar**: "View all files" row fix; `FileSidebarFileItem`
migrated off `display:flex` + `gap` on the Button root (which no longer
reaches the nested label) onto `leftSection` / `rightSection` + a
stacked label.
## Portal / auth polish
- Portal button consolidation and styling across Header, SettingsModal,
Home, Infrastructure, ApiKeyCard, and PopularUseCases.
- **Login**: onboarding text now shows the default starting username /
password; login carousel uses the new CarouselDots; desktop OAuth
styling tweak.
## Verification
- Storybook: button sizes measure exactly on the control-height scale
and match `ActionIcon`; icon-only buttons stay square and centered;
`fullWidth` loading buttons hold full width; disabled dark-mode primary
buttons render as a muted accent rather than grey.
- Single-line buttons are pixel-identical before/after; only buttons
whose content previously overflowed a fixed height render differently
(they now fit rather than clip).
- `task frontend:lint` clean; typecheck shows only the pre-existing
third-party `node_modules` noise also present on `main`.
## Overview
Adds **AI document classification** and a **classification-aware Files
sidebar**: uploaded documents are automatically tagged with
document-type labels (Invoice, Contract, Lab report, …), and the sidebar
groups files under editable parent categories so a large library stays
navigable.
> [!IMPORTANT]
> **This feature only runs in the SaaS build.** Classification depends
on the AI engine and team-scoped label storage, so it's gated to SaaS
end-to-end:
> - The sidebar grouping is a `saas/`-layer override of the
`fileSidebarGrouping` seam; every other build (OSS core, self-hosted
proprietary, desktop) gets the null stub and renders the **unchanged
flat, recency-sorted list** — no categories, no "Other", no picker.
> - The classify/labels backend endpoints are gated on
`policies.enabled` (on in SaaS) and live in `app/proprietary`, so
they're absent from pure OSS and dormant in self-hosted unless
explicitly enabled.
> - The Python classifier is reached only via that gated path.
>
> Shared-layer changes that do compile everywhere are inert without the
engine (dormant schema/field additions) or intentional (`GetInfoOnPDF`
surfacing custom metadata).
## What it does
- **Classifier (engine):** reads the first/last two pages of a PDF and
assigns document-type labels from an allowed vocabulary. Labels are
deliberately document-*type* descriptors — no deep-content/PII
detection, since only a page window is read.
- **Team label vocabulary:** ~270 built-in defaults across ~15 families,
seeded per team. Editable by team leaders/admins in the Classification
policy settings (import/export/reset). Team-scoped and shared;
**per-user personal labels are intentionally out of scope** — the
vocabulary is team-level only.
- **Sidebar categories:** files group under parent categories
(Financial, Legal, Medical, …), busiest-first, collapsible, with a
"Recent" group on top and an "Other" group for anything uncategorised.
The category structure (names, icons, membership, custom categories) is
**device-local and user-editable** via a "Customize" picker — the only
per-user personalization; it never changes the team's label vocabulary.
- Classification results are written to PDF metadata
(`StirlingPDFClassification`), read back to keep files in their groups
without re-parsing.
## Architecture
Spans all three layers, mirroring the existing policy/source subsystem
conventions:
- **`frontend/editor`** — sidebar grouping seam + SaaS override,
category manager, labels editor, icon palette, file grouping, tests,
`en-US` i18n.
- **`app/proprietary` + `app/common` + `app/core`** —
`ClassifyLabelController`, team-scoped `ClassificationLabelStore` (Jpa +
in-process impls, same shape as `PolicyStore`/`SourceStore`), metadata
read/write.
- **`engine`** — the document-classifier agent, contracts, routes,
tests.
## Screenshots
**Files sidebar — grouped by category (SaaS)**
### Loading view
<img width="2056" height="1046" alt="Screenshot 2026-07-07 at 5 12
56 PM"
src="https://github.com/user-attachments/assets/1d712da5-50ae-4349-b0cd-e62665c3ec0c"
/>
### Organized in the sidebar
<img width="2056" height="1045" alt="Screenshot 2026-07-07 at 5 14
05 PM"
src="https://github.com/user-attachments/assets/3ea4fe21-da51-4cea-bc3a-18ce040d3d05"
/>
**Customize categories picker**
### Personal settings to change how labels are grouped in an individual
users editor
<img width="2056" height="1044" alt="Screenshot 2026-07-07 at 5 52
42 PM"
src="https://github.com/user-attachments/assets/40be03ce-0f63-4d1e-b58b-cec045d01cb2"
/>
**Classification labels editor (team settings)**
<img width="2056" height="1042" alt="Screenshot 2026-07-07 at 5 53
00 PM"
src="https://github.com/user-attachments/assets/337b0739-15c9-4749-9c6b-22e3b20825b8"
/>
## Testing
- Frontend `task frontend:check` — green (editor + portal tests,
typecheck across all flavors, lint, label-drift guard).
- Backend `task backend:check` (proprietary) and `:saas:test` — green.
- Engine `task engine:check` — green.
• Removed colors from policies to make them look more professional.
• upgraded to enterprise link to contact us.
• Hid inactive policies from users (Kept for admin and team lead).
• Closing policies had wrong arrow, made a standard component for chat,
tools and policies header.
## What
Make the **SaaS** build always use the modern logo, so the classic logo
can no longer appear anywhere in the SaaS app.
This is a minimal, SaaS-only alternative to the full classic-logo
removal PR (~80 files). **OSS (`core`) and proprietary builds are
untouched** — they keep the full modern/classic variant system,
including the admin _Logo Style_ picker.
## How
A single SaaS-layer override shadows the core hook:
- `frontend/editor/src/saas/hooks/useLogoVariant.ts` → returns
`"modern"` unconditionally.
In the SaaS build the `@app/*` alias cascade resolves
`@app/hooks/useLogoVariant` to `src/saas/*` before `src/core/*`, so this
shadows the core implementation (which otherwise resolves the variant
from the stored user preference or the server `logoStyle`).
## Why one file is enough
All logo rendering funnels through `useLogoVariant()`:
- `useLogoAssets()` → favicon, web manifest, apple-touch icon, wordmark,
`logo512`, tooltip logo — consumed by `BrandingAssetManager` (which sets
`<link rel="icon|manifest|apple-touch-icon">`), `Wordmark`, `LogoIcon`,
`Tooltip`.
- `useLogoPath()` → the no-text logo SVGs.
- The login-carousel slides (`buildLoginSlides`) receive the variant
from `AuthLayout`, which calls the hook.
Everything else that references a logo in SaaS already hardcodes
`modern-logo` (`index.html`, the SaaS
`Login`/`Signup`/`AuthCallback`/`OAuthConsent` routes, cloud onboarding,
account/MFA QR logos).
The only hardcoded `classic-logo` reference — the admin _Logo Style_
picker in `AdminGeneralSection` — is **not shipped in SaaS**:
`createSaasConfigNavSections` builds from the core nav sections and
never includes the proprietary admin sections.
`manifest-classic.json` and the classic assets remain in the shared
`public/` folder (served by all builds) but are never referenced in the
SaaS bundle.
## Test plan
- [x] `task frontend:typecheck:saas` — clean
- [x] `eslint` on the new file — clean
AI PDF creation ("create a PDF for me") has been broken since the
Policies backend (#6527) introduced PolicyExecutor as the tool execution
pipeline. PolicyExecutor runs normal single-input tools with a per-file
loop, but generator tools like `create-pdf-from-html-agent` take no
input file and build their output purely from parameters. With zero
input files the loop ran zero times, so the endpoint was never called
and the step silently produced nothing. The chat reported success
("Created Purchase Order") while no document ever appeared.
This adds an `else if (inputFiles.isEmpty())` branch so a generator tool
is called once with an empty file list, matching what the multi-input
branch already does for an empty input. Two files changed: the
one-line-ish fix in `PolicyExecutor`, and a regression test covering the
no-input case.
---------
Co-authored-by: James Brunton <jbrunton96@gmail.com>
Fix issues with the theme of the app that caused some things to persist
in light mode/dark mode whilst the rest of the app was the opposite
theme.
Removed dead rainbow mode code.
Added system theme option to settings.
add en-US changes to SaaS, previously merged into main. So this is
effectively a main -> SaaS PR also. It seems to be all additive.
Also take the 230 ish missing translations from en-GB over to en-US
using a script, and also make and english spellings American when adding
them to the en-US file, and fix any existing American spellings in the
en-GB file.
Added the create agent. Use [these
prompts](https://github.com/Stirling-Tools/Stirling-PDF-SaaS/blob/main/docgen/backend/default_templates/sample_prompts.md)
to test or try your own :)
Here’s the one I use
```
Hey, I need to generate an employee expense report for reimbursement.
Company: Summit Consulting Partners Company address: 88 Riverside Plaza, Suite 1400, New York, NY 10069 Accounting department email: expenses@example.com
Employee details:
* Employee Name: Michael Tran
* Employee ID: EMP-1047
* Department: Client Services
* Report Date: January 20th, 2026
* Reporting Period: January 5th, 2026 – January 16th, 2026
* Manager Approver: Laura Simmons
Trip purpose: Client onsite meetings with Atlantic Energy Solutions in Boston, MA.
Expense items:
* Flight (NYC to Boston roundtrip) — $325.40 — January 5th, 2026 — Airline ticket
* Hotel (3 nights at Harborview Hotel) — $822.75 — January 5th-8th, 2026
* Taxi from airport to hotel — $48.00 — January 5th, 2026
* Client dinner (3 attendees) — $186.20 — January 6th, 2026
* Parking at JFK Airport — $72.00 — January 5th-8th, 2026
* Breakfast (per diem not used) — $18.50 — January 7th, 2026
* Uber to client office — $22.10 — January 7th, 2026
* Printing + presentation materials — $46.90 — January 8th, 2026
* Lunch with client — $39.75 — January 8th, 2026
* Office supplies (notebooks, pens) — $27.60 — January 10th, 2026
* Mileage reimbursement (client visit in NJ, 42 miles @ $0.67/mile) — $28.14 — January 14th, 2026
* Team lunch meeting (internal) — $64.30 — January 15th, 2026
Reimbursement method should be direct deposit.
Add a notes section stating: "All receipts attached. Expenses are business-related and comply with company travel policy."
```
---------
Co-authored-by: Anthony Stirling <77850077+frooodle@users.noreply.github.com>
## Summary
Addresses two review comments from #6507:
- **`timeUtils.ts`** — route relative time strings (`just now`, `Xm
ago`, `Xh ago`, `Xd ago`) through i18n by accepting a `TFunction`
parameter and using new `time.relative.*` keys in `en-GB`
- **`ChatPanel.tsx`** — replace `ReturnType<typeof useTranslation>["t"]`
with `TFunction` from `i18next`
## Test plan
- [x] `task frontend:check` passes (695 tests)
### To test
- Ask the agent to “list all the things you can do and put them in a
markdown table”. I know we’re explicitly asking it for markdown, but I
don’t want to update the system prompt to ask it to make tables when
necessary because it’ll probably turn everything into a table, not sure
though, we can test in future.
- Notice how the loading is different
- Notice how the user chat is in a bubble but the agent chat is flat
(super standard design practice in AI tools, and looks much better when
the agent outputs mardown, expecially tables and needs room to do so)
- Ask it to do something different, then close the chat, and see that
the agent is marked as running and has a green outline and a green dot.
- Play around with resizing the chat to make it bigger/smaller
Open to any and all criticisms on any of the design choices, and of
course the usual, code etc.
Resizing
<img width="1572" height="812" alt="Screenshot 2026-06-01 at 2 47 53 PM"
src="https://github.com/user-attachments/assets/ec0ac1d0-01da-4025-bf7e-eea4eb544181"
/>
Loading (cool animation not visible through screenshot obviously)
<img width="559" height="141" alt="Screenshot 2026-06-01 at 2 53 41 PM"
src="https://github.com/user-attachments/assets/99f0b1f5-1719-4d78-8947-21b142293052"
/>
Removed bubbles for agent chat (maybe controversial, let me know) and
markdown now renders properly again
<img width="654" height="1060" alt="Screenshot 2026-06-01 at 2 55 01 PM"
src="https://github.com/user-attachments/assets/445f0889-a632-4751-9a16-f80ae388c632"
/>
## Fix 1 — Viewer bug (8 tools)
8 tools called `useFileSelection()` directly instead of routing through
`useBaseTool`. In the viewer, this meant they operated on **all selected
files**
instead of only the one being viewed. For example: 10 files loaded,
viewing
file 3, running Add Stamp — all 10 files got stamped.
**Root cause:** These tools had no view-scope awareness.
`useFileSelection()`
returns the raw workbench selection with no knowledge of which file is
active in
the viewer.
**Fix:** A new hook `useViewScopedFiles` was introduced:
```ts
// Viewer → only the active file
// Everywhere else → all loaded files
const selectedFiles = useViewScopedFiles();
```
The 8 tools were updated to call this instead of `useFileSelection()`.
**Tools fixed:** Add Stamp, Add Watermark, Add Password, Add Page
Numbers,
Add Attachments, Reorganize Pages, OCR, Convert
---
## Fix 2 — Page selector / active files context (all tools)
`useBaseTool` returned `selectedFiles` (checked files only) in
non-viewer
contexts. In the page selector this is typically empty or stale — not
the full
set of loaded files that tools should operate on.
**Fix:** `useBaseTool` was updated to use `useViewScopedFiles`, which
returns
all loaded files in non-viewer contexts. This affected every tool via
`useBaseTool`.
---
## Workarounds for Compare & Merge
Two tools intentionally need all loaded files regardless of view, so
they use
`ignoreViewerScope: true` in `useBaseTool`.
**Compare** — needs exactly 2 files for its Original/Edited slots.
Scoping to
one file would break the comparison entirely. `ignoreViewerScope: true`
is set
and `disableScopeHints: true` hides the "(this file)" button label hint.
The
slot auto-mapping logic was also improved alongside this fix.
**Merge** — needs 2+ files; merging a single file is meaningless. Rather
than
leaving the button silently disabled, Merge now:
- Auto-redirects to the active files view on first open from the viewer
- If the user navigates back to the viewer, shows a disabled button with
a hint
and a "Go to active files view" shortcut button
---
## How to Test
---
## Fix 1 — 8 tools (viewer scoping)
### Test steps (same for each)
1. Load 3 PDFs into workbench
2. Open viewer, navigate to file 2
3. Open the tool, configure settings, run
4. ✅ Only file 2 is in the results
5. ✅ Button label shows **"[Action] (this file)"**
6. ✅ A note below the button reads **"Only applying to: [filename]"**
| Tool | What to configure |
|---|---|
| **Add Stamp** | Enter any text stamp or upload an image stamp |
| **Add Watermark** | Select text watermark, enter any text |
| **Add Page Numbers** | Leave defaults |
| **Add Password** | Enter any owner + user password |
| **Add Attachments** | Attach any small file |
| **Reorganize Pages** | Enter a page range e.g. `1,2` |
| **OCR** | Leave default language |
| **Convert** | Convert PDF → any format |
---
## Fix 2 — All tools (page selector context)
### Test steps
1. Load 3 PDFs into workbench
2. Open the page selector view
3. Open any tool from the sidebar, run it
4. ✅ All 3 files are processed (not zero or a stale subset)
---
## Compare (intentionally ignores view scope)
**A — Auto-fill with exactly 2 files**
1. Load exactly 2 PDFs
2. Open Compare from either the viewer or active files view
3. ✅ Both slots are filled automatically (Original + Edited)
4. ✅ No scope hint appears on the button
**B — Manual selection with 3+ files**
1. Load 3+ PDFs
2. Open Compare
3. ✅ The first 2 files fill the slots
4. ✅ A 3rd file does not add a 3rd slot (capped at 2)
**C — File removed mid-session**
1. Load 2 PDFs, let Compare auto-fill both slots
2. Remove one file from the workbench
3. ✅ The corresponding slot clears; the other slot is unchanged
**D — Viewer mode**
1. Load 2 PDFs, open viewer
2. Open Compare from the viewer sidebar
3. ✅ Both files are still available for slot selection (not scoped to
current file)
---
## Merge (intentionally ignores view scope, disabled in viewer)
**A — Auto-redirect on first open from viewer**
1. Load 2+ PDFs, open the viewer
2. Open Merge from the viewer sidebar
3. ✅ Immediately redirected to the active files view
**B — Viewer mode disabled state (after navigating back)**
1. From the active files view, open Merge, then navigate back to the
viewer
2. ✅ Execute button is **disabled** with tooltip "Switch to the file
editor to select multiple files"
3. ✅ A note appears: *"Merge needs 2 or more files. Head to the file
editor to select them."*
4. ✅ A **"Go to active files view"** button is shown; clicking it
navigates back
**C — Active files view works normally**
1. Load 3 PDFs, open Merge from the active files view
2. ✅ All 3 files appear in the merge list
3. ✅ Button shows **"Merge (3 files)"**
4. Run the merge
5. ✅ Output is a single PDF containing all 3 files
---
## Button label behaviour (all tools)
| Context | Expected button text |
|---|---|
| Viewer, 1 file loaded | `[Action]` (no suffix) |
| Viewer, 2+ files loaded | `[Action] (this file)` |
| Active files view, 1 file loaded | `[Action]` (no suffix) |
| Active files view, 2+ files loaded | `[Action] (N files)` |
| Merge in viewer | disabled — no suffix |
| Compare | never shows scope suffix (`disableScopeHints: true`) |
---------
Co-authored-by: Reece Browne <74901996+reecebrowne@users.noreply.github.com>
* Text box/notes movement improvements ✅
* Fix the issue where hiding, then showing annotations looses progress ✅
* Fix the issue where hidig/showing annotations jumps you back up to the
top of your open document ✅
* Support ctrl+c and ctrl+v and backspace to delete ✅
* Better handling when moving to different tool from annotate ✅
* Added a color picker eyedropper button ✅
* Auto-switch to Select after note/text placement, so users can quickly
place and type ✅
I have been trying to get this perfect for a while, but this is the best
I can do. I tried to get the open files button to be sticky at the
bottom, but beyond certain zoom levels, this causes it to disappear. Not
worth spending more time on it right now.
Co-authored-by: James Brunton <jbrunton96@gmail.com>
Added an optional flag in settings.yml to hide the settings button in no
login servers. When hidden, users can no longer:
- Open the Settings modal at all (gear button is hidden)
- Change General preferences (tool picker mode, hide unavailable
tools/conversions, auto‑unzip and file limit)
- Configure keyboard shortcuts (Hotkeys / Keyboard Shortcuts section)
- Use the in‑app update checker UI (see current/latest version, check
for updates, view update details)
- Note: When enableLogin === true, the flag is ignored and the Settings
button remains visible.
---------
Co-authored-by: Anthony Stirling <77850077+Frooodle@users.noreply.github.com>
Co-authored-by: Reece Browne <74901996+reecebrowne@users.noreply.github.com>
Also added `enableDesktopInstallSlide` flag in `settings.yml` to hide
the download for desktop page in the onboarding.
---------
Co-authored-by: James Brunton <james@stirlingpdf.com>
EthanHealy01