`sudo pacman -U <url>` also fetches `<url>.sig` and failed (404) on
CachyOS: remote packages fall under RemoteFileSigLevel (signature
required) and we don't sign the package. A downloaded file installs under
LocalFileSigLevel (optional): `curl -LO <url> && sudo pacman -U ./…`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Reported on CachyOS: "Check for Updates → The update downloaded but
couldn't be installed … Permission denied (os error 13) at path
/usr/bin/tauri_current_app…". The app was installed from the Arch package;
Tauri's Linux updater can only replace an AppImage.
With cinny-desktop's new update_install_kind command:
- pacman / deb installs: the toast says the update is available and opens
Settings → General → App Updates, which shows the package-manager command
(`sudo pacman -U …pkg.tar.zst`, or the .deb + `sudo apt install`) with
Copy command and Download package — no Install & Restart that can't work.
"Copied" only when the clipboard write actually succeeded.
- other distros: a link to the downloads page.
- Windows, AppImage, and desktop builds without the command: unchanged
in-app update.
- a native "package-managed" refusal shows the same help.
Tests: unit (kinds, commands, refusal detection); in the real client with a
simulated desktop bridge: pacman → toast + Settings command/buttons,
install never attempted; older desktop → in-app flow as before. Unit 1321,
Playwright 26 passed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
When the homeserver, voice calls or sign-in break, or maintenance is under
way, say so in the app — driven by the Kuma status page
(isitup.lotusguild.org/status/matrix), managed from Kuma's UI. The client
asks Kuma directly (not via our servers) so it still hears "the server is
down" when our servers can't tell it.
- config.json `statusPages`, keyed by homeserver: users of other servers
never contact Kuma.
- utils/kumaStatus.ts (pure, unit-tested): parse Kuma 2.x's public JSON;
a group is down when any monitor fails two checks in a row (down+down or
pending+down); maintenance = windows under way; announcements = incidents.
One strip at a time: server down (connection lost AND Kuma confirms) >
server having problems > maintenance > calls down > announcement;
sign-in problems on the login screen only.
- Wording about the user's own connection: "Connection lost … our status
checks say the server is up, so it may be your internet connection" ONLY
when Kuma checked the server after this client's connection dropped and
it passed; a stale "up" (Kuma needs a minute or two to notice an outage)
keeps the plain "Connection Lost!".
- useServerStatus: polls only while visible; 5 min, 60 s while something is
wrong or the connection is lost, at once when it drops; backoff; any
failure = no banner (Kuma being unreachable never looks like Matrix
being down); GET only, no cookies.
- UI in the existing banner slot and style (ContainerColor/Line like the
sync and clock banners); Details expands; dismiss for calls-down and
announcements (an edited announcement comes back); calls-down note above
Join; phone: one line + Details.
Needs the CSP connect-src to allow https://isitup.lotusguild.org (matrix
repo) before it can fetch in production; until then it fails quiet.
Tests: 15 unit tests (live page layout, two-check rule, any-monitor rule,
unknown/garbage, UTC beat times, stale-vs-fresh "up", priorities, login vs
client, maintenance, announcements + dismiss/edit); e2e (fixtures for Kuma):
calls-down strip + dismiss across reload, other homeservers make no
requests, Kuma 500 → nothing, lost connection + Kuma down → critical strip
instead of "Connection Lost", + fresh "up" → "may be your connection",
+ stale "up" → plain "Connection Lost". Unit 1315, Playwright 26 passed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Incident 2026-09-29: the homeserver's host ran out of memory and stalled for
~2 minutes. The /sync that finally went out carried events whose `age` was
computed ~30 s before it arrived, so every client showed "Your computer's
clock is 30 seconds ahead of the server" while the real problem was the
server (all host clocks were within 0.25 s the whole evening).
The skew estimate was the median of the last 5 samples, and a sample is
local skew + delivery delay, so one late /sync with a handful of events
tripped it.
- Estimate = the LOWEST sample of the last 5 minutes: delay only ever adds,
so the fastest-delivered event is the truest.
- "Behind" (which a delay can't cause) is reported as soon as there are 3
samples, like before. "Ahead" must hold across samples received at least
a minute apart, so a single late burst never trips it.
- Samples are aged on the monotonic clock, and a change of the local clock
(someone fixing it) resets the measurement, so the warning clears at once.
- Only events stamped by our own homeserver are sampled: a federated event's
origin_server_ts is the other server's clock.
- Wording: "This device's clock is … Voice calls and encrypted messages can
fail until it's corrected." / call bar "Device clock … : calls may fail"
(was "will fail").
Unit tests: the incident (late burst after normal traffic, and a fresh
client whose first samples are all late), mixed slow/fast deliveries,
ahead only after a minute, behind at once, hysteresis, clock fixed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Until now a send that failed (offline, homeserver down, a blip) went
straight to "Failed to send": nothing retried it, and a reload dropped it
without trace (chronological pending ordering keeps local echoes in memory
only).
- Outbox (utils/outbox.ts + features/outbox/OutboxFeature): own message
sends (text, stickers, reactions, polls; not call signalling or
redactions) are mirrored to localStorage from their first local echo until
the server confirms them or the user cancels.
- After a reload they come back as local echoes via room.addPendingEvent,
same shape as the SDK's own. Recent ones (< 1 h) are sent again with the
same txnId; older ones come back as "Failed to send" for the user to
retry or cancel. Ones the server already has (transaction id seen in
/sync) are dropped, so no duplicates.
- Retries: network failures (ConnectionError, 408/429/5xx) are re-sent when
the connection returns (sync recovers or the browser goes back online),
and after a blip while online (5 s, backing off, max 10 per message).
Oldest first, in order per room. 4xx / consent / encryption failures are
left to the user.
- UI: a network failure while offline shows a clock, "Queued. Will send
when you're back online" (thread view too), not the red ✕. The ✕ is now
a button: click to retry.
- Logout wipes the outbox with the other plaintext caches (the content is
decrypted, like drafts).
Tested end to end against a local Synapse (Chromium): offline → queued →
sent once on reconnect; homeserver unreachable → queued → sent once; failed
send → reload → sent once and shown once; server accepted but response lost
→ reload → no duplicate; 2 h old entry → failed, not sent, click ✕ → sent;
cancel → gone after reload; encrypted room → restored message goes out as
m.room.encrypted with no plaintext and decrypts; one-off failure retried by
itself in ~5 s; no page errors. Unit tests for the pure parts; Playwright
20 passed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
The Linux desktop app runs on WebKitGTK, which ships without WebRTC (2.52
has no RTCPeerConnection; 2.54 disables it outright pending a libwebrtc
backend around 2.56), so calls can't work there. Until now the call button
just disappeared, the call room said "Your browser does not support WebRTC"
with Join disabled, and an incoming call couldn't be answered.
In the desktop app (isTauri) without WebRTC:
- call rooms: "Calls aren't available in the desktop app on Linux yet: its
web engine has no WebRTC" + an "Open in browser" button;
- incoming-call overlay: the same, with "Answer in browser";
- room header: the call button stays, and opens the room in the browser.
The link is the room in the web app (config.json `webAppUrl`, https only,
new key); the user presses Join there. Deliberately not an auto-join link:
a crafted URL must not be able to join a call and open someone's mic. It
opens through the desktop's new-window handler (web/mail schemes only → the
system browser). Without `webAppUrl` the explanation shows with no button;
browsers without WebRTC keep the old message.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
markAsRead ran on every room visit (timeline at the bottom and focused) and
sent a threaded receipt for every unread thread, so a reply in a thread you
started or replied in lost its unread badge the moment you glanced at the
room, without opening the thread.
Reads from just viewing the timeline are now "passive":
- threads you follow (started, replied in, or were mentioned in) stay unread
until their panel is opened; other threads are still cleared so they don't
keep the room dot lit forever;
- while a followed thread has an unread reply, the main receipt is scoped to
the main timeline instead of unthreaded, because an unthreaded receipt
also reads every older thread reply (the next main message would clear the
thread anyway). The check also asks whether the latest reply is read, since
the thread's count lags when the reply and a main message share a sync;
- the thread open in the panel is skipped, as the panel sends its own
receipt (was two identical receipts per reply).
Explicit "mark as read" (room menu, Escape, bulk actions) still clears
everything. Unit tests for each rule plus a local-homeserver e2e that checks
the server's per-thread count survives a reply + newer main message.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Thread images opened the room media lightbox, which looks the event up in
the room's detached media timeline. Thread replies never reach that
timeline, so a thread image always showed "1 / 1" with no prev/next, after
paging the room's media up to six times. The thread panel now builds the
viewer's items from its own root + loaded replies, and "Go to message"
scrolls the thread panel instead of the room.
The viewer gains a "Copy image" button: fetches the displayed media (blob
URL for E2EE, authenticated URL otherwise), re-encodes to PNG when needed,
and writes it via ClipboardItem with a promise so Safari keeps the click's
user activation. Hidden where ClipboardItem is missing. No "open in new
tab": an E2EE blob URL is revoked when the viewer closes and authenticated
media 401s in a bare tab.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Probed how real providers fail before picking signals:
- X renders an EMPTY frame for a deleted/private/suspended post and says
so only via postMessage `twttr.private.no_results`. The post embed now
swaps to "This post isn't available…" with an "Open on X" link.
- A hung frame never fires `load`. After 20 s every player (media, rich
posts, TikTok, Steam widget, X) overlays "This embed is taking too long
to load" with Retry (remounts the iframe) and "Open on <site>". A late
`load` clears it.
- Instagram and Bluesky show their own "removed / not found" page, and a
refused request still fires `load` (browser error page), so neither needs
or can use a guess. A missing height message is not treated as failure.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
When a multi-line paste looks like code, a chip above the composer asks
"That looks like code. Format it as a code block (js)?" with Format as code /
No thanks. It never converts on its own; typing, No, or 8 s dismiss it, and
Settings → Editor → "Offer to Format Pasted Code" turns it off.
- utils/looksLikeCode.ts (pure, tested): ≥ 3 lines and ≥ 2 of indentation
with depth changes, statement terminators, operators + brackets, keywords,
monospace/<pre> clipboard HTML, or SQL clause lines. Prose guards: quoted
replies, URL lists, long sentence-punctuated lines, plain-word lines,
markdown lists. Language guess only when fairly sure (js/ts/python/sql/
rust/c/php).
- Accept rebuilds the pasted paragraphs as one code block from their plain
text (code lines hold text only) and leaves the caret after it.
- Code blocks carry an optional lang → <code class="language-js"> (whitelisted
identifier only).
Verified in Chromium: pasting a JS function shows the chip; Format as code →
sent formatted_body is <pre><code class="language-js"> with indentation intact;
a three-line prose paste shows no chip; typing after a paste dismisses it.
Unit tests: 14 detector fixtures + 2 output tests; chromium e2e 19 passed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
- sendRoomMessage (composer's fire-and-forget sends: text, location, voice,
files, GIFs) and sendRoomEvent (polls, poll responses/ends, forwards,
reactions, edits) in utils/room.ts carry the one cast each needs
(`keyof TimelineEvents` / `RoomMessageEventContent`, no `any`).
- sendRoomMessage also swallows the rejected promise: a failed send already
shows on the local echo (Failed to send + Retry, or the consent prompt), so
it no longer surfaces as an unhandled error in the console.
- getAccountData narrows to `keyof AccountDataEvents`; ForwardMessageDialog's
guard now narrows `contentToSend` itself (same behaviour).
- `as any` 39 → 23; eslint warnings 46 → 36, ratchet tightened to 36.
Verified in Chromium on a local Synapse: a text message, a quick reaction and
an edit all reach the server with the right content; a consent-blocked send
no longer logs an unhandled MatrixError. 1219 unit tests pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
A homeserver with a consent requirement (Synapse user_consent) rejects sends
with 403 M_CONSENT_NOT_GIVEN until the user accepts its current terms. The
message just showed "Failed to send" with no reason.
Listen for the SDK's HttpApiEvent.NoConsent and show a dialog naming the
user's own homeserver (the client works with any server, so no Lotus-specific
wording), with "Review and accept" opening the server's consent_uri (http(s)
only; anything else is dropped) and "I've accepted — retry sending" resending
every event that failed for this reason. "Later" snoozes it for 10 s so
background retries don't re-open it immediately.
Verified in Chromium against a local Synapse with the send endpoint answering
M_CONSENT_NOT_GIVEN: dialog shows, link opens, retry delivers the message.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Status messages are saved per device and re-sent with every presence
heartbeat (a presence write without status_msg clears it on Synapse). A
device never receives its own user's presence changes made on other
devices, so the DP3 fix in db864326 — mirroring remote changes from the
Profile page — could never fire: device B kept a status that device A had
cleared and re-published it on its next state change.
Heartbeats now reconcile with the server first: GET our own presence,
send the server's current status_msg and bring the local copy in line.
Falls back to the local copy when the read fails, when the server shows
us offline (invisible mode clears the status by design), and for 15 s
after this device saved/cleared its own status (a server read that
hasn't caught up yet can't override a fresh save).
Verified with two sessions of the same user against local Synapse:
B sets "dp3 old status" → A clears it → B goes hidden→visible → server
stays "" and B's local copy is removed (before: back to "dp3 old status").
A sets "dp3 new from A" → B heartbeat keeps it and adopts it locally.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Steam double-encodes its meta tags (`&quot;`), so after Synapse's
single decode the card showed "The "Perpetual Testing
Initiative" …". The preview's og:title / og:description /
og:site_name are now decoded once where the preview is fetched, so every
card (about 20 read those fields directly) gets clean text. Rendered as
React text only, so decoding can't inject markup; exactly one level is
decoded.
Verified on the Portal 2 store link: "The \"Perpetual Testing
Initiative\" has been expanded…", no literal " left on the page.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
The comment said on.soundcloud.com links need an oEmbed round-trip. They
don't: Synapse follows the redirect for the preview and the og:url
fallback re-parses the canonical track URL. Checked with a real short
link (the play facade loads the w.soundcloud player).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
A friend's update failed ten times in a row ("Update check failed: error
sending request for url (…nsis.zip)") before the 11th went through, and
he didn't know what to do. The label was also wrong: the check had
worked; the download failed.
- Progress: "Downloading update… 28% (14.3 MB of 49.9 MB)", "The update
server didn't respond. Trying again in 3 s (attempt 2 of 4)…", from the
native `lotus-update-progress` events (cinny-desktop retries itself).
- Failures name the step (check / download / install, from the native
error prefix) in plain language, with Try again and a Download
installer button (Windows: the setup .exe; else the release page), and
the raw error under "Details".
- Installing from the update toast now shows a "Downloading update"
toast, and on failure a sticky "Update didn't install" toast that
retries on click and points at Settings → General → App Updates.
Before, the toast vanished and the failure was only visible in Settings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Notification sounds called `audio.play()` without handling the promise,
so every message that arrived before the user interacted with the page
(e.g. right after launch) logged an uncaught NotAllowedError — 15 in a
short test run. Same pattern in the video thumbnail loader, the voice
preview (which now also resets its Play button) and useMediaPlay.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
A "Mark all read" chip appears in the Threads panel header whenever any
thread in the room has unread replies, and sends one threaded receipt per
unread thread (never the root — a root lives in the main timeline and a
receipt there would drag the MAIN read marker backwards, the P6
regression). Honours the private-receipt settings.
Verified: two threads with unread replies → the chip appears, the rows
say "unread" in their labels; after clicking, no unread rows and the chip
is gone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Verifying the KaTeX checklist: `$x^2$` in inline code was turned into
math and the backticks were left as literal text, because the math split
runs before markdown. The splitter now skips backtick code spans (N ticks
close with N), so markdown's inline code wins: wire is
<code>$y^2$</code>. Fenced blocks were already fine.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Room Settings → General now reads e.g. "Encrypted · Invite only ·
History visible to members since they joined" under the name — derived
from m.room.encryption, m.room.join_rules and m.room.history_visibility
with the same words the preview-card chips and the history setting use.
One Text line, no card, no icons, no controls; spec defaults (invite-only,
shared history) when a state event is absent. Unit-tested.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
utils/haptics.ts: tick('ptt-on' | 'ptt-off' | 'reaction') → 10/10/8 ms
navigator.vibrate, a no-op without the API (iOS), when the system prefers
reduced motion, or when the new Settings → Calls "Haptic Feedback" switch
(default on, only rendered where the API exists) is off. PTT is observed
once through pttActiveAtom so the keyboard, global-hotkey and on-screen
paths all tick; reactions tick where the reaction event is sent in the
room and thread timelines (quick bar, hover bar, sheet and emoji board
all funnel there).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Room context menu → "Add to Section" submenu: every u.<name> tag in use
across your rooms as a checkable item, plus a "New section…" field
(validated: non-empty, ≤ 40 chars, no dots, unique). Toggling writes or
deletes the standard u.<name> room tag (order 0.5), so sections sync
across devices and other clients see the same tags. The menu row reads
"Sections: Raids, Off-topic" once a room is in any.
Home renders each section as a collapsible category between Favorites
and Rooms (alphabetical; members by tag order then name; the same
closed-state store and unread-only-when-collapsed behaviour as the built-in
categories). A sectioned room leaves the plain Rooms list but keeps a
Favorite / Low Priority placement. Empty sections don't exist by
construction; rename is retag (v2). Derivation in utils/roomSections.ts
with unit tests.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
One tile under Settings → General → Storage: total local usage vs the
browser's quota from navigator.storage.estimate(), the Chromium
usageDetails breakdown when available (IndexedDB = sync cache + encryption
keys + search index; Cache Storage = offline app files), and whether the
browser granted persistent storage — green "Protected" or amber "May be
evicted" with a "Keep my data" button that calls storage.persist(). That
last bit is the useful diagnostic for the KE-1 storage-eviction cluster.
No clear button: media lives in the browser's own HTTP cache (not in the
estimate, and not clearable from a page), the crypto store must never be
casually cleared, and the search-index clear already lives in Message
Search. The About page's "Clear Cache & Reload" now says it deletes this
device's encryption keys too. Hidden entirely when estimate() is missing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
While joined, a device-local record {roomId, deviceId, joinedAt,
lastSeen, mic, video} is written and refreshed every 30 s; a deliberate
hangup (HangupCall/Close) and logout clear it. On the next start, once
the first sync is in and the room's MatrixRTC session has reported its
members (waited for up to 10 s — it fills a moment after sync), a fresh
record (< 10 min) for this device with someone still in the call yields
either a sticky "Rejoin voice?" toast (tap to rejoin) or an automatic
rejoin, per the new Settings → Calls → After a Restart (Ask / Rejoin
automatically / Do nothing; default Ask). Skipped when our own membership
is already live from another device, or only our stale one is left.
Mic/camera state comes from the record (camera still gated by
cameraOnJoin); PTT is applied by startCall as usual.
Pure decision in utils/callRejoin.ts with tests. Verified headless:
reload mid-call → toast → tap → back in the call; hangup → reload → no
toast; auto mode → back in without a prompt.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
A reply to an image, video or sticker used to quote just the filename.
The quote (timeline) and the composer's reply-draft preview now carry a
36 px thumbnail from the event's own thumbnail, decrypted for E2EE media
via the same hook the gallery uses — never the full-size file. Clicking
still jumps to the original.
useDecryptedMediaUrl and getThumbMxc moved out of MediaGallery into
hooks/ and utils/ so components/message can use them without a cycle.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Metadata was only dropped as a side effect of opt-in compression, so a
phone photo carried its GPS fix, camera model and timestamp into the room
and the media store. utils/stripImageMetadata.ts now removes it at the
container level, without touching pixels: JPEG drops APP1/APP13/COM
(writing back a minimal EXIF holding only Orientation when it isn't 1,
so sideways-stored photos still display upright), PNG drops eXIf and the
text chunks XMP lives in, WebP drops EXIF/XMP and clears the VP8X flags.
Other types pass through.
Applied before encryption on every composer path (attach, paste, drop,
share target) and to user/room avatar picks; GIF upload is excluded.
Setting → General → Privacy "Remove Photo Metadata Before Sending",
default on. The upload card says "Photo metadata removed".
Unit tests on generated fixtures with a GPS IFD (JPEG orientation 6,
JPEG + comment, PNG with eXIf + XMP, WebP with EXIF); verified end to
end: the bytes stored by Synapse decode fine and carry only Orientation.
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
public/manifest.json declares share_target (POST multipart to
/share-target: title/text/url + image/video/audio/pdf/text files). The
service worker answers that POST itself: it stashes the form in a Cache
API bucket and 303s to the in-app /share page, which lists what arrived,
offers a room search, and on pick writes the files into that room's
upload-board atom (encrypting first for E2EE rooms via the composer's
shared filesToUploadItems) and the title/text/url into its draft, then
opens the room — the user still presses Send. The stash is cleared once
placed; reopening /share afterwards says so.
nginx/caddy examples and the prod image config gain a 303 for
/share-target so a POST that reaches the origin before the worker
controls the page lands on /share instead of a 405. iOS has no share
target support and ignores the manifest entry.
Verified headless against the built preview: SW-controlled page → POST
/share-target (two PNGs + title + text) → /share lists both files and the
text → pick the DM → composer shows both files on the upload board and
the text in the draft → /share reports nothing pending.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Client-side only: every file is still its own standard m.image/m.video
event, so Element and friends keep seeing N plain images. In Lotus a run
of media from one sender — contiguous, ≤ 60 s apart, no reply/thread/edit
relation, up to 10 — renders once, at its last event, as a 2–4 column
grid of square thumbnails (blurhash placeholder, video play badge,
tap-to-load when media auto-load is off). A member with reactions or a
thread closes its group so those stay visible under the rendered event.
Tapping a tile opens the lightbox on just that group in send order
(←/→, zoom, download, jump). "Show separately" splits a group back into
individual messages for the session; "Show as gallery" undoes it.
Planning is lazy per render pass (utils/mediaGroups.ts, unit-tested):
the first media event met plans its whole run in both directions, so a
virtual window that starts mid-run agrees with one that starts before it.
Verified: 5 files dropped at once in an encrypted room — both sender and
recipient see one 5-tile grid with decrypted thumbnails; desktop + phone;
a reaction on photo 3 yields [1–3]+👍 and [4–5].
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Audit of every rendered time found five families of ad-hoc formatting:
the shared Time component + copies of its today/yesterday branch
(forwarded header, thread summary, read receipts, device tile, moderation
alerts, edit history), locale-default toLocale*String calls that ignored
the user's 12/24 h and date-format settings (scheduled tray, reminders,
schedule preview, notification snooze, bookmarks, threads list, search
cache line, room insights, media gallery), a hard-coded en-US date in the
activity log, and three relative-age variants.
utils/formatTimestamp.ts now holds the rules — today → time; yesterday /
tomorrow → day word + time; last 6 days → weekday + time; older → date +
time in dateFormatString — plus autoDate / time / date / dateTime styles,
formatDayDivider (full weekday), formatShortAge (room list) and
formatRelativeAge (list rows). useTimestampFormatter binds them to the
settings. 11 unit tests with an injected 'now'.
Visible changes are limited to consistency: 12 h times keep the existing
zero-padded hh:mm A; the a11y label and Created-by line use the user's
date format instead of a fixed long month.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Many distinct reactions used to wrap into a tall stack (16 reactions =
4 rows at phone width). Now only the first row is shown, ending in a
"+N" chip; tapping it expands inline and a "less" chip collapses again.
Expanded state is remembered per message for the session. Nothing changes
when the reactions already fit on one row.
Overflowing chips stay in the DOM (invisible, aria-hidden, untabbable,
clipped by max-height) so the container keeps its natural width — which
keeps shrink-to-fit bubble layout stable — and each chip stays measurable.
utils/reactionOverflow.ts holds the unit-tested fit calculation; a
ResizeObserver re-fits on width changes. "+N" is forced LTR for RTL UIs.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Hovering or focusing "(edited)" shows a tooltip with only the most recent
edit as a word diff — removed words struck, added words bold — plus a
+N/−N summary. Clicking still opens the full history viewer. On touch, a
long-press on the label shows the same diff as a popout (a plain tap opens
the viewer; the message's own long-press action sheet is not triggered).
utils/wordDiff.ts is a unit-tested LCS over words that ignores whitespace-
only changes and gives up past 400 words. Only plain-text bodies are
diffed; formatted edits fall back to the viewer.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
One line under the handle in the user profile popover, from the presence the
SDK already receives (currently_active + last_active_ago): 'Active now',
'Active just now', 'Last active 12 min ago / 3 hours ago / yesterday / 5 days
ago / over a month ago'. Nothing is shown without presence data or for an
offline user with no timestamp (Hide Online Status users), so nothing new is
exposed; re-renders on the existing presence events, no polling. Wording
unit-tested; verified headless ('Active now' for an online member).
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
Reported: 'Failed to save status — server may be rate limiting. Try again.'
on the first status change in a month. Root cause: Synapse rate-limits
PUT /presence/{user}/status to ONE request per 10 s per user by default
(rc_presence.per_user: per_second 0.1, burst_count 1), shared across all of
the user's devices, and our presence heartbeat (online/away on visibility and
activity changes, from every open tab/device) spends that budget — so a manual
save that lands within 10 s of a heartbeat gets a 429, which the form showed
as a dead end.
Two fixes: (1) the status save waits out Retry-After (bounded to ~25 s) via
setPresenceWithRetry instead of failing, and the error text now says what
actually happened (rate limit / server text / offline); (2) the heartbeat
dedupes — it only sends when presence or status actually changes (/sync
already keeps us online), so it stops burning the budget in the first place.
Unit-tested; reproduced headless with a routed 10 s limiter: heartbeat ok →
save 429 → retried 8 s later → saved.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
On deafen the participant set is snapshotted; on undeafen it is diffed and,
only if it changed and you were deafened for at least 10 s, one auto-dismissing
toast says 'While you were deafened: Alice, Bob joined · Cole left' (names
capped at 3 + N more). Rides the membership stream that already drives the
join/leave sounds — no new subscriptions; PTT holds don't touch deafen so they
can't trigger it. Verified headless: short deafen → nothing; bob leaves during
an 11 s deafen → 'bob left'; long deafen with no change → nothing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Elbow-on-the-keyboard guard: a PTT hold longer than 5 min (fixed, not a
setting) is released exactly like a keyup — pttActive off, mic muted — with a
toast 'Push to talk released after 5 minutes — press the key again to keep
talking.' A fresh press re-engages normally. One watchdog shared by the DOM
path and the desktop-global hotkey path; keydown auto-repeat no longer
restarts anything; non-PTT mode, deafen and the mic button are untouched.
Helper unit-tested with fake timers; verified headless with Playwright's clock:
held → still open at +4 min → released + toast at +5 min → re-press works.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Room Settings → General → Privacy: 'Your messages in this room' with a
Delete all… flow. The confirm dialog first counts your events with a
server-side sender-filtered /messages walk (live count), then asks to confirm
with the number — typing the room name above 50 — and offers 'Leave the room
afterwards'. Files are called out as not purged by a redaction.
The job runs outside React (closing settings is fine): sequential redactEvent
with 429 back-off, 404/already-redacted skipped, progress on the tile with
Cancel, pending ids persisted per room so a reload shows Resume/Discard, a
toast when done. State events are never touched; reactions, edits and thread
replies you sent are included; encrypted rooms work the same (nothing is
decrypted). Own events need no power level, so it is purely self-service.
Unit tests cover candidate filtering, the server filter, pagination, 429/404
handling and cancel. Verified headless: 62 of bob's events (60 messages, a
reaction, a thread reply) redacted in ~34 s while alice's 10 stayed; cancel at
17/40 → reload → Resume → 'Deleted 40 messages.'
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Message notifications shown through the service worker now carry a text-input
'Reply' action (Chrome desktop/Android). On notificationclick with
action==='reply' the SW sends the typed text itself — it already holds the
newest session's access token for authenticated media — as m.room.message
(threaded when the notification was for a thread), so it works with the tab in
the background or closed; a failed send shows a 'Reply not sent' notification
that opens the room. Not offered for encrypted rooms (the SW cannot encrypt).
The sender lives in swReply.ts so it is unit-tested; verified headless that
the SW notification carries actions + {roomId, threadId}.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
'Unable to decrypt message' now carries one sentence per matrix-js-sdk
DecryptionFailureCode (describeDecryptionFailure, unit-tested against every
code so no raw code can leak into the copy) and, where something fixes it,
one button: no key backup → 'Set up key backup'; backup exists but this
session can't open it / key withheld for an unverified session → 'Unlock key
backup' / 'Verify this session' (both open Settings → Devices via a new
settingsRequestAtom that SettingsTab consumes); backup working or unknown
session (rust-crypto re-requests keys itself) → 'Retry', which re-runs
decryptEventIfNeeded. Sender-side problems are plain text. The raw code sits
in the placeholder's tooltip for support.
Verified headless on a fresh session in the encrypted seed room: each event
shows 'Sent before you signed in here, and no key backup exists…' with
tooltip HISTORICAL_MESSAGE_NO_KEY_BACKUP; the button opens Settings → Devices.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Incident 2026-09-17: a wrong Windows clock broke calls and media keys while
the server answered 200 to everything, with no hint in the UI.
Measurement needs no extra requests and no CORS-exposed headers: every live
event carries origin_server_ts and unsigned.age (our server's now − ts at
response time), so localTimestamp − origin_server_ts is the skew. Only
RoomEvent.Timeline live events count (cache replays have stale age and are
already flagged liveEvent=false by the SDK); the initial network sync
qualifies, so a wrong clock is flagged within seconds of startup. Median of
the last 5 samples, ≥3 needed; warn at |skew| > 30 s, clear below 15 s.
UI: a banner in the sync-status slot — "Your computer's clock is 14 minutes
ahead of the server. Encrypted messages and voice calls will fail until it is
fixed." with a per-OS How-to-fix hint and Dismiss for 24 h — plus the same
line in the call status bar while in a call. Never auto-corrects anything.
Unit-tested (median, hysteresis, stale-age rejection, wording); verified
headless with Playwright's clock skewed +14 min and −3 h (banner, hint,
in-call line, dismiss) and in sync (nothing shown).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
The upload card printed the SDK's toString — 'MatrixError: [413] nope
(http://<hs>/_matrix/media/v3/upload?filename=…)'. describeUploadError() maps
the common cases to one sentence: 413/M_TOO_LARGE → 'This file is larger than
the server allows (limit N)' using m.upload.size when known, 429 → 'Slow down —
try again in a moment.', 401/403 → 'The server refused this upload: <server
text>', 5xx/transport after the retry loop → 'Couldn't reach the server. Check
your connection and retry.', other 4xx → the server's own sentence, URL
stripped. Both card renderers use it; the raw error is still console.warn-ed
by uploadContent for debugging. Unit-tested; verified headless with routed
413/403/503 responses.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
rateLimitedActions now collects non-429 failures (and a 429 that exhausted its
retries) and returns them instead of swallowing them; existing callers ignore
the return. /kick and /ban turn the list into a CommandError whose message
names who and why, using the server's own sentence (MatrixError.data.error),
never the URL-bearing toString(); RoomInput's toast shows it verbatim.
Verified headless as a non-moderator: '/kick @alice' → "Could not kick
@alice:localhost: You cannot kick user @alice:localhost."; '/ban @nobody
@alice' → "Could not ban @nobody:localhost, @alice:localhost: You don't have
permission to ban".
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Verified first on the local calls stack: a participant who left and rejoined
~3 s later played "leave" then "join" every time. Cues now go through
createCallSoundDebouncer (per USER, not per device):
- a leave cue waits 5 s; if the same user is back before it fires, the leave is
cancelled and no join cue is played either;
- a join cue is suppressed for a user who left < 60 s ago;
- same-kind cues within 250 ms collapse, so a batch of leaves still sounds once;
- a second device of someone already present is not a new arrival, and a
device switch is quiet.
Only the sound is debounced; membership UI is unchanged. Sound style, volume
and PTT interplay untouched (the style is read at play time). Timers are
injected — unit-tested with a manual clock. Re-ran the flap scenario headless:
join → flap → (silence) → real leave → one "leave" 5 s later → rejoin within
60 s → silence.
Also enables msc4133 (custom profile fields → in-call avatar decorations) on
the dev Synapse.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
matrix-js-sdk rejects an upload whose XHR ends with status 0 (offline,
connection reset, DNS) with DOMException('AbortError') to mimic fetch,
the same name mx.cancelUpload() produces. isRetryableUploadError bailed
on any AbortError, so the one failure class the retry loop was built
for was never retried. Decide by our own cancel AbortSignal instead.
Verified with Playwright routing the upload endpoint: 502 → network drop
→ ok now completes in 3 attempts (1 s, 2 s back-off) and the image sends;
413 still fails fast after 1 attempt; persistent 503 gives up after 4.
Unit tests in utils/uploadRetry.test.ts.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
RoomTimeline renders a numeric index window into the live timeline's
event arrays; SDK back-pagination prepends, so any side panel calling
paginateEventTimeline(room.getLiveTimeline()) shifted the visible
messages into the past on the next render and broke at-bottom tracking.
New utils/detachedTimeline.ts builds a timeline set that mirrors the
already-loaded history and paginates independently: a room-registered
filtered set (server-side contains_url / types filter) when the filter
is usable, else a private EventTimelineSet seeded from the live timeline.
useRoomMediaTimeline wraps it for the gallery (live events + redactions
handled); RoomActivityLog uses a type filter (safe in encrypted rooms);
ExportRoomHistory pages a private set so a full export no longer parks
thousands of events in the live timeline.
Verified with Playwright against a local Synapse in a 400-message plain
room and a 200-message encrypted room: timeline stays at the bottom
through gallery pages, activity load-more and a full export; live
messages keep auto-scrolling; all media found in both rooms.
Also adds scripts/dev-homeserver.sh + scripts/dev-seed.py (local
throwaway Synapse for driving the real UI) and documents them.
Closes#163
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
LOTUS_TODO.md → LOTUS_REFERENCE.md (design laws, decided deferrals,
server-blocked features, operational reference only). Every open task was
filed: cinny #195–#210, cinny-desktop #15–#18, matrix #8–#10.
LOTUS_TESTING.md keeps the automated-coverage map, the Playwright notes and
the deploy tip; every manual checklist is now a `qa` issue under the
'Manual QA backlog' (cinny #170–#194, #198) and 'Desktop QA backlog'
(cinny-desktop #11–#14, #18) milestones.
Repointed the README, LOTUS_FEATURES, CI and source comments that
referenced LOTUS_TODO.md.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PPmy3tPq869XDW4njjVaKA
Lotus CIandClaude Opus 5.5
jaredandClaude Opus 5