The grounding metadata went to the API response and the logs, which is no use to
someone holding the deck. A teaching resource shown to trainees should carry its
own provenance, so a grounded one now ends with a References section — the final
slide in a presentation — listing the library excerpts it actually used, by
title and page.
Restricted deliberately: only excerpts actually drawn on, nothing invented. That
was worth checking rather than trusting. Generated a deck and compared every
citation against the source metadata: "Kliegman R. Nelson Textbook of
Pediatrics, 22nd ed., 2024, p. 2604" against a stored title of "Kliegman R.
Nelson Textbook of Pediatrics 2-Volume Set 22ed 2024" at page 2604, and the same
for Fleisher & Ludwig, Rosen's, Understanding Pathophysiology and the AAP
compendium. The model reformatted filename-derived titles into readable
citations using only what it was given.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
Learning generated everything from the model alone. A deck on bronchiolitis was
whatever the model remembered about bronchiolitis, with no connection to the
documents this institution actually indexed — while the assistant had been
searching that corpus all along.
Same collection, deliberately. mcp_bge_m3_1024 is already embedded with
openrouter-bge-m3 at 1024 dimensions; a second index over the same documents
with the same embedder would be a copy that drifts. What differs is the budget:
a chat answer wants a few tight excerpts because the reader is waiting, a
teaching resource synthesises a whole topic. So learning.search_limit and
learning.context_chars default to 30 and 2500 against the assistant's 8 and
1400, and are separate keys so tuning one cannot move the other.
Not unbounded, though. "No limit" only moves the ceiling from a setting to the
model's context window, where overflow truncates the middle of the prompt
silently — the worst place to lose source material. 60 results and 8000
characters per excerpt are the caps.
Opt in per generation: a resource on something the library does not cover is
better written without it than padded with the nearest unrelated excerpts.
Retrieval never fails a generation — the resource is then written from the model
alone, which is what happened before this existed — and every response reports
what it was grounded on, so a caller can say "24 excerpts" or "the library had
nothing on this" rather than quietly serving ungrounded material.
Verified against the live corpus: bronchiolitis, neonatal jaundice and febrile
seizure each returned 12 excerpts and ~23k characters from Nelson, Rudolph and
the Pediatric Clinical Practice Guidelines. A deck generated through the full
chain came back with textbook specificity that is not general recall —
bronchiolar diameter, birth-weight thresholds, the full pathogen list.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
npm audit reports 0 vulnerabilities. It reported 2 high and 2 moderate this
morning.
@google-cloud/vertexai was the last source of findings — gaxios and a uuid with
a missing buffer bounds check, neither reachable in this deployment because
GOOGLE_VERTEX_PROJECT is unset and the require sits inside that check. Dormant
is not the same as gone, and the provider is available through the gateway
anyway, so the direct path has been removed rather than left to rot:
- the SDK client and callVertex, which without the package could never run
- the dispatch and discovery branches that reached them
- VERTEX_MODELS, a list of ids nothing could route any more, and the two
places in adminConfig that concatenated it into the built-in set
- the health endpoint's vertex line, and the env vars documented for it
AI_PROVIDER=vertex now says where to configure the model instead of quietly
becoming something else. The Google STT and TTS paths keyed off the same
variable are untouched; neither ever used this SDK.
Verified after deploy: provider litellm, the assistant answers with 8 sources,
/api/models returns 10, and @aws-sdk/s3-request-presigner — which documents.js
needs for presigned MinIO URLs — is still declared and resolvable.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
Generated a deck with ds-deepseek-v4-flash and rendered it to look at. The model
produced exactly the six headings it was asked for; the deck came out with eight
slides. The extra ones were pandoc's, not the model's.
Two rules, both found by rendering rather than reading:
- pandoc splits a slide after a table. Anything following one becomes a new
slide with no title — that was the stray "Key differentials to consider:"
slide floating with no heading.
- A table with no blank line before it is not parsed as a table at all. It
renders as literal pipe characters in the preceding paragraph.
And one that was visible on the slide itself: a nested ordered list inside a
bullet ran off the bottom.
None of these are the model failing. A cheap model writes perfectly good slide
markdown — bold, italics, nested lists and a table with a subscript all came
through correctly. It just needs to be told the shape the renderer wants.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
pptxgenjs is gone, and with it 269 lines of hand-rolled markdown parsing.
It stretched every image. Reading the slide XML it emitted shows why: it writes
the target box verbatim with <a:stretch/> and a no-op srcRect, so a 200x800
image handed an 11.8x3.9 box came out 1:4 squashed to 3:1. It could not do
better — it never measures an image, and its own getSizeFromImage is commented
out and marked "currently unused", reaching for a package called sizeof that
does not exist. pandoc measures them: a 300x175 source renders at aspect 1.714
and a 160x360 at 0.445, verified by rendering the deck to PDF and looking at it.
Tables, ordered and unordered lists, bold, italic and subscripts all come out
natively, and the fonts, palette and slide layouts come from
assets/learning/slides-reference.pptx. Design now lives in that file: restyling
the decks means editing it in PowerPoint, not editing this route.
Only images the requester owns can reach a deck. pandoc resolves an image link
against the filesystem, so a markdown link naming any local path would read that
file into the presentation. Images are fetched by id through the ownership
check, written into a per-request temporary directory under names we choose, and
every image link that did not resolve is removed rather than passed through. The
directory is removed in a finally block, and the conversion has a 60s timeout so
it cannot hang a request.
pandoc is in the image rather than a sidecar, because an export must not fail
for reasons outside this container. It costs 197MB (307 -> 504).
Removing pptxgenjs also removed image-size, and with it both high-severity
advisories — GHSA-w3rx-r6r6-pgpr and GHSA-5p2g-fcmc-qvqq, ICNS/JXL/HEIF parser
denial of service, ranged <=2.0.2 with no fixed release to upgrade to. npm audit
goes from 2 high and 2 moderate to 2 moderate.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
Every generated image in an exported deck was distorted. PowerPoint scales an
image to whatever extent it is given, and the export handed it the content box
verbatim. pptxgenjs has a `sizing: contain` option that looks like it solves
this; reading the emitted slide XML shows it does not — a 200x800 image in an
11.8x3.9 box came out as cx=10789920 cy=3566160 with <a:stretch/>, stretched
from 1:4 to 3:1. It cannot do better: it never measures the image, and its own
getSizeFromImage is commented out and marked "currently unused".
So the export measures the image itself and hands PowerPoint an extent that
already has the right shape, centred in the space available. Verified: a 200x800
image now places 0.97x3.90 and a 4x3 places 5.20x3.90, both matching their
source aspect exactly, neither overflowing. An image that cannot be measured
keeps the old behaviour rather than failing the export.
image-size becomes a real dependency rather than one borrowed transitively, and
an override collapses it to a single copy — pptxgenjs declares it but the string
appears in none of its four shipped bundles, so npm was placing a second copy in
the production image that nothing could load.
Its ICNS, JXL and HEIF parsers have open denial-of-service advisories against
every published version (GHSA-w3rx-r6r6-pgpr, GHSA-5p2g-fcmc-qvqq, both ranged
<=2.0.2), so there is no release to upgrade to. They are disabled instead:
this application measures PNG, JPEG, WebP and GIF and nothing else. An ICNS
buffer is now refused and falls back to the box rather than entering the parser.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
Removed at the owner's request, entirely rather than switched off: the route's
allow-list, anonymous identity and flag lookup; the client's entry path, the
authFetch exception that let four endpoints out without an account, and the
workspace guard; the CSS, the in-page note, the admin flag and its save/load;
the test file and the assertions elsewhere that pinned it. Both settings rows
are deleted from app_settings.
Two things were checked rather than assumed on the way out. Removing the
anonymous identity collapsed every `if (!req.user.preview)` branch to its
authenticated side, so image tools, audit logging and citation storage now run
unconditionally — which is what they did before preview existed. And the route's
gate went back to a bare router.use(authMiddleware), which on a /api mount gates
every path below it in server.js; it is scoped to /clinical-assistant again, the
guard test catches it either way.
Verified after deploy: signed out, status, examples and chat all refuse with
401; signed in, chat still answers with 8 sources; extensions, encounters,
documents and admin remain shut.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
The card holds more than one Save button and nothing said so. "Save image
settings" is injected directly above "Save model & retrieval settings", with no
indication of where one stops and the other starts, and the page saves nothing
automatically. It now states that up front, and the bottom button says which
settings it applies.
"Retry loading settings" sat beside Save looking like an ordinary control,
because it did: a bare button with the hidden attribute, which the browser's own
[hidden] rule could not hide once .btn-sm set a display. It is now inside an
error message that exists only on failure, says what failed, and says that
nothing typed has been lost.
The status line used to read "Settings ready." forever, which answers a question
nobody asks. It now reports the thing an admin actually wants to know when they
come back: whether the last save went through, and at what time. A toast is gone
in three seconds; this stays on the page.
The signed-out preview moves to Feature Flags, where it belongs. It was a second
checkbox under a row labelled "Sources", followed by two paragraphs, the first
about preview and the second about citations — so neither paragraph clearly
belonged to either checkbox. It is stored as feature.assistant_preview now, with
the old clinical_assistant.preview_enabled still honoured when the new key has
never been written. That also means an ordinary admin can toggle it under
ADMIN_LOCKDOWN: clinical_assistant.* is locked, and putting a day-to-day switch
behind host access was never the intent.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
Eleven routers are mounted on '/api' and called router.use(authMiddleware) with
no path. Mounted that way, the gate applies to every /api request that reaches
the router — including routes belonging to routers mounted further down
server.js. extensions.js did it from line 295; the assistant is mounted at 305.
So a signed-out request to /api/clinical-assistant/status was refused ten lines
before the preview middleware could look at it, whatever the admin setting said.
server.js line 250 already warned about this shape.
Each gate now names its own prefix, so a router protects its own routes and
nothing else. Verified afterwards that every namespace which must stay shut
still answers 401 signed out: extensions, encounters, memories, notes, diagrams,
generated images, image jobs, documents, audio backups, ED encounters,
don't-miss, patient education, billing, well visit, admin, transcribe and the
rest. Two of these routers were gating routes nobody realised they were gating.
Second defect in the same path: authMiddleware only ever looks for a token, so
calling it unconditionally after the preview identity had been assigned rejected
exactly the requests preview exists to serve. Only that identity may skip it;
authMiddleware stays strict everywhere else.
Preview now answers with a real cited answer, and stays as narrow as it was
designed to be — four allow-listed paths, no identity, nothing ownable. A test
now walks every /api router and fails on a blanket gate, which is how the last
six were found.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
The recordings were never the problem. Six stored recordings were pulled back
out of object storage and examined: valid webm/opus, 3-5 seconds, 1.5-2s of
continuous speech-shaped audio each. Every one came back from
mistral-voxtral-mini-transcribe as an empty string, while the same model
transcribed synthesised speech perfectly — including a one-word clip, and
including that speech attenuated to the same level, so neither length nor
loudness explains it. Re-encoding to wav, mp3, flac, ogg and a remuxed webm
changed nothing; groq-whisper-large-v3-turbo transcribed all six. stt.model is
set to that now, and the real recording round-trips through /api/transcribe as
"Hello." instead of "".
So the server now says something when a model answers 200 with no words for a
non-trivial amount of audio. That silence is what made this look like lost
recordings; the log names the backup id, so the kept audio can be tried against
another model directly instead of suspecting the microphone.
Also: browsers report "audio/webm;codecs=opus", and deriving the extension by
splitting on "/" alone named the upload "audio.webm;codecs=opus". This gateway
tolerates it. A provider dispatching on extension would not.
And registration is one card again: enable it, decide whether it needs an
invitation, hand out codes — top to bottom. The invite-only switch sat in a
separate card far below the enable/disable toggle, which made one decision look
like two unrelated settings.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
Measured in a real browser against the app rather than reasoned about.
The transcript boxes are contenteditable divs, and an editable div zooms on
focus exactly like an <input>. The earlier 16px sweep covered input, textarea
and select, so every workspace tab still zoomed while the calculators did not —
which is exactly what was reported. Every focusable text control in every tab
now measures 16px at phone width; the count of ones below it is zero.
Three ways a recording could end with nothing to show for it:
- Safari supports none of the audio/webm types and throws NotSupportedError
when handed one. Six modules built their own recorder on resume with
"opus, else audio/webm", so resuming threw there and the recording stopped.
There is now one codec chain in the app, and no module constructs a
MediaRecorder of its own.
- audio-recorder-failed is dispatched on document, and the encounter tab
stopped its recording on any of them. The assistant's microphone failing
ended a consultation being recorded in another tab. The recorder now
travels with the event and the listener checks it is its own.
- The server answers {success:true, text:''} for silence, and five modules
assigned that straight into the transcript — emptying the box the browser
had been filling live. It reads as a recording that vanished. Text is now
required before overwriting, and a recording that captured nothing says so
instead of resetting the button over an empty box.
Also: the citation counters were registered on prom-client's default registry
while the app serves its own, so they were never scraped. They read zero at
/metrics now instead of being absent, which is what the Grafana panels need.
And the reference linter passes for the first time, so scripts/e2e.sh gets past
its preflight: KaTeX is vendored (it was referenced by the assistant's LaTeX
rendering but never shipped — three 404s a page load and no math), and the
JavaScript left behind by the removed image picker, saved-chats toggle, image
gallery and visual-output panel is gone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
Citation quality
- A citation naming a source that never came back is never rendered as a
link, so it appears as plain text and nobody learns it happened. It is now
measured on the server, where the answer and the sources both exist, so it
is seen whether or not a browser rendered it.
- Four Prometheus counters feed a Grafana dashboard (Ped-AI Citation
Quality): answers, citations written, answers affected, and individual
unresolved markers. Only answers with at least one unresolved citation are
stored, with the question and the titles retrieval returned, so an operator
can judge whether retrieval came back thin or the model over-cited. Rows
expire after 30 days: this is a quality signal, not a transcript log.
- Both answer paths are covered. /chat/stream is normal; /chat is the
fallback the client uses when streaming fails, so auditing only the first
would have hidden exactly the answers produced under failure.
- The tracker is resolved on demand and allowed to be absent. Seven test
files load this route with a hand-built list of permitted imports, and
adding a hard dependency would mean editing all seven — and the eighth
written later would break. Observation must never be able to fail an
answer, so a missing module simply means no tracking.
- Metric registration reuses an already-registered counter, because this
module can legitimately load twice in one process.
SSO settings on mobile
- Six rows were laid out inline: flex with a 160px label and an input that
would not shrink, so on a phone the row was wider than the screen with
nothing to scroll and no way to reach the rest. They use .admin-row now,
which already stacks below 640px. Verified at 390px and 360px: nothing
off-screen, no sideways overflow.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
With more than one admin, every setting in the panel was editable by all of
them — prompts, model policy, retrieval budgets, SMTP, email templates.
ADMIN_LOCKDOWN separates running the service from changing how it behaves.
It is an environment variable on purpose: a setting could be switched off by
the very admin it restrains, so lifting this needs host access and a restart.
The server is the control. One gate refuses configuration writes rather than
a check in each of the fifteen write routes, because that list grows and a
route added later would quietly miss it. Reads always pass — lockdown hides
nothing. Day-to-day operation stays available: invitations, announcements,
registration, feature flags, and the test endpoints, which persist nothing.
A setting invented later is locked until someone deliberately makes it
editable, rather than defaulting to open.
The panel disables what it cannot save and says why, but that is courtesy;
the refusal is what enforces it.
Two things this taught me, both fixed: my first version painted the panel
from an IIFE, which the module conventions forbid, and fetched the whole
config a second time just to read one flag — breaking the test that pins
admin loaders firing exactly once. The state now rides on the invites
response the panel already requests.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
registration_enabled was a single switch: open to anyone, or closed to
everyone. This adds the setting an operator actually wants in between — open
to people you invited.
A code is single-use, expires (7 days by default, 90 maximum), and can be
revoked or deleted. It is stored hashed with only its last four characters
kept, because an invite grants account creation and a database dump should
not hand someone a working one. The code is readable exactly once, in the
response that creates it.
The claim is a single conditional UPDATE carrying every condition, so two
registrations racing the same code cannot both succeed. It happens after the
account exists, so a code is never spent on a failed registration — and if
the race is lost, the just-created account is removed rather than left behind
as a free registration. The rejection never says which of the four reasons
applied; distinguishing them would tell someone probing codes which guesses
were closer.
Codes avoid I, L, O and U so they survive being read aloud or copied off a
screen, and matching ignores case and separators.
The sign-up field appears only when the server says a code is required. The
admin card creates, lists, revokes and deletes, and carries the toggle.
Verified against the live database: create, claim, second claim refused,
unknown code refused, revoking a used code refused, delete. 684 tests pass.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
The nc_semantic_search alias is gone from the MCP server, so accepting it
here would point retrieval at a tool that no longer exists. A stale
override now stops the app at startup instead of silently retrieving
nothing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
Three S3 configurations had grown separately — S3_* for documents,
GENERATED_IMAGES_S3_* for images, and AUDIO_BACKUPS_S3_* after them — with
different key names and their own client construction. That is why moving
storage meant hunting through several files.
src/utils/objectStorage.js now resolves settings for any purpose: its own
variables first, then the shared S3_* ones, with a per-purpose bucket name
(S3_BUCKET_AUDIO_BACKUPS). One endpoint plus three bucket names is enough
for the whole app, and a purpose that needs its own account still overrides
everything. Audio backups and documents use it; generated images keeps its
own tested storage module, whose variable names the resolver already
understands.
Nothing existing has to change: S3_ACCESS_KEY_ID, S3_SECRET_ACCESS_KEY and
the AWS_* fallbacks still resolve, and path-style addressing keeps each
purpose's previous default — off for documents, so a Backblaze endpoint
behaves as before, on where a custom endpoint implies MinIO. A _FILE
credential now always beats an inline one, so a mounted secret cannot be
shadowed by an inherited environment variable.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
Recording
- A screen wake lock is held for as long as a recording runs. Browsers drop
the lock whenever the page is hidden, so it is taken again on return —
without that, one glance away ended it for the session. The lock is
reference counted (two recorders cannot release each other's), never
requested while hidden (the request would just be rejected), and a denial
or an unsupported browser leaves the recording running.
- Signing out releases it and stops the recording; nothing is sent, because
the session that owned the audio is gone.
- start() on an already-running recorder is now a no-op instead of replacing
the MediaRecorder and silently dropping everything captured so far.
- A recording that ends by itself — recorder error, or the microphone taken
by another app, unplugged or revoked — takes the same path as pressing
Stop, so it is transcribed and stored rather than left in a tab that still
says "recording". Moving around the workspace already kept recording.
Retention
- Every recording is kept for 24 hours now, not only the ones whose
transcription failed. /api/transcribe already has the audio, so this costs
no second upload, and a storage failure is logged rather than thrown: it
must never lose the transcription someone is waiting for.
- One store (src/utils/audioBackupStore.js) is shared by /api/transcribe and
/api/audio-backups so the two cannot drift. Payload goes to object storage
when AUDIO_BACKUPS_S3_* is set and to the encrypted Postgres column
otherwise; metadata always stays in Postgres, so listing, ownership and
expiry behave the same either way. Object keys are scoped by owner, and
the expiry sweep deletes the object with the row.
Verified against the live database: round trip byte-identical, another user
reads null, 950 -> 48 bytes compressed, expired rows take their objects.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
- adminMilestones was reached only because adminConfig is mounted on
/api/admin ahead of it and guards the whole path. adminMiddleware checks
req.user.role and nothing else, so it failed closed (403) rather than
open — but on mount order, not intent. It now states the requirement,
with a test covering all four admin routers.
- TODO.md records the whole audit: what was verified working (live
transcription round trip, voice mode wiring), what was fixed, the two
advisories that are unreachable and why, and the CI/CD and Kubernetes
work worth doing before scaling out.
Security review found nothing else exploitable: parameterised SQL
throughout (the one interpolated table name is allowlisted), CORS refuses
to start open in production, JWT_SECRET refuses to start unset in
production, rate limits on /api and each auth route, a real CSP, and no
secrets in the repo.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
Security
- nodemailer 9.0.1 -> 9.1.1, clearing four high advisories, two of which
are delivery bugs that matter for an app that sends mail: recipient-domain
validation bypass via RFC 5322 comments, and an IDN/punycode allow-list
bypass, both of which can route mail to an attacker-controlled domain.
Live transcription
- The Settings picker was a hardcoded list of six ids
(local-whisper-*, local-parakeet-v3, gemini-*). None of them resolve on
this gateway, and /api/transcribe prefers the user's choice over the admin
default, so picking one broke every recording with "Invalid model name".
Verified against the live gateway: local-whisper-large-v3-turbo -> 400.
- The picker now lists what /model/info advertises as audio_transcription,
cached for five minutes, with the built-in list kept only as a fallback
and the admin default marked.
- The pipeline itself is healthy: local-kokoro-tts produced 92KB of speech
and mistral-voxtral-mini-transcribe returned the sentence back verbatim.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
Feature audit
- Settings claimed you could reference a template by saying "use my normal
physical exam" in dictation. No phrase handling exists anywhere, and the
prompt says the opposite: "Never copy clinical content from a template —
only formatting and structure." So a template can never supply findings.
The text now says what happens, and keeps the true privacy statement that
only template categories go to the AI (Custom is filtered out in
/memories/context by AI_CONTEXT_CATEGORIES).
- Templates themselves are real: CRUD plus /memories/context, injected as
style hints by hpi, soap, sickVisit, wellVisit, edEncounters and
hospitalCourse, behind the `memories` feature flag.
Docs
- docs/CLINICAL_ASSISTANT.md listed six settings and offered `deepl`, which
no longer exists in the code. The table now covers all seventeen keys the
server reads, with their fallbacks, plus how a model reaches a user.
Testing
- Every test file's stdout is now pure TAP, which is the stream node:test
parses results from. Three sources: a leftover debug console.log dumping
600 characters of HTML, page modules logging into a JSDOM without a
virtual console, and the server startup banners. The banners are guarded
by NODE_TEST_CONTEXT, set only inside node:test children, so production
and `node server.js` output is unchanged (verified both ways).
- Three consecutive full-suite runs at 671/671.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Dv6sqaY6Vq3ChZHMem3cnU
Image models
- The Clinical Assistant "Image models" list waited on an old
#assistant-image-model dropdown that no longer exists, so discovery never
reached it and only four hard-coded fallbacks appeared — with no way to
add any of the gateway's 50 image models.
- Image Generation search rows now have + Add / Added. Added models are
saved as clinical_assistant.image_model_roster (validated server-side:
up to 100 ids) and appear in the Clinical Assistant list at once; ticking
one there offers it to users. Anything already allowed or configured
stays listed. Unsaved ticks survive an add.
- The roster notification is guarded, so it can never fail the settings load.
Phone top bar
- The page is drawn under the status bar (viewport-fit=cover) and its
theme colour was the removed header's blue, so on an iPhone content showed
scrolling at the top of the screen. The row is now a real fixed element
that extends behind the status bar (env(safe-area-inset-top), 0 in a
normal tab), the menu button, sources pill and drawers clear it, and the
theme colour is white.
Verified in Chromium: + Add -> saved roster -> listed unticked; tick kept;
remove works. Top bar is the only thing in the top 48px on all 22 pages;
phone menu positions unchanged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01W4rptBNvn6RYieQw54GXNS
Gallery tiles are 56px but were downloading the full ~280kB original. Previews
are now rendered with sharp and stored beside the originals in the same MinIO
bucket under a thumbs/ prefix, so nothing about credentials, lifecycle or backup
changes. Measured on live assets: 216-294kB originals become 13-19kB at 256px,
about 16x smaller; 640px is about 4x.
Both paths, as asked:
- Rendered when a job completes, so the first viewer never waits for a resize.
A preview failure never unmakes a finished job.
- Rendered on demand for anything that has none — the existing 26 images work
immediately with no backfill required, and the result is stored for next time.
Boundaries that matter more than the speed:
- Only 256 and 640 are honoured. An open width parameter would let a caller
drive arbitrary resizes.
- Permission is checked against the ORIGINAL before a preview is served, so a
preview can never widen who can see an image.
- Previews carry their own SHA-256 and owner headers, because the client
verifies both on every asset; sending the original's checksum would be
rejected as tampering, which is that check working correctly.
- Still private, no-store. The client asset pattern was widened to exactly
?w=256 and ?w=640 and nothing else.
Client-side downscaling stays as the fallback when a preview cannot be produced.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018e1PLqrKgAM9jQhFKRnbLd
Sources (correcting what I built earlier)
The previous toggle branched the SYSTEM PROMPT, so the same question could get a
different answer depending on a display setting — the bias this was meant to
avoid. The prompt is now unconditional: buildSystemPrompt takes no display
argument and is byte-identical either way. Hiding sources happens on the way out
— the server omits them and strips the now-orphaned [n] markers from the copy it
sends. The answer is generated, stored and exported with citations intact, so
turning the setting back on restores them without re-asking anything. Renamed to
clinical_assistant.show_sources; the old key is still honoured.
Signed-out preview (admin opt-in, default off)
A visitor may try the assistant; reaching for the workspace asks them to sign in.
Deliberately narrow:
- Reachable paths are an exact allow-list, not a pattern, so a new endpoint is
private unless someone adds it on purpose.
- A preview visitor gets no identity at all (id: null), so nothing can be owned,
saved, billed or addressed to them.
- The image tool is withheld rather than left to fail on a null owner, and no
audit rows are written.
- A caller presenting a token is authenticated normally, so preview can never
downgrade a real session; if the setting cannot be read, authentication is
required.
- Actions needing an account are hidden rather than offered and refused.
Composer
The bar above the transcript is gone. Patient take home, Export PDF, Download
transcript and Attach images moved into a + menu in the composer, and the model
selector moved beside send — shown only when there is more than one model, as
before. Both views now start at the same top edge, so switching modes cannot
nudge the page up or down. On an empty transcript the tiled ground runs behind
and below the composer, which floats on it above centre.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01GmpYHPSLGmXGZMyLpn2Lbe
Multimodal removal
The multimodal path called nc_multimodal_search against a second hardcoded
collection whose embedding service (multimodal-embeddings:7999) was never
deployed and ENABLE_MULTIMODAL_RAG has always been false, so it only ever logged
"multimodal search skipped". Removed rather than left as dead weight:
- clinicalRetrieval: normalizeMcpMultimodalResponse, isVisualSourceQuery,
isRadiologyQuery, buildMultimodalSearchQuery, classifyAndRerankMultimodalResults,
selectMultimodalResults, visualIntent, visualMetadataScore,
shouldRejectVisualSource, allowsFrontMatterQuery, looksLikeFrontMatterPage,
looksLikeTextOnlyPage and MULTIMODAL_CANDIDATE_LIMIT (~140 lines).
- clinicalMcpClient: multimodalSearch.
- The route's visual/text slot split is gone; the whole search limit is text.
- The "[visual PDF page match]" prompt label and the "visual PDF page" source
badge are gone with it.
Adding models
Model availability could only be ticked from what the gateway advertised, so an
admin could never offer a model discovery did not list. Each list now has a text
field: a typed id joins the same checkbox list, is enabled by default, is
de-duplicated, and persists through the normal allowed_models save.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BkfrkQwA4YGrGw9LZSpeAq
New setting clinical_assistant.citations_enabled (default true, admin checkbox).
With it off, retrieval, grounding and every other rule are unchanged — answers
are still built only from retrieved sources — but:
- buildSystemPrompt swaps only the citation block: the "cite factual claims with
[1]" rules are replaced with "do not include citations, source numbers or
bracketed markers", and the note that the sourcing requirement itself is
unchanged. Grounding, scope, table formatting and tone rules are byte-identical
between the two modes.
- The server strips any stray [n] the model emits anyway, from the stored answer
rather than only the view, so saved chats and exports match what was shown.
- No sources are sent to the client at all, and the status endpoint reports the
mode so the UI hides the Sources panel and gives its 330px column back to the
chat instead of showing an empty rail.
Validated as a boolean in adminConfig, like the feature.* keys.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BkfrkQwA4YGrGw9LZSpeAq
Mobile
body.assistant-workspace .assistant-layout is both later in the file and more
specific than the .assistant-layout rule inside @media (max-width:640px), so
phones were inheriting height: calc(100vh - 64px) — an offset for the app header
that workspace mode already hides — plus grid rules on a flex container. The
topbar then overlapped the content. Added a matching-specificity mobile override
restoring 100dvh and the flex layout.
Workspace menu in the rail
The saved-chats rail gains a collapsible Workspace section listing the app's own
tabs, so the assistant sidebar carries both its chats and the workspace
navigation. It is built from the real .tab-btn elements rather than a second copy
of the menu, so tabs added, renamed or hidden in index.html follow automatically
and admin-only tabs stay hidden.
Prompt
Rewritten as principles instead of an enumerated rulebook: "respond to the user's
latest message, not to an earlier one", "if it carries no question, ask what they
would like you to look up", "never repeat a previous answer", and for the tool
"use it when the latest message asks for a picture, or for a change to one you
just made, and not otherwise". No example words in any language remain — the
model reads the message as written. 1032 -> 788 characters.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BkfrkQwA4YGrGw9LZSpeAq
The previous fix put a hand-maintained multilingual word list in the route. That
does not generalise — the bug report itself was "Окей", and the next one would be
a language not on the list. The model already has the tool, the system prompt and
the whole conversation, so the policy belongs where it reads it.
Removed from src/routes/clinicalAssistant.js:
- GREETING_RE and its short-circuit (an ASCII keyword list that decided, before
the model ever saw the message, that a greeting deserved a canned reply).
- The IMAGE_NOUN / IMAGE_REPEAT vocabulary lists and the answer-repetition
backstop added earlier today.
Moved into the prompt and the tool definition:
- DEFAULT_BEHAVIOR now says that a greeting, acknowledgement or thanks in ANY
language gets a one-sentence request for a clinical question, and that a
previous turn's answer must never be repeated to justify a second image.
- The generate_image description says to call it ONLY when the user's latest
message asks for a picture or a change to one just made, and that an
acknowledgement of an existing image is not such a request.
Both are admin-editable (clinical_assistant.system_behavior), so this can now be
tuned without a deploy.
dispatchImageRequestFallback and IMAGE_REQUEST_PATTERN stay: that is the
compatibility path for a serving model that writes the image prompt as text
instead of calling the tool. It reads the USER's message only, never the model's
answer, so it cannot replay a previous turn — it was not the cause of this bug.
The DEFAULT_BEHAVIOR byte-hash lock in prompt-administration.test.js is updated
deliberately, which is what that guard is for.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BkfrkQwA4YGrGw9LZSpeAq
Saved chats
Grouped the way Open WebUI groups them: Pinned, Today, Yesterday, Previous 3
days, Previous 7 days, Previous 30 days, then calendar months (with the year
once it is not the current one). Chats with no usable timestamp land in Undated
rather than disappearing.
Sidebar toggle
The collapse button used <i class="fas fa-sidebar">, which is a Font Awesome PRO
icon; on the Free 6.5.0 build this app loads it rendered nothing, so the toggle
has been an invisible button since it shipped. It now uses fa-table-columns,
keeps aria-expanded/aria-controls in sync, flips its label between "Hide saved
chats" and "Show saved chats" so the collapsed state is escapable, and animates.
Acknowledgements
"Окей", "Nice", "Perfect", "gracias" and friends now get the existing
"What clinical question would you like me to look up?" reply instead of a
retrieval and a paid generation. "yes", "sure", "no" and "more" are deliberately
excluded: answers end by offering more detail, so those must still be answered.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BkfrkQwA4YGrGw9LZSpeAq
Translation formatting
LibreTranslate's text mode destroys markdown syntax. Verified against the live
container: table pipes come back as "←", the |---| delimiter row is translated
as prose ("Silencio."), and "**bold**" returns as "** bold**" which no longer
renders. Its html mode leaves tags — and bare [n] markers — completely intact.
Messages and the patient take home are now rendered to HTML, simplified (maths
and UI chrome flattened to text), and translated as HTML. Citation chips are
re-linked from the returned markers afterwards, which is the step the original
html path was missing. A text-mode fallback remains for builds that reject html.
Repeat image generation
Typing "Окей" or "Nice" after an image turn produced another image every time:
the model saw its own "I'll generate an educational image…" in the history and
repeated it. Recognising acknowledgements in every language is not possible, so
the rule is inverted — a short follow-up (<=3 words) that mentions nothing about
a picture does not get the image tool offered at all when the previous assistant
turn produced an image. Terse repeat requests ("again", "ещё", "another one")
still work. The worst case is that a terse question is answered in text.
In-chat images
Generated images render as a 320x240 thumbnail instead of filling the bubble,
and the image itself opens the full-resolution preview.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BkfrkQwA4YGrGw9LZSpeAq
Translation
- Stop scrubbing markdown before sending it to LibreTranslate. The scrub
deleted ordered-list numbering ("1. Give amoxicillin" -> "Give amoxicillin"),
flattened tables into ambiguous whitespace and ate underscores inside
identifiers. Raw markdown now goes to the translator unchanged.
- Render the translation through the same markdown pipeline as the original
bubble, with the message's own sources, so [n] markers come back as the usual
clickable .assistant-cite chips instead of escaped literal text. Headings,
lists and tables survive with them.
- When the translator drops citation markers, surface the affected sources in a
recovery block rather than letting the evidence disappear.
- Image cards are live nodes: they are now re-attached on every path out of a
translation (success, failure and Show original), so a failed translation no
longer silently removes a generating image from the message.
Patient take home
- Add a language selector to the take-home modal, reusing the existing
/translate endpoint and offering only what the local LibreTranslate reports.
- Copy, Export and Email carry what the caregiver is actually reading; the
original stays canonical behind "Original".
Conversation budget
- The admin field no longer prefills with the environment value, which turned
the next Save into an accidental override and made the documented "leave
empty to use the environment" path unreachable. The effective limit is shown
as a placeholder instead.
- Report source 'default' honestly instead of naming an unset env var.
- The load-failure notice now lands on the <p> instead of an <input>'s
textContent, where it rendered nothing.
- One validator for the budget everywhere: conversationLimit() replaces a
parseInt that accepted "120000abc".
Other
- /assistant is addressed by its URL, not by ped_last_tab, so "/" no longer
reopens the assistant; the URL follows tab changes and Back leaves it.
- Remove the dead DeepL path (it referenced an undefined DEEPL_BASES) and stop
offering admins a provider the server silently ignores.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01BkfrkQwA4YGrGw9LZSpeAq
