Removed every `as any` in src/, then restored only the ones tsc still
needs: 163 → 36. 112 were in lotus-terminal.css.ts (`'… !important' as
any` etc. — vanilla-extract's style types accept these strings as-is),
so its file-wide eslint-disable goes too. The casts were type-only, so
emitted code and generated CSS are unchanged.
eslint warnings 49 → 46; the ratchet is tightened to match.
tsc clean; 1202 unit tests pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
useCallControlState kept the previous control's state (the all-off
default before a call) for one render after the control changed, until
its effect caught up. The thumbbar, SMTC and tray hooks pushed that render
to the OS, so every join flashed "muted + deafened" first (recorded:
idle → active+muted+deafened → active). The state now remembers which
control it belongs to and reads the new control directly when they differ
(now: idle → active).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Verified first: the embed hides Element Call's footer, so its in-call
settings sheet (which has the output picker) is unreachable from Lotus;
the cinny call bar had no output control.
Now a speaker button next to Deafen (desktop bar only; hidden where
setSinkId is unavailable — Firefox, Safari, Android Chrome) opens a menu
of enumerateDevices() audio outputs with the current one checked; picking
one sends io.lotus.set_audio_output to the fork (≥ 0.25.0-lotus.11),
which selects it, and the choice is re-sent with the rest of the sticky
fork state after an EC remount.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Event handlers typed as React.MouseEvent, join errors as Error, the
service-worker session setter as strings, the UIA policy map and the
webkitAudioContext probe as narrow shapes. No behaviour change.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
One line in the existing toast style when a call you were in ends: the
duration from our own join clock, plus the fork's io.lotus.call_summary
readout (fork ≥ 0.25.0-lotus.10) when it arrives — "connection was
good", "3 reconnects", "connection was poor for 4 min". Nothing is
stored or sent; the summary is one postMessage at hangup. Without the
fork summary the toast still shows the duration.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Verified upstream first: the Copy chip (with a 'Copied' state) already exists
on every code block. Added the missing half: a Wrap chip that switches long
lines to pre-wrap (default stays no-wrap + horizontal scroll); the choice is
remembered for the session, not persisted. Markdown/HTML output, layout widths
and the syntax theme are untouched. Verified headless: scrollWidth 1016 →
612 px with wrapping on, 'Copied' state on copy.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
utils/linkSafety.ts compares the registrable domain the visible text claims
(when it looks like a URL/host) with the href's; a mismatch, or a punycode
(IDN) destination, renders the anchor as SuspiciousLink, whose click opens a
confirm — "It shows matrix.lotusguild.org but goes to evil.example." with the
full URL, Cancel / Open anyway (opens in a new tab with noopener). Honest links
are untouched: same registrable domain (youtube.com text over www.youtube.com,
bbc.co.uk over news.bbc.co.uk), plain-word text, mailto:, matrix.to and Lotus
permalinks, anchors with non-text children. Comparator unit-tested (incl. a
Cyrillic-а paypal homograph); verified headless that the phish and IDN
messages are flagged, the honest ones are not, the click shows the confirm and
does not navigate, Cancel keeps you put, Open anyway opens the real target.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Host half of the talking-while-muted feature. The fork (0.25.0-lotus.9)
reports speakingWhileMuted for the local participant in io.lotus.call_state —
a level tap on the muted track, so cinny captures no audio. useMutedTalkWarning
polls that flag while muted; after 1.5 s of it being true it shows one sticky
toast, "You're muted — click to unmute", whose click unmutes; it fires at most
once per mute session (re-arms after unmute → mute), is retired the moment the
mic comes back on by any means, and never runs in Push-to-Talk mode. Setting
'Warn me when I talk while muted' (default on) sits next to AFK Auto-Mute.
Verified headless with a tone mic: no toast at +1 s, toast at +4 s, click
unmutes and clears it, second mute re-arms, unmute via the button clears it,
PTT mode → nothing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Answering from the in-call banner is the only call-to-call switch path; it
started the new embed and let the atom dispose the old one, which just removed
the iframe — our m.call.member in the old room lingered ~17 s (until the
delayed leave expired), so everyone there still saw us in the call.
hangupAndWait moves out of LogoutDialog into plugins/call/hangup.ts and
handleAnswer now hangs up, waits for our membership to clear (bounded 4 s),
disposes the old embed itself (its HangupCall echo would otherwise land after
startCall and clear the NEW embed from the atom — seen in testing), then joins
the new call. Measured headless: old membership gone in 1 s, new call live
with both participants.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
CallControl.setSound()/applyScreenshareAudioMuted() no longer poke <audio>.muted
inside the EC frame, and the per-membership re-apply (useCallMemberSoundSync)
is gone: with the pin at 0.25.0-lotus.4 the fork applies deafen and the
screenshare-audio mute through its own renderer, including for late joiners.
Verified headless: deafen set before a second participant joins mutes their
track on arrival; deafen/undeafen leaves the screenshare-audio mute in place.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
When the Element Call frame's navigation fails (offline, blocked) it becomes a
cross-origin error page and `iframe.contentWindow.document` throws a
SecurityError. Every DOM-driven call hook (CallControl selectors, deafen DOM
fallback, useCallSpeakers/useRemoteAllMuted observers, CallEmbed theme
injection) read it unguarded, so the moment the 25 s load watchdog fired the
page logged a burst of uncaught SecurityErrors. Reproduced headless by aborting
the EC index request. One `getCallDocument()` helper now treats "can't read" as
"not loaded yet"; the watchdog overlay itself was already correct.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Deafen muted the mic (correct) but undeafen left you muted, so every
deafen cycle silently turned into a mute. Remember whether the mic was on
when deafening and turn it back on when undeafening (Discord semantics).
Verified in a real two-party LiveKit call on the local stack.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
matrix.to cannot target this deployment (Cinny adapter hard-codes
app.cinny.in; web-instance[] is Element-only, allowlisted), so add a
"Copy Lotus Link" next to every Copy Link (message menu, space header
menu, sidebar space tab) producing https://<this origin>/home/<room>/<event>
?viaServers=… via plugins/lotus-permalink.ts. Lotus links in messages are
rewritten to their matrix.to form inside the HTML parser so they render as
room/event mentions and navigate in place.
/home/<room> for a joined room that belongs to a space or Direct now
redirects to its own route (was a preview card with a View button; also
the form matrix.to → Cinny links use). ?via= is accepted as an alias of
?viaServers= (what the matrix.to Cinny adapter emits). A deep link
visited while logged out is now honoured after an OIDC login: the OIDC
callback reloads at the app root, which discarded the stored path — the
index loader consumes it via the shared takeAfterLoginPath().
Verified end-to-end with Playwright on a local Synapse: logged-out cold
link → login → lands on the event under the space route; menu copies the
expected link; a pasted Lotus link renders as a mention and jumps in
place; both space menus copy the space link.
Closes#130
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Shared links routinely carry ad/analytics identifiers (utm_*, fbclid, gclid,
YouTube si=, Amazon ref=/tag=, X s=/t=, TikTok _r/_t, …) that tie every
recipient's click back to the person who shared the link. New
src/app/utils/urlTracking.ts is a pure, local stripper: a global list +
utm_/pk_/matomo_ prefixes, plus host-scoped rules so e.g. `si` is only
removed on youtube/spotify. matrix.to and non-http(s) schemes are never
rewritten; unparseable input is returned unchanged; Amazon's `th`/`psc`
variant selectors are deliberately kept. 13 unit tests.
Wired at three points, all behind a new Settings → Privacy toggle
(`stripTrackingParams`, default on):
- paste: plain-text pastes are cleaned and re-inserted through Slate's own
insertData so multi-line pastes still split into paragraphs;
- send: RoomInput submit + schedule paths and MessageEditor saves clean both
`body` and `formatted_body` (the HTML variant unescapes `&` around each
URL and re-escapes it so the markup is untouched);
- render: linkify `formatHref`/`format` and explicit `<a href>` in
formatted_body are cleaned, so links sent from other clients are safe to
click too. LINKIFY_OPTS is spread into memoised per-timeline objects, so
the toggle is a module flag kept current by ClientNonUIFeatures.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
81a6d9c9 wired getRtcTransports() but its .well-known fallback read
mx.getClientWellKnown(), which the SDK only populates when the client
is started with clientWellKnownPollPeriod — cinny never sets it, so on
the live client the fallback returned [] and calls still failed with
MISSING_MATRIX_RTC_TRANSPORT after deploy. (The unit test faked a
populated cache and hid this.)
The driver now fetches the client .well-known itself, once, via
AutoDiscovery.getRawClientConfig(domain) — the same resolver Element
Call used before 0.22 — and only when the SDK has nothing cached. Test
added for the uncached path (asserts a single fetch); verified the
resolver returns the livekit focus for matrix.lotusguild.org.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Since the upstream v0.25.0 merge, Element Call in widget mode no longer
reads .well-known for its LiveKit transport; it asks the HOST via the
widget API (org.matrix.msc4515.get_rtc_transports, capability
org.matrix.msc4515.rtc_transports). cinny never granted the capability
nor implemented WidgetDriver.getRtcTransports(), so discovery returned
nothing and every join failed with "Call is not supported"
(MISSING_MATRIX_RTC_TRANSPORT) — the [LocalMembership] Multiple
Transport Errors line in the browser console.
- matrix-widget-api 1.17.0 -> 1.18.0 (adds MSC4515; also changes the
sendDelayedEvent driver signature, adapted — parent delay ids were
removed from the draft).
- Grant MSC4515RtcTransports in getCallCapabilities.
- CallWidgetDriver.getRtcTransports(): homeserver /rtc/transports
(MSC4143) first, then the .well-known org.matrix.msc4143.rtc_foci list
(what matrix.lotusguild.org advertises today). Unit-tested.
Server side needs no change: the well-known already carries the livekit
focus and the JWT service answers.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
- io.lotus.request_state: when the fork's lotus handlers (re)register
(an EC-side remount that doesn't unmount us) we re-send deafen, quality
and the focus pin, and the decoration pusher re-pushes its roster —
decorations and the pin no longer vanish for the rest of the call
(element-call#17).
- focus_participant carries the per-device media id from call_state
(speaking device preferred) so a multi-device user pins the right
device (element-call#30).
- injectAudio returns the fork's reply; when it refuses with
reason:"muted" the soundboard shows "Unmute your microphone…" instead
of playing the clip locally as if it went out (element-call#13).
All backwards compatible with the 0.25.0-lotus.1 bundle (unknown action
is acked; missing reply fields default to "played").
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
clearFocusParticipant() had no callers, so a spotlight pin was permanent.
CallControl now tracks focusedUserId, the member menu toggles, and
dispose() clears the pin. Removes _pipMode/setPipMode (never read).
Fixes#56Fixes#59
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
npm audit --omit=dev went from 11 advisories (6 high) to 0:
- pdfjs-dist 5.7.284 -> 6.3.289 (GHSA-hq66-cqwq-w95j, arbitrary JS on
opening a malicious PDF; the client has an inline PDF viewer). Only API
change for us: getDocument() no longer accepts a bare string.
- react-router-dom 7.15.1 -> 7.18.3 (open redirect, DoS, RSC advisories)
- sanitize-html 2.17.4 -> 2.17.7 (javascript: URI bypasses; not
exploitable in our config per the audit, bumped anyway)
- katex 0.16.11 -> 0.16.47 (\htmlData attribute validation)
- uuid under @giphy/js-util overridden to >= 11.1.1 (giphy imports v4,
which is unchanged)
- npm audit fix for transitive brace-expansion / js-yaml / nanoid / postcss
Verified: tsc clean, 1000 tests pass, production build succeeds and
dist/pdf.worker.min.js is still copied from the new package.
Fixes#24
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Test-coverage expansion (2-agent reviewed, both SHIP). The named candidates
(roomToUnread, markedUnread, serverAcl, plaintextCaches, recent*) were already
tested, so this targets genuinely-untested pure logic.
- dom.test.ts: getThumbnailDimensions (scaling math incl. just-over-cap
boundaries), tryDecodeURIComponent, syntaxErrorPosition, and the three
scroll-view geometry helpers (via duck-typed element mocks — no jsdom).
- emoji.test.ts: getHexcodeForEmoji (astral codepoints, 4-digit zero-pad,
FE0F/FE0E/200D stripping on and off, keycap sequences, degenerate inputs)
and the pre-load `undefined` contract for getShortcode(s)For.
Fix (found while writing the tests): syntaxErrorPosition required whitespace
AFTER the digits (`/position\s(\d+)\s/`), but real V8/Node JSON.parse errors
put the number at end-of-string ("... at position N"), so it returned
undefined for every real error and the three dev-tools JSON editors silently
pointed their cursor at position 0. Dropped the trailing `\s`; tests now assert
extraction at end-of-string.
Gates: tsc 0, eslint 0, prettier clean, 891 tests, build ok.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Mobile-audit batch 1. All fixes reuse cinny's own responsive primitives and are
mobile-gated so desktop is unchanged.
- Message tables: wrap <table> in an overflow-x container so a wide table scrolls
instead of overflowing the message column / page body.
- Composer toolbar: let the before|editable|after row and the toolbar wrap on
phones (@media <=750px) instead of squeezing the editable to zero and pushing
Send off-screen.
- In-call control bar: collapse to the compact/stacked layout on a mobile
viewport (ScreenSize.Mobile) too, not just when the bar's own container is
<500px — fixes the 500-750px band where the control row overflowed.
- URL-preview card: base width toRem(400) -> min(25rem, 92vw) so a single card
fits a narrow phone (still exactly 400px on desktop).
- Explore card grid: drop to one column at <=750px (was a fixed 3-col grid).
Two review passes: desktop behavior provably unchanged (all gated by @media /
ScreenSize.Mobile; the table wrapper only contains previously-overflowing tables).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
COR-3 (CallEmbedProvider): the incoming-call lifetime guard distrusted a
caller's sender_ts only when it was >20s AHEAD of the server ts. A caller clock
that ran SLOW left sender_ts in the past, so the ring auto-dismissed/never
showed for a fresh invite. Trust sender_ts only within ±20s of the server ts,
else fall back to it (also fixes a NaN path when sender_ts is missing).
COR-6 (CallControl): forceState rebuilt CallControlState with 5 args, silently
defaulting screenshareAudioMuted to false; pass this.screenshareAudioMuted.
COR-5 (uploadContent + useBindUploadAtom): cancelling during the retry back-off
was a no-op (mx.cancelUpload only aborts an in-flight request), so the upload
resurrected on the next attempt. Thread an AbortSignal: the back-off sleep
resolves early on abort and the loop stops with an abort error; the hook aborts
a per-upload AbortController on cancel (alongside mx.cancelUpload for the
in-flight case).
All verified by two review passes (no double-settle / no resurrection); includes
their suggested abort-listener cleanup on normal sleep resolution.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add centralized typed helpers in src/app/utils/accountData.ts:
- getAccountData<T>(mx, eventType): T | undefined (returns content)
- setAccountData<T>(mx, eventType, content): Promise<void>
These wrap the single `as any` cast needed because matrix-js-sdk's typed
overloads reject the fork's custom account-data event names. Every call site
now stays fully typed on its content shape.
Route all account-data reads/writes that previously used
`(mx as any).getAccountData/setAccountData` or `mx.getAccountData(... as any)`
through the helpers (or, where a MatrixEvent is needed, through the existing
utils/room.ts getAccountData whose param is widened to accept string keys).
No behavior change: same event types, same content shapes.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- C-H1: forceState only on FIRST join; on EC reconnect re-arm the fork handlers
(resendForkState — deafen+quality only) instead of clobbering live mic/video/
deafen back to the join-time snapshot.
- C-H2: AFK auto-mute reads the fork's io.lotus.call_state VAD of the LOCAL
published track instead of getUserMedia on the browser DEFAULT mic (which could
measure silence while the user spoke on another device → auto-mute an active
speaker). Fails safe (never mutes) when call_state is null OR empty.
- C-H3: control observer re-binds after EC re-renders (body subtree:true + 100ms
debounce) with an early-return so unchanged state doesn't re-render.
- C-M3 setQuality join-gated; C-M4 hangup 4s fallback dispose (idempotent);
C-M5 PTT no longer silently un-deafens; C-M6 screenshare-audio mute resets on
stop; C-L4 deafen key works in the iframe; C-L6 setState-after-unmount guards.
Reviewed (C-H2 [] fail-safe + C-H3 re-render guard applied). tsc/eslint/prettier
clean, build OK, 677 tests.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- index.tsx: request navigator.storage.persist() for logged-in sessions so the
browser can't evict the IndexedDB rust-crypto store (eviction while the
localStorage session survives resurrects the device with a blank store → the
KE-1 "one time key already exists" upload storm). Guarded, checks persisted()
first, best-effort.
- Docs: remove HANDOFF_ELEMENT_CALL_FORK.md, LOTUS_E2EE_INVESTIGATION.md, and
LOTUS_BUGS.md. Port their live content into the three kept docs — verification
backlog → LOTUS_TESTING; open bugs + E2EE (KE-1..4) + an Element Call fork
operational reference (publish steps + io.lotus action catalog) → LOTUS_TODO.
Fix all dangling references (README, code comments, cross-doc links). Full
history of the removed docs remains in git.
Gates: tsc/eslint/prettier clean, build OK, 665 tests.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
CallControl now sends the new io.lotus.set_deafen action (join-gated via
forceState) on every deafen / screenshare-audio-mute toggle + on join, ALONGSIDE
the retained iframe-DOM .muted hack (transitional). Against the current pinned
bundle the action is immediately error-replied + swallowed by .catch — inert, no
timeout. Reordered toggleSound() to commit state before setSound() so the sent
deafen value isn't inverted.
Phase 2 (after the fork is published): bump the pin lotus.1 -> lotus.2 and delete
the DOM hack. Docs: HANDOFF §12.4, LOTUS_TODO P6-2, LOTUS_BUGS.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Each message is role="article"; collapsed messages (consecutive from one
sender) now carry an aria-label with sender + time — previously a screen
reader heard only the body with no attribution (the biggest a11y gap).
Pure messageAriaLabel() reuses the existing time utils (+3 tests).
- Editing a message announces "Editing message from <sender>" (ariaLabel
threaded MessageEditor → CustomEditor; the main composer is unaffected).
- System emoji get role="img" + aria-label from the shortcode; custom
emoticons always have an accessible name.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
HEAD-checks the copied denoise worklet/wasm/model assets for the selected model
and console.warns a single line listing anything missing — a silent asset skew
between the EC fork's expectations and vite's copied files would otherwise
disable noise suppression with no signal. Fire-and-forget; never blocks call
setup.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- emojibase (~965 KB) is now fully lazy: plugins/emoji.ts loads compact data +
shortcode maps via a memoized dynamic import (rejections reset the memo so a
mid-deploy chunk 404 can retry); reaction labels degrade to the raw glyph
until loaded. Consumers get FRESH array references on load (the module arrays
populate in place — same-ref state updates would skip re-render and leave
emoji search empty; reviewer-caught). Verified out of the eager graph.
- Service worker precaches hashed assets (workbox precacheAndRoute, 82 entries
~10.8 MB incl. the crypto wasm): repeat visits stop re-downloading the app.
index.html is NOT precached — navigations stay network-first so deploys are
picked up immediately; the media-auth fetch handler is untouched.
- ReactPrism: curated 21-language set — chunk 574 KB → 71 KB.
- Timeline inline images get loading="lazy".
- Removed dead dompurify (+types); sanitize-html is the real sanitizer.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Soundboard v2 — a near-parallel of the custom-emoji image-pack system for
in-call audio clips.
- Data model: 3-tier packs mirroring MSC2545 — room/space pack (state event
io.lotus.soundboard, inherited by child rooms via parent-space aggregation),
global refs (io.lotus.soundboard_rooms), and the personal pack
(io.lotus.soundboard account data; the v1 flat-list content is migrated to the
pack shape on read). New plugins/soundboard/ (readers, SoundboardPack, utils) +
hooks/useSoundboardPacks (useRelevantSoundboardPacks = user U global U room,
deduped). Unit-tested (migration + slug).
- Management: reusable SoundboardPackEditor (name + emoji + per-clip volume +
delete + upload + batched save), power-level-gated for room packs like emoji
packs; a Soundboard page wired into Room + Space settings.
- In-call: CallSoundboard rewritten as a Discord-style grid grouped by pack
(emoji + name tiles), sourcing room+parent-space U personal clips; a Manage
toggle embeds the editors; per-clip volume x master volume on playback.
- Spam guard: host gates on a playing key (fork enforces one clip at a time).
- Control bar: Mute-Screenshare moved next to the Screenshare button.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Renders LaTeX via spec data-mx-maths spans/divs (KaTeX render of the attr,
children as fallback) and conservative $…$ / $$…$$ text detection (escape-aware,
currency-guarded, never inside code/pre). KaTeX + CSS load lazily on first math
(ReactPrism pattern) — verified absent from the eager bundle. Sanitizer
unchanged by design (we render post-sanitize from attr/text; no incoming MathML
accepted). +14 unit tests.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- CallEmbed sets `autoGainControl=false` for the ML noise-suppression tier so
the browser's auto gain control doesn't fight the in-source ML model; the
browser/off tiers keep AGC on.
- Docs: refresh the LOTUS_FEATURES noise-suppression section (browser-native
default, quality-ordered dropdown, DFN3 ML default, attenuation floor,
gate-after-ML, DFN level 60, AGC-off, the reliability fixes) and LOTUS_TODO
P5-30 (mark tuning/reliability/AGC done; record GTCRN as researched-and-deferred).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Model dropdown is now ordered by quality/CPU, best first (DeepFilterNet 3 →
DTLN → RNNoise → Speex); fix RNNoise's inaccurate "High" voice-quality label.
- When a user opts into the ML tier, default to the highest-quality model
(DeepFilterNet 3). The tier default stays browser-native (known-good, best
perceived in testing so far).
- Wire the "Series Suppression" (native-NS-before-ML) toggle into the real call
path — it was applied only in the settings tester, so the tester could sound
better than the actual call. Default it OFF (a single NS stage is best
practice; it's an opt-in test aid).
- isMLDenoiseSupported now also requires WebAssembly, so ML isn't offered on
strict-CSP shells where it would silently fall back to the raw mic.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Element Call is now consumed as our self-built fork
(@lotusguild/element-call-embedded); wire up its previously-dormant
capabilities and document the fork as live.
Soundboard (P5-15): a call-bar button plays user-uploaded audio clips into the
call as a real published track (io.lotus.inject_audio) plus local playback.
Clips are uploadable like emoji/sticker packs, stored in io.lotus.soundboard
account data (synced across devices). Gated by a Settings toggle + volume.
Quality controls (P5-31): per-user mic/screenshare bitrate + screenshare
framerate (Settings -> Calls), applied via io.lotus.set_quality clamped to any
room cap. Room admins set caps and hard call-permissions (allow_screenshare /
allow_camera) in Room Settings -> Voice; the call bar hides blocked buttons.
- New: CallSoundboard, useSoundboard, soundboardClips; RoomQuality,
useCallQuality, callQuality (+ unit tests).
- Optimistic-write RoomQuality admin UI (no stale-state clobber).
- Docs: mark EC fork live across README/FEATURES/TODO/BUGS/TESTING; add D2
manual-test steps.
Numeric quality caps are client-cooperative; screenshare/camera permissions are
hard-enforced server-side (see LotusGuild/matrix voice-limit-guard).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Coverage work found a 3rd real bug: isMacOS() compared os.name against the
legacy 'Mac OS' string, but ua-parser-js v2 reports 'macOS' — so it was dead,
and Mac users saw "Ctrl + k" instead of "⌘ + k" in the editor toolbar, search,
and settings shortcut hints. Now accepts both 'macOS' and 'Mac OS'.
Suites (via subagent, verified): via-servers (10 — power/popularity server
selection), bad-words (9), syntaxHighlight tokenize (14), plugins/utils
getEmoticonSearchStr (5), imageCompression formatFileSize/isCompressible (5),
user-agent (6, now asserting the fixed behavior).
Full suite now 501 tests, all passing.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Resolve the eslint/prettier failures from the previous commit (non-blocking
in CI, but real): drop the banned `void` operator on fire-and-forget
transport.send().catch() calls, prefix the now-unused _denoiseNativeNS
param, and run prettier on the touched files.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Switch to @lotusguild/element-call-embedded@0.20.1-lotus.1 (our self-built
fork) and turn on the source-level features it adds:
- #1 denoise CUTOVER: in-source ML denoise (lotusDenoiseSource=1) replaces
the build-time getUserMedia shim — removed the shim injection from
vite.config.js (denoise/ assets still shipped; the processor loads them).
Survives reconnects (fixes A7).
- #2 call-state: CallEmbed consumes io.lotus.call_state; useCallSpeakers /
useRemoteAllMuted prefer it over scraping EC's DOM (DOM fallback kept;
empty payloads ignored).
- #4 focus: CallControl.focusCameraParticipant sends io.lotus.focus_participant
(works during screenshare), replacing the DOM tile-click hack.
- #5 theming: lotusTransparent=1 (native transparent background).
- #6 decorations: LotusDecorationPusher sends each member's decoration URL
via io.lotus.decorations -> rendered on in-call tiles.
#3 soundboard / #7 quality ship dormant (EC-ready; no host UI sends them yet).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- N122: setMediaState resolves on EC's transport ACK instead of waiting for a
DeviceMute state-echo that EC may elide or skip during teardown — which
previously stranded the promise forever and silently skipped the initial
deafen state + first StateUpdate on join. Dropped the single-slot
mediaStatePromiseResolver; onMediaState remains the authoritative sync path.
- N123: focusCameraParticipant now waits for a spotlight videoTile to mount via
a MutationObserver (with a 600ms hard-timeout fallback) instead of a fixed
2-frame delay that EC's React commit can exceed on slower devices.
- N126: PiP position restored from localStorage is shape+finiteness validated,
so corrupt data can't feed NaN into the position math (invalid 'NaNpx' CSS).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- CallEmbed watchdog now SELF-HEALS: a genuine ready/joined signal arriving
after the 25s timeout clears the error and notifies subscribers with
undefined, so a slow-but-successful EC load no longer strands the user on
the recovery screen over a live call. Listener dispatch wrapped in try/catch.
- ringtones: synth notes route through a per-session master gain; stop() ramps
it to 0 so the ring is silenced instantly on answer instead of letting the
last scheduled phrase ring out over call audio.
- IncomingCallBanner: ping fires exactly once per incoming call (guarded by
refEventId) instead of re-pinging when ringtone settings change mid-banner.
- focusCameraParticipant: try multiple tile selectors (EC labels vary by
version), defer the tile click past EC's async spotlight layout switch
(rAF x2), and dev-warn when no tile matches so testers get signal.
- uploadContent: a cancelled upload (mx.cancelUpload -> AbortError) is no
longer treated as retryable — previously the retry loop could resurrect an
upload the user just cancelled. Also retry on 408.
- addRoomIdToMDirect/removeRoomIdFromMDirect: guard against a corrupt m.direct
whose values aren't arrays.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- CallEmbed: 25s load watchdog that fails fast on iframe error / preparing-error /
timeout instead of hanging on a permanent spinner; additive onLoadError API,
cleared on ready/capabilities/joined.
- CallView: user-visible "call failed to load" overlay with Retry/Leave (folds +
tokens) via a new useCallLoadError hook.
- CallMemberCard: wrap the participant avatar in AvatarDecoration so decorations
render in the call roster (the tile rendered UserAvatar bare while member lists
already wrapped it).
Addresses LOTUS_BUGS item 3 (avatar decorations in calls) and EC iframe failure monitoring.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Pure formatting reflows (multi-line wrapping of long lines/imports/tables);
no behavior change. Clears the working tree of pending prettier diffs.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Four changes to match screenshare full-screen UX for camera feeds:
1. Fullscreen button always visible
CallControls.tsx: remove `screenshare &&` gate — the ⛶ fullscreen
button now appears in camera-only calls, not just during screenshare.
2. Per-participant camera focus (CallControl.focusCameraParticipant)
Finds the target's video tile in the EC iframe DOM via:
[data-testid="videoTile"] / [data-video-fit]
closest ancestor of [aria-label="${userId}"]
Enables spotlight mode if not already active, then clicks the tile
so EC's internal focus handler runs. Falls back gracefully if the
tile is not in the DOM (camera off).
3. MemberGlance participant popup
Clicking a participant avatar in the call status bar now shows a
small menu: "Focus camera" (calls focusCameraParticipant) and
"View profile" (existing behaviour). Previously it opened the
profile immediately with no way to focus the camera.
4. PiP fullscreen button
A ⛶/⊡ icon button appears in the PiP overlay top-right area,
letting users go fullscreen directly from PiP mode without
navigating back to the call room first.
UNTESTED — requires a real multi-participant call to verify tile
clicking behaviour and fullscreen transitions.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace raw error object logging (which may contain Matrix event
payloads, user IDs, or message bodies) with e.message-only strings
in three files:
- CallEmbed.ts: state update and event widget feed errors
- msgContent.ts: image/video element load failures and thumb errors
- RoomInput.tsx: GIF send failure
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Implement a flexible, multi-model noise suppression pipeline for Element Call/LiveKit integration:
- ML Engines: Added support for RNNoise, Speex, DTLN, and DeepFilterNet 3 models.
- Pipeline Architecture: Implemented modular audio processing in lotus-denoise.js, supporting 'Series Suppression' (running browser-native NSNet2 before ML) and a hardware-style Noise Gate.
- UI & UX Enhancements:
- Settings UI: Added model comparison chart with CPU/Quality metadata.
- Tuning: Added Live Microphone Meter for calibrating Noise Gate thresholds.
- Reporting: Added LotusToast system to alert users when ML suppression fails or falls back to raw input.
- Robustness & Quality:
- Capture Fidelity: Removed forced 48kHz capture constraints to allow native-rate capture (solving static issues with high-end audio interfaces).
- Performance: Added WASM SIMD detection with transparent fallback.
- Capability Detection: Added browser feature detection to disable unsupported ML modes.
- Build Integration: Updated Vite config to self-host all model WASM/tflite assets in /denoise/ directory.
Replace the boolean call noise-suppression setting with a 3-way control
(Off / Browser-native / ML beta) in Settings -> General -> Calls.
- Off: noiseSuppression=false to Element Call
- Browser-native: EC's built-in WebRTC suppressor (prior default)
- ML (beta): on-device RNNoise (@sapphi-red/web-noise-suppressor)
Element Call captures the mic inside its iframe and publishes to LiveKit,
so the host can't reach that track; LiveKit's Krisp filter is Cloud-only
(we self-host the SFU) and EC's own RNNoise PR #3892 is unmerged. The ML
tier instead injects a same-origin pre-init shim into the vendored EC
index.html (build/lotus-denoise.js, wired by the lotusDenoise vite plugin)
that patches getUserMedia and routes the captured mic through an RNNoise
AudioWorklet before LiveKit sees it -- the same post-capture pipeline as
#3892, with no EC fork/AGPL/rebase burden. Falls back to the raw mic if
setup fails; keeps echoCancellation/AGC on the raw capture.
- settings.ts: callNoiseSuppression -> 'off'|'browser'|'ml' + legacy
boolean migration (true->browser, false->off)
- CallEmbed/useCallEmbed: tier maps to noiseSuppression param and appends
lotusDenoise=ml (native suppressor off in ML mode)
- vite.config.js: copy RNNoise worklet/wasm + shim into the EC bundle and
inject the shim <script> before EC's module entry
- docs: LOTUS_FEATURES.md, LOTUS_TODO.md (P5-30 done)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Lotus CIandClaude Opus 5.5
jaredandClaude Opus 5
KrishanandAjay Bura