Bump to v2026.9.28

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

.claude/launch.json

@@ -1,17 +0,0 @@
-{
-  "version": "0.0.1",
-  "configurations": [
-    {
-      "name": "frontend",
-      "runtimeExecutable": "npm",
-      "runtimeArgs": ["--prefix", "frontend", "run", "dev"],
-      "port": 5173
-    },
-    {
-      "name": "frontend-preview",
-      "runtimeExecutable": "npm",
-      "runtimeArgs": ["--prefix", "frontend", "run", "preview"],
-      "port": 4173
-    }
-  ]
-}

.gitignore

@@ -17,6 +17,7 @@ dist-ssr/
 ## Editors / OS
 .vscode/
 .idea/
+.claude/
 .DS_Store
 
 ## Test / coverage

CONTRIBUTING.md

@@ -0,0 +1,51 @@
+# Contributing to DS Chat
+
+Thanks for considering a contribution.
+
+## Reporting bugs and requesting features
+
+Open an issue on this project's issue tracker. Include steps to reproduce
+for a bug, or the problem you're trying to solve for a feature request —
+that's usually more useful than a proposed solution.
+
+## Development setup
+
+See the root [README.md](README.md)'s Quickstart, plus
+[backend/README.md](backend/README.md) and
+[frontend/README.md](frontend/README.md) for the full local dev setup
+(Postgres, Redis, Python venv, migrations, the Vite dev server).
+[ARCHITECTURE.md](ARCHITECTURE.md) covers the overall system design if
+you're orienting yourself before a larger change.
+
+## Before opening a pull request
+
+- **Tests**: run `pytest` in `backend/` for any backend change, and add
+  tests for new behavior rather than just the happy path — see
+  `backend/README.md`'s "Run tests" section. For frontend changes, run
+  `npx tsc -b` in `frontend/` and confirm `npm run build` succeeds.
+- **Style**: match the conventions already in the file you're editing
+  rather than introducing a new pattern — this codebase doesn't have a
+  separate style guide beyond "look at what's already there."
+- **Scope**: smaller, focused PRs are easier to review than large ones
+  that mix unrelated changes.
+
+## Contributor terms
+
+By submitting a contribution (a pull request, patch, or similar), you
+agree that:
+
+1. Your contribution is licensed under the project's own license,
+   AGPL-3.0-or-later ([LICENSE](LICENSE)), and
+2. You grant the project's maintainer(s) a perpetual, worldwide,
+   non-exclusive right to also relicense your contribution under different
+   terms — for example, as part of a separately-licensed commercial
+   offering built on this project.
+
+This keeps the option of a future dual-licensed (open-source +
+commercial) version of the project available, without requiring a
+separate signed agreement for every contribution.
+
+*This is a lightweight starting point, not a substitute for legal advice —
+if you're contributing something substantial, or maintaining a fork with
+your own commercial plans, it's worth having this reviewed by a lawyer
+rather than relying on the paragraph above alone.*

DEPLOYMENT.md

@@ -132,35 +132,45 @@ sudo -u ds-chat ssh-keygen -t ed25519 -f /srv/ds-chat/.ssh/id_ed25519 -N ""
 sudo cat /srv/ds-chat/.ssh/id_ed25519.pub
 ```
 
-Add that public key as a **read-only deploy key** on the Gitea repo
+This project is hosted at
-(Settings → Deploy Keys), then:
+**[github.com/ds-ksmith/DS-Chat](https://github.com/ds-ksmith/DS-Chat)**.
+Add that public key there as a **read-only deploy key** (Settings → Deploy
+Keys on the repo), then:
 
 ```bash
-sudo -u ds-chat ssh-keyscan git.darksingularity.org >> /srv/ds-chat/.ssh/known_hosts
+sudo -u ds-chat ssh-keyscan github.com >> /srv/ds-chat/.ssh/known_hosts
-sudo -u ds-chat git clone git@git.darksingularity.org:DarkSingularity/ds-chat.git /srv/ds-chat
+sudo -u ds-chat git clone git@github.com:ds-ksmith/DS-Chat.git /srv/ds-chat
 ```
 
-(If your Gitea's SSH is on a non-default port, adjust the clone URL and
+(Deploying from your own fork instead? Substitute its clone URL — the same
-`ssh-keyscan -p <port>` accordingly.)
+deploy-key/access-token steps work the same way on GitHub, GitLab, Gitea,
+and most other git hosts.)
 
 **Alternative: a personal/deployment-user access token instead of a deploy
 key** — skip the `.ssh`/`ssh-keygen`/`ssh-keyscan` commands above entirely
 and clone over HTTPS with the token embedded in the URL:
 
 ```bash
-sudo -u ds-chat git clone https://<TOKEN>@git.darksingularity.org/DarkSingularity/ds-chat.git /srv/ds-chat
+sudo -u ds-chat git clone https://<TOKEN>@github.com/ds-ksmith/DS-Chat.git /srv/ds-chat
 ```
 
 The token then lives in plaintext in `/srv/ds-chat/.git/config` (`git
 remote -v` shows it) — readable by root and the `ds-chat` user, not by
 anyone else under normal file permissions. `deploy/upgrade.sh`'s later
-`git pull`s reuse this same authenticated URL automatically, no extra
+`git fetch`es reuse this same authenticated URL automatically, no extra
 setup needed. Fine as long as the token is scoped to read-only access on
 just this repo.
 
-Either way, now that the repo is cloned:
+Either way, now that the repo is cloned, check out the latest release tag
+rather than deploying whatever the default branch's tip happens to be —
+`deploy/upgrade.sh` follows the same rule on every later upgrade (see §6),
+so this keeps the very first deploy consistent with all the ones after it:
 
 ```bash
+cd /srv/ds-chat
+sudo -u ds-chat git fetch --tags
+LATEST_TAG="$(sudo -u ds-chat git tag --sort=-creatordate | head -n1)"
+sudo -u ds-chat git checkout --detach "$LATEST_TAG"
 sudo -u ds-chat mkdir -p /srv/ds-chat/uploads
 ```
 
@@ -232,7 +242,16 @@ admin sets it up.
 `/ws`) whenever that directory exists — that's what lets Nginx Proxy
 Manager forward the whole domain to one port with no custom path routing.
 
+Before building, copy `frontend/.env.example` to `frontend/.env.production`
+and set `VITE_SOURCE_URL` to wherever *your* copy of the repo lives — see
+that file's own comment for why this matters (AGPL-3.0 source-availability
+compliance). Vite bakes this in at build time, so it needs to be in place
+before `npm run build` runs, and needs re-running after any future change
+to it.
+
 ```bash
+sudo -u ds-chat cp /srv/ds-chat/frontend/.env.example /srv/ds-chat/frontend/.env.production
+sudo -u ds-chat nano /srv/ds-chat/frontend/.env.production  # set VITE_SOURCE_URL
 sudo -u ds-chat bash -c 'cd /srv/ds-chat/frontend && npm ci && npm run build'
 ```
 
@@ -320,12 +339,15 @@ This is config in NPM's own UI/database, not a file this repo ships:
 sudo -u ds-chat /srv/ds-chat/deploy/upgrade.sh
 ```
 
-Pulls latest `main`, reinstalls backend deps, runs `alembic upgrade head`,
+Fetches tags and checks out whichever one sorts newest (`git tag
-rebuilds the frontend, restarts `ds-chat`, and curls `/api/health` to
+--sort=-creatordate`) — deliberately not the default branch's tip, so
-confirm it came back up. Fails loudly (`set -euo pipefail`) and stops
+running this between releases is a safe no-op rather than pulling in
-before restarting anything if an earlier step — most importantly a failed
+whatever's mid-flight on `main`. Then reinstalls backend deps, runs
-migration — errors out, so a bad deploy doesn't take down the previously
+`alembic upgrade head`, rebuilds the frontend, restarts `ds-chat`, and
-working one.
+curls `/api/health` to confirm it came back up. Fails loudly
+(`set -euo pipefail`) and stops before restarting anything if an earlier
+step — most importantly a failed migration — errors out, so a bad deploy
+doesn't take down the previously working one.
 
 Active users get disconnected for a few seconds during the restart and
 reconnect automatically (same reconnect logic as §4's NPM-timeout note) —

LICENSE

@@ -629,8 +629,8 @@ to attach them to the start of each source file to most effectively
 state the exclusion of warranty; and each file should have at least
 the "copyright" line and a pointer to where the full notice is found.
 
-    <one line to give the program's name and a brief idea of what it does.>
+    DS Chat, a self-hosted, real-time team chat service.
-    Copyright (C) <year>  <name of author>
+    Copyright (C) 2026  Keith Smith
 
     This program is free software: you can redistribute it and/or modify
     it under the terms of the GNU Affero General Public License as published by

backend/README.md

@@ -822,10 +822,13 @@ preview card fetched from that page's Open Graph tags (`og:title`,
   `Content-Type` is `text/html`.
 - **Cached by URL, not by message** (`link_previews` table, unique on
   `url`) — a URL posted by five different people in five different rooms
-  fetches once. A row also gets written on a *failed* fetch
+  within the same short window fetches once. A row also gets written on a
-  (`fetch_failed=True`) so a URL that genuinely doesn't unfurl (SSRF
+  *failed* fetch (`fetch_failed=True`) so a URL that genuinely doesn't
-  rejection, timeout, no usable title) isn't re-attempted on every message
+  unfurl (SSRF rejection, timeout, no usable title) isn't re-attempted on
-  that references it; both kinds expire after 7 days (`_CACHE_TTL`).
+  every message that references it; both kinds expire after 5 minutes
+  (`_CACHE_TTL` — #70: was 7 days, confirmed live as far too long, a
+  re-posted URL whose title/content had genuinely changed kept showing
+  the stale first-fetch preview for up to a week).
 - Parsed with stdlib `html.parser.HTMLParser`, not a new dependency — only
   meta-tag scraping is needed, not general HTML parsing.
 - Editing a message re-extracts the URL; if it changed or was removed, the

backend/alembic/versions/339b78011a4f_add_user_text_scale_preference.py

@@ -0,0 +1,32 @@
+"""add user text_scale preference
+
+Revision ID: 339b78011a4f
+Revises: a318850726ee
+Create Date: 2026-08-30 18:03:16.159924
+
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision: str = '339b78011a4f'
+down_revision: Union[str, Sequence[str], None] = 'a318850726ee'
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    """Upgrade schema."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.add_column('users', sa.Column('text_scale', sa.String(length=20), nullable=True))
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    """Downgrade schema."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.drop_column('users', 'text_scale')
+    # ### end Alembic commands ###

backend/alembic/versions/e81c9bcc82b9_add_user_emoji_scale_preference.py

@@ -0,0 +1,32 @@
+"""add user emoji_scale preference
+
+Revision ID: e81c9bcc82b9
+Revises: 339b78011a4f
+Create Date: 2026-08-30 18:14:26.045186
+
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision: str = 'e81c9bcc82b9'
+down_revision: Union[str, Sequence[str], None] = '339b78011a4f'
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    """Upgrade schema."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.add_column('users', sa.Column('emoji_scale', sa.String(length=20), nullable=True))
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    """Downgrade schema."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.drop_column('users', 'emoji_scale')
+    # ### end Alembic commands ###

backend/app/main.py

@@ -58,8 +58,12 @@ async def lifespan(app: FastAPI) -> AsyncIterator[None]:
     # ASGITransport, used by the plain REST test fixtures -- only
     # TestClient's websocket_connect-based tests actually run lifespan).
     # The background listener task genuinely needs a running event loop
-    # though, so that part stays here.
+    # though, so that part stays here. run_forever() (not listen() directly)
-    listener_task = asyncio.create_task(app.state.broadcaster.listen())
+    # -- #76: a bare listen() here died permanently and silently the moment
+    # its Redis connection ever dropped (e.g. Redis itself restarting),
+    # with nothing to notice or reconnect; run_forever() supervises it and
+    # restarts on failure instead.
+    listener_task = asyncio.create_task(app.state.broadcaster.run_forever())
 
     yield
 

backend/app/models/user.py

@@ -19,6 +19,17 @@ class User(Base):
     is_active: Mapped[bool] = mapped_column(Boolean, default=True, nullable=False)
     display_name: Mapped[str | None] = mapped_column(String(50))
     theme: Mapped[str | None] = mapped_column(String(20))
+    # #71: null means "normal" (the pre-existing default before this
+    # setting existed) -- a preset name, not a raw scale factor, so it's
+    # validated/enumerable the same way `theme` already is rather than
+    # accepting an arbitrary float.
+    text_scale: Mapped[str | None] = mapped_column(String(20))
+    # #71: independent of text_scale above -- scales emoji rendered in
+    # message text specifically, not the whole UI (see
+    # frontend/src/components/MessageContent.tsx's --emoji-scale, scoped
+    # to message content only so it can't also inflate the emoji picker's
+    # grid or reaction pills).
+    emoji_scale: Mapped[str | None] = mapped_column(String(20))
     # Only meaningful when theme == "custom" -- which of this user's saved
     # CustomTheme rows (app/models/custom_theme.py) is currently active.
     # Cleared explicitly (not via a DB-level ON DELETE) whenever that theme

backend/app/routers/auth.py

@@ -100,6 +100,10 @@ async def update_profile(
         current_user.display_name = display_name or None
     if "theme" in updates:
         current_user.theme = updates["theme"]
+    if "text_scale" in updates:
+        current_user.text_scale = updates["text_scale"]
+    if "emoji_scale" in updates:
+        current_user.emoji_scale = updates["emoji_scale"]
     if "appear_offline" in updates:
         current_user.appear_offline = updates["appear_offline"]
     await db.commit()

backend/app/routers/rooms.py

@@ -39,12 +39,14 @@ from app.schemas.webhook import (
     WebhookIncomingRead,
 )
 from app.services.link_preview_service import get_link_previews_for_urls
-from app.services.message_events import broadcast_room_added
+from app.services.message_events import broadcast_new_message, broadcast_room_added
 from app.services.message_service import (
+    create_message,
     get_reactions_for_messages,
     list_recent_messages,
     list_room_attachments,
 )
+from app.services.system_user_service import get_or_create_system_user
 from app.services.upload_settings_service import format_mb, get_upload_settings
 from app.services.room_service import (
     AlreadyMemberError,
@@ -655,7 +657,37 @@ async def add_member_endpoint(
         raise HTTPException(status_code=404, detail="No user with that ID")
     except AlreadyMemberError:
         raise HTTPException(status_code=409, detail="That user is already a member")
+
+    # room_added first -- the new member's client needs to know this room
+    # exists before it can make sense of an unread_update for it, which the
+    # welcome message below would otherwise trigger out of order.
     await broadcast_room_added(request.app.state.broadcaster, data.user_id, room)
+
+    # #74: posted as the auto-provisioned System account, not the admin who
+    # did the adding -- "Welcome, bob!" reads as coming from the room/app
+    # itself, not as something the admin personally typed.
+    system_user = await get_or_create_system_user(db)
+    welcome_name = membership.user.display_name or membership.user.username
+    welcome_message = await create_message(
+        db, room.id, system_user.id, f"Welcome to #{room.name}, {welcome_name}!"
+    )
+    # Same "sending implies having seen the room" reasoning as ws/chat.py's
+    # own live-message path -- the admin is the one who caused this message,
+    # and is presumably already looking at this room's member management, so
+    # without this their own client would show it as unread regardless.
+    await mark_room_read(db, room.id, current_user.id)
+    await broadcast_new_message(
+        db,
+        request.app.state.broadcaster,
+        request.app.state.presence,
+        request.app.state.focus_presence,
+        request.app.state.global_presence,
+        str(request.base_url),
+        room.id,
+        welcome_message,
+        system_user,
+    )
+
     online_ids = await request.app.state.global_presence.online_user_ids([membership.user_id])
     return RoomMemberRead(
         user_id=membership.user_id,

backend/app/schemas/user.py

@@ -23,6 +23,8 @@ class UserRead(BaseModel):
     is_site_admin: bool
     display_name: str | None
     theme: str | None
+    text_scale: str | None
+    emoji_scale: str | None
     # Resolved, not just an id -- the frontend needs the actual palette to
     # paint on load without a second round trip (see lib/theme.ts).
     active_custom_theme: CustomThemeRead | None
@@ -57,6 +59,10 @@ class ProfileUpdate(BaseModel):
     # ownership check; that's POST /api/custom-themes/{id}/activate, not a
     # bare theme name with nothing to point it at.
     theme: Literal["dark", "light", "midnight", "sunset"] | None = Field(default=None)
+    # #71: kept in sync with frontend/src/lib/theme.ts's TEXT_SCALE_PERCENT map.
+    text_scale: Literal["small", "normal", "large", "xlarge"] | None = Field(default=None)
+    # #71: kept in sync with MessageContent.tsx's EMOJI_SCALE_MULTIPLIER map.
+    emoji_scale: Literal["small", "normal", "large", "xlarge"] | None = Field(default=None)
     appear_offline: bool | None = Field(default=None)
 
 

backend/app/services/link_preview_service.py

@@ -27,8 +27,18 @@ _TRAILING_PUNCTUATION = ".,;:!?)'\">"
 _FETCH_TIMEOUT_SECONDS = 5.0
 _MAX_BYTES = 512 * 1024
 _MAX_REDIRECTS = 3
+# #70: was 7 days -- confirmed live as too long for how this app actually
+# gets used: re-posting a URL whose title/content had genuinely changed
+# kept showing the stale first-fetch preview for up to a week. Short
+# enough that it's effectively "always fresh" for any realistic human
+# posting cadence, while still doing the one thing a cache here is
+# actually for -- collapsing a burst of near-simultaneous fetches of the
+# same URL (several people pasting the same link within moments of each
+# other, or the same person's message history being loaded repeatedly)
+# into one, and not hammering a URL that just failed on every message
+# that references it.
 _USER_AGENT = "ds-chat-link-preview/1.0"
-_CACHE_TTL = timedelta(days=7)
+_CACHE_TTL = timedelta(minutes=5)
 
 
 def extract_first_url(content: str | None) -> str | None:

backend/app/services/system_user_service.py

@@ -0,0 +1,45 @@
+from sqlalchemy import select
+from sqlalchemy.exc import IntegrityError
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.models import User
+from app.security import generate_token, hash_password
+
+# #74: one well-known, auto-provisioned bot account the app itself posts as
+# for automated first-party messages (the #72 welcome message, and whatever
+# comes next) -- distinct from bot_service.py's admin-created integration
+# bots, which each need a human actor and audit-log entry for creating them.
+# There's no actor here: this account is provisioned lazily, the first time
+# something needs to post as it.
+SYSTEM_USERNAME = "system"
+
+
+async def get_or_create_system_user(db: AsyncSession) -> User:
+    result = await db.execute(select(User).where(User.username == SYSTEM_USERNAME))
+    user = result.scalar_one_or_none()
+    if user is not None:
+        return user
+
+    # Same placeholder-email/discarded-password shape as bot_service.create_bot
+    # -- this account never logs in, email just satisfies the NOT NULL/unique
+    # column.
+    user = User(
+        username=SYSTEM_USERNAME,
+        email=f"{SYSTEM_USERNAME}@bots.example.com",
+        password_hash=hash_password(generate_token()),
+        is_bot=True,
+    )
+    db.add(user)
+    try:
+        await db.flush()
+    except IntegrityError:
+        # Two concurrent requests both found no existing row and raced to
+        # create one -- the loser just reads back the winner's row instead
+        # of erroring.
+        await db.rollback()
+        result = await db.execute(select(User).where(User.username == SYSTEM_USERNAME))
+        return result.scalar_one()
+
+    await db.commit()
+    await db.refresh(user)
+    return user

backend/app/ws/broadcaster.py

@@ -1,13 +1,26 @@
+import asyncio
+import contextlib
 import json
+import logging
 import uuid
 
 from redis.asyncio import Redis
 
 from app.ws.connection_manager import ConnectionManager
 
+logger = logging.getLogger(__name__)
+
 ROOM_CHANNEL_PREFIX = "room:"
 USER_CHANNEL_PREFIX = "user:"
 
+# #76: a run shorter than this before dying doesn't count as "stable" for
+# resetting the backoff in run_forever() below -- long enough that a real,
+# working connection has clearly settled in, short enough not to leave a
+# just-recovered listener exposed to a full backoff climb if Redis blips
+# again shortly after reconnecting.
+STABLE_RUN_SECONDS = 60
+MAX_BACKOFF_SECONDS = 30
+
 
 class Broadcaster:
     """Cross-instance message fan-out (ARCHITECTURE.md phase 5).
@@ -36,6 +49,43 @@ class Broadcaster:
     async def publish_to_user(self, user_id: uuid.UUID, payload: dict) -> None:
         await self._redis.publish(f"{USER_CHANNEL_PREFIX}{user_id}", json.dumps(payload))
 
+    async def run_forever(self) -> None:
+        """Supervises listen(), restarting it if it ever dies.
+
+        #76's real root cause, found after the per-message try/except below
+        turned out not to be enough: this is the *one* pubsub listener for
+        the whole process (see main.py's lifespan), created once at startup
+        with nothing watching over it afterward. A connection-level failure
+        -- Redis restarting, a network blip between the app and Redis --
+        raises from the `async for` iteration in listen() itself, outside
+        that per-message guard, and previously killed this task permanently
+        and silently: HTTP/REST and login kept working fine (neither
+        touches this), but live message delivery was gone for every room
+        and every user on this instance until the whole process was
+        restarted. Confirmed live in production: Redis got restarted by an
+        OS update, and `PUBSUB NUMPAT` against it afterward showed zero
+        pattern subscriptions across all 4 gunicorn workers -- every one of
+        them had silently lost its listener with nothing to bring it back.
+        """
+        backoff_seconds = 1.0
+        while True:
+            started_at = asyncio.get_event_loop().time()
+            try:
+                await self.listen()
+                logger.warning("Broadcaster.listen() exited without error -- restarting")
+            except asyncio.CancelledError:
+                raise
+            except Exception:
+                logger.error("Broadcaster.listen() failed -- restarting", exc_info=True)
+            ran_for = asyncio.get_event_loop().time() - started_at
+            # Sleep at the *current* backoff first, then grow it for next
+            # time -- computing the next value before sleeping would skip
+            # straight past the first (shortest) retry delay.
+            await asyncio.sleep(backoff_seconds)
+            backoff_seconds = (
+                1.0 if ran_for > STABLE_RUN_SECONDS else min(backoff_seconds * 2, MAX_BACKOFF_SECONDS)
+            )
+
     async def listen(self) -> None:
         pubsub = self._redis.pubsub()
         await pubsub.psubscribe(f"{ROOM_CHANNEL_PREFIX}*", f"{USER_CHANNEL_PREFIX}*")
@@ -43,6 +93,16 @@ class Broadcaster:
             async for message in pubsub.listen():
                 if message["type"] != "pmessage":
                     continue
+                # #76: this is the one pubsub listener for the whole process
+                # (see main.py's lifespan) -- ConnectionManager.broadcast/
+                # send_to_user already guard against one dead socket taking
+                # the rest of a single fan-out down, but this belt-and-
+                # suspenders catch is for anything else unexpected (a
+                # malformed payload, e.g.) doing the same. An uncaught
+                # exception here previously meant *this whole async for*
+                # loop died silently -- no more live delivery to any room on
+                # this instance, for anyone, until the process restarted.
+                try:
                     channel = message["channel"]
                     payload = json.loads(message["data"])
                     if channel.startswith(ROOM_CHANNEL_PREFIX):
@@ -51,6 +111,19 @@ class Broadcaster:
                     elif channel.startswith(USER_CHANNEL_PREFIX):
                         user_id = uuid.UUID(channel.removeprefix(USER_CHANNEL_PREFIX))
                         await self._manager.send_to_user(user_id, payload)
+                except Exception:
+                    logger.error("Error handling pubsub message on channel %s", message.get("channel"), exc_info=True)
         finally:
+            # #76: reached via a connection that's already dead just as
+            # often as a clean shutdown (that's the whole failure mode
+            # run_forever() now recovers from) -- punsubscribe/aclose over
+            # a broken connection would themselves raise, masking whatever
+            # actually killed the loop above and, since run_forever() only
+            # starts backing off *after* listen() returns/raises, delaying
+            # the reconnect for no benefit (a fresh pubsub object is about
+            # to be created on retry regardless of whether this cleanup
+            # succeeded).
+            with contextlib.suppress(Exception):
                 await pubsub.punsubscribe(f"{ROOM_CHANNEL_PREFIX}*", f"{USER_CHANNEL_PREFIX}*")
+            with contextlib.suppress(Exception):
                 await pubsub.aclose()

backend/app/ws/chat.py

@@ -1,3 +1,4 @@
+import asyncio
 import uuid
 
 from fastapi import APIRouter, Depends, WebSocket, WebSocketDisconnect
@@ -31,6 +32,18 @@ router = APIRouter(tags=["ws"])
 
 WS_UNAUTHENTICATED = 4401
 
+# #76: comfortably under the ~60s idle-connection timeout common to reverse
+# proxies/load balancers sitting in front of this endpoint, and short enough
+# that the frontend's own zombie-detection watchdog (useChatSocket.ts) can
+# notice a dead connection and reconnect well within a session -- a socket
+# can go silently dead (no close frame at all, e.g. a laptop sleeping or a
+# NAT dropping an idle mapping) without this, and no live message reaches it
+# again until *something* else finally triggers a close, which used to be
+# "switch rooms and back" purely by accident (that remounts the message
+# list, which refetches history over plain REST -- unrelated to the socket
+# itself actually recovering).
+WS_PING_INTERVAL_SECONDS = 30
+
 
 class ClientEnvelope(BaseModel):
     type: str
@@ -124,7 +137,18 @@ async def chat_endpoint(websocket: WebSocket, db: AsyncSession = Depends(get_db)
 
     try:
         while True:
-            raw = await websocket.receive_json()
+            try:
+                raw = await asyncio.wait_for(
+                    websocket.receive_json(), timeout=WS_PING_INTERVAL_SECONDS
+                )
+            except asyncio.TimeoutError:
+                # Nothing received in a while -- ping rather than assume the
+                # connection's still good. A genuinely dead socket fails this
+                # send (raising, caught below same as any other disconnect);
+                # a merely quiet one just gets proded and the loop continues
+                # waiting.
+                await websocket.send_json({"type": "ping"})
+                continue
             try:
                 try:
                     envelope = ClientEnvelope.model_validate(raw)
@@ -146,6 +170,14 @@ async def chat_endpoint(websocket: WebSocket, db: AsyncSession = Depends(get_db)
                     joined_rooms.add(envelope.room_id)
                     await websocket.send_json({"type": "joined", "room_id": str(envelope.room_id)})
 
+                elif envelope.type == "pong":
+                    # #76: the client's reply to our "ping" above -- nothing
+                    # to do with it. Its only purpose is resetting the
+                    # frontend's own liveness watchdog; the timeout on the
+                    # receive_json() above already gets reset by receiving
+                    # any frame at all, this one included.
+                    pass
+
                 elif envelope.type == "leave":
                     if envelope.room_id is None:
                         await websocket.send_json({"type": "error", "detail": "room_id required"})

backend/app/ws/connection_manager.py

@@ -1,8 +1,11 @@
+import logging
 import uuid
 from collections import defaultdict
 
 from fastapi import WebSocket
 
+logger = logging.getLogger(__name__)
+
 
 class ConnectionManager:
     """Local, single-process WebSocket socket registry.
@@ -43,8 +46,22 @@ class ConnectionManager:
 
     async def broadcast(self, room_id: uuid.UUID, payload: dict) -> None:
         for websocket in list(self._rooms.get(room_id, ())):
+            # #76: one dead/racing socket (e.g. its own disconnect cleanup
+            # hasn't finished removing it from _rooms yet) must not take the
+            # rest of this broadcast down with it -- this runs inside
+            # Broadcaster.listen()'s single, never-restarted per-process
+            # pubsub loop, so an unhandled exception here previously meant
+            # one bad socket silently killed *live delivery for every room
+            # on this instance*, not just the one socket, until the process
+            # was restarted.
+            try:
                 await websocket.send_json(payload)
+            except Exception:
+                logger.warning("Dropping dead socket during broadcast to room %s", room_id, exc_info=True)
 
     async def send_to_user(self, user_id: uuid.UUID, payload: dict) -> None:
         for websocket in list(self._users.get(user_id, ())):
+            try:
                 await websocket.send_json(payload)
+            except Exception:
+                logger.warning("Dropping dead socket sending to user %s", user_id, exc_info=True)

backend/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ds-chat"
-version = "1.1.0"
+version = "2026.9.28"
 description = "DS Chat backend service"
 license = { text = "AGPL-3.0-or-later" }
 requires-python = ">=3.11"

backend/tests/test_broadcaster_supervision.py

@@ -0,0 +1,61 @@
+import asyncio
+
+import pytest
+
+from app.ws.broadcaster import Broadcaster
+
+
+async def test_run_forever_restarts_after_listen_raises(monkeypatch):
+    # #76: listen() dying (a dropped Redis connection, most commonly -- e.g.
+    # Redis itself restarting) must not permanently kill live delivery for
+    # the whole process. run_forever() has to notice and restart it, not
+    # just let the background task disappear silently while everything else
+    # (login, REST endpoints) keeps working fine.
+    broadcaster = Broadcaster(redis=None, manager=None)
+    calls = 0
+
+    async def fake_listen():
+        nonlocal calls
+        calls += 1
+        if calls < 3:
+            raise ConnectionError("redis went away")
+        # Stop the loop once the retry behavior below is proven, rather
+        # than looping forever in a test.
+        raise asyncio.CancelledError()
+
+    monkeypatch.setattr(broadcaster, "listen", fake_listen)
+    sleep_calls = []
+
+    async def fake_sleep(seconds):
+        sleep_calls.append(seconds)
+
+    monkeypatch.setattr(asyncio, "sleep", fake_sleep)
+
+    with pytest.raises(asyncio.CancelledError):
+        await broadcaster.run_forever()
+
+    assert calls == 3
+    # Slept once after each of the first two failures (exponential:
+    # 1s, then 2s) -- not after the third, which is the CancelledError
+    # that stops the loop before reaching the sleep.
+    assert sleep_calls == [1.0, 2.0]
+
+
+async def test_run_forever_does_not_swallow_cancellation(monkeypatch):
+    # A genuine shutdown (the app's own lifespan cancels this task) must
+    # propagate immediately, not get treated as just another failure to
+    # retry after a backoff.
+    broadcaster = Broadcaster(redis=None, manager=None)
+
+    async def fake_listen():
+        raise asyncio.CancelledError()
+
+    monkeypatch.setattr(broadcaster, "listen", fake_listen)
+
+    async def fail_if_called(seconds):
+        raise AssertionError("should not sleep/retry on cancellation")
+
+    monkeypatch.setattr(asyncio, "sleep", fail_if_called)
+
+    with pytest.raises(asyncio.CancelledError):
+        await broadcaster.run_forever()

backend/tests/test_link_previews.py

@@ -1,9 +1,13 @@
 import asyncio
 import uuid
+from datetime import datetime, timedelta, timezone
 
 import pytest_asyncio
+from sqlalchemy import select
 
+from app.database import async_session_factory
 from app.database import engine as _link_preview_engine
+from app.models import LinkPreview
 from app.schemas.user import UserCreate
 from app.services.auth_service import register_user
 from app.services.link_preview_service import extract_first_url
@@ -294,3 +298,66 @@ async def test_link_preview_reused_across_messages_with_same_url(client, db_sess
     history = (await client.get(f"/api/rooms/{room['id']}/messages")).json()
     assert len(history) == 2
     assert all(m["link_preview"]["title"] == "Example Article" for m in history)
+
+
+async def test_link_preview_refetches_after_cache_expires(client, db_session, monkeypatch):
+    # #70: a real report -- re-posting a URL whose title had genuinely
+    # changed kept showing the stale first-fetch preview, because the
+    # cache TTL used to be 7 days. Simulates that expiry directly (rather
+    # than actually sleeping 5+ minutes) by backdating the cached row's
+    # fetched_at past the TTL, then confirms a second post of the same URL
+    # picks up new content instead of the stale cached title.
+    captured_tasks: list[asyncio.Task] = []
+    real_create_task = asyncio.create_task
+
+    def fake_create_task(coro):
+        task = real_create_task(coro)
+        captured_tasks.append(task)
+        return task
+
+    monkeypatch.setattr("app.services.message_events.asyncio.create_task", fake_create_task)
+    calls: list[str] = []
+    monkeypatch.setattr(
+        "app.services.link_preview_service.httpx.AsyncClient",
+        _fake_client_factory(call_log=calls),
+    )
+    url = _unique_url()
+
+    await register_and_login(client, db_session, username="alice")
+    room = (await client.post("/api/rooms", json={"name": "general"})).json()
+    webhook = (await client.post(f"/api/rooms/{room['id']}/webhooks/incoming", json={})).json()
+
+    resp1 = await client.post(
+        f"/api/webhooks/incoming/{webhook['token']}", json={"content": f"see {url}"}
+    )
+    assert resp1.status_code == 204
+    await asyncio.gather(*captured_tasks)
+    captured_tasks.clear()
+
+    async with async_session_factory() as session:
+        row = (
+            await session.execute(select(LinkPreview).where(LinkPreview.url == url))
+        ).scalar_one()
+        row.fetched_at = datetime.now(timezone.utc) - timedelta(minutes=10)
+        await session.commit()
+
+    updated_html = _OG_HTML.replace(b"Example Article", b"Updated Article")
+    monkeypatch.setattr(
+        "app.services.link_preview_service.httpx.AsyncClient",
+        _fake_client_factory(html=updated_html, call_log=calls),
+    )
+
+    resp2 = await client.post(
+        f"/api/webhooks/incoming/{webhook['token']}", json={"content": f"again: {url}"}
+    )
+    assert resp2.status_code == 204
+    await asyncio.gather(*captured_tasks)
+
+    assert len(calls) == 2  # the expired cache forced a second real fetch
+
+    # Cached by URL, not by message (see link_preview_service.py) -- the
+    # row was refreshed in place, so *both* messages referencing this URL
+    # now show the new title on a history reload, not one each.
+    history = (await client.get(f"/api/rooms/{room['id']}/messages")).json()
+    assert len(history) == 2
+    assert all(m["link_preview"]["title"] == "Updated Article" for m in history)

backend/tests/test_profile.py

@@ -102,6 +102,62 @@ async def test_theme_custom_rejected_on_generic_profile_update(client, db_sessio
     assert resp.status_code == 422
 
 
+async def test_update_text_scale_persists(client, db_session):
+    await register_and_login(client, db_session, username=_unique("alice"))
+
+    resp = await client.patch("/api/auth/me", json={"text_scale": "large"})
+    assert resp.status_code == 200, resp.text
+    assert resp.json()["text_scale"] == "large"
+
+    me = await client.get("/api/auth/me")
+    assert me.json()["text_scale"] == "large"
+
+
+async def test_invalid_text_scale_rejected(client, db_session):
+    await register_and_login(client, db_session, username=_unique("alice"))
+
+    resp = await client.patch("/api/auth/me", json={"text_scale": "huge"})
+    assert resp.status_code == 422
+
+
+async def test_updating_text_scale_does_not_clobber_theme(client, db_session):
+    await register_and_login(client, db_session, username=_unique("alice"))
+    await client.patch("/api/auth/me", json={"theme": "sunset"})
+
+    resp = await client.patch("/api/auth/me", json={"text_scale": "xlarge"})
+    assert resp.status_code == 200
+    assert resp.json()["theme"] == "sunset"
+    assert resp.json()["text_scale"] == "xlarge"
+
+
+async def test_update_emoji_scale_persists(client, db_session):
+    await register_and_login(client, db_session, username=_unique("alice"))
+
+    resp = await client.patch("/api/auth/me", json={"emoji_scale": "xlarge"})
+    assert resp.status_code == 200, resp.text
+    assert resp.json()["emoji_scale"] == "xlarge"
+
+    me = await client.get("/api/auth/me")
+    assert me.json()["emoji_scale"] == "xlarge"
+
+
+async def test_invalid_emoji_scale_rejected(client, db_session):
+    await register_and_login(client, db_session, username=_unique("alice"))
+
+    resp = await client.patch("/api/auth/me", json={"emoji_scale": "huge"})
+    assert resp.status_code == 422
+
+
+async def test_updating_emoji_scale_does_not_clobber_text_scale(client, db_session):
+    await register_and_login(client, db_session, username=_unique("alice"))
+    await client.patch("/api/auth/me", json={"text_scale": "large"})
+
+    resp = await client.patch("/api/auth/me", json={"emoji_scale": "small"})
+    assert resp.status_code == 200
+    assert resp.json()["text_scale"] == "large"
+    assert resp.json()["emoji_scale"] == "small"
+
+
 async def test_avatar_upload_succeeds_and_persists(client, db_session):
     await register_and_login(client, db_session, username=_unique("alice"))
 

backend/tests/test_rooms.py

@@ -484,6 +484,27 @@ async def test_add_member_directly(client, db_session, monkeypatch):
     assert "added" in calls[0]["subject"].lower()
 
 
+async def test_add_member_posts_welcome_message(client, db_session, monkeypatch):
+    # #74: posted as the auto-provisioned "system" account, not the admin
+    # who added them -- mirrors test_add_member_directly's setup.
+    _fake_send_email(monkeypatch)
+    await register_and_login(client, db_session, username="alice")
+    room_id = (await client.post("/api/rooms", json={"name": "general"})).json()["id"]
+
+    await client.post("/api/auth/logout")
+    bob = await register_and_login(client, db_session, username="bob")
+
+    await client.post("/api/auth/logout")
+    await login_as(client, "alice")
+    resp = await client.post(f"/api/rooms/{room_id}/members", json={"user_id": bob["id"]})
+    assert resp.status_code == 201, resp.text
+
+    history = (await client.get(f"/api/rooms/{room_id}/messages")).json()
+    welcome_messages = [m for m in history if m["username"] == "system"]
+    assert len(welcome_messages) == 1
+    assert welcome_messages[0]["content"] == "Welcome to #general, bob!"
+
+
 async def test_add_member_requires_admin_role(client, db_session, monkeypatch):
     _fake_send_email(monkeypatch)
     await register_and_login(client, db_session, username="alice")

backend/tests/test_system_user.py

@@ -0,0 +1,13 @@
+from app.services.system_user_service import SYSTEM_USERNAME, get_or_create_system_user
+
+
+async def test_get_or_create_system_user_creates_bot_account(db_session):
+    user = await get_or_create_system_user(db_session)
+    assert user.username == SYSTEM_USERNAME
+    assert user.is_bot is True
+
+
+async def test_get_or_create_system_user_is_idempotent(db_session):
+    first = await get_or_create_system_user(db_session)
+    second = await get_or_create_system_user(db_session)
+    assert first.id == second.id

backend/tests/test_ws_chat.py

@@ -109,6 +109,24 @@ def test_history_tiebreaks_identical_timestamps_by_id(ws_client):
     assert [m["content"] for m in history] == [m["content"] for m in expected_order]
 
 
+def test_ws_sends_heartbeat_ping(ws_client, monkeypatch):
+    # #76: with nothing else sent on the connection, a "ping" must arrive
+    # proactively once the idle timeout elapses -- shrunk here so the test
+    # doesn't actually wait the real 30s. Replying "pong" (the frontend's
+    # own behavior) must be accepted silently and leave the connection
+    # fully usable afterward, not treated as an unknown message type.
+    monkeypatch.setattr("app.ws.chat.WS_PING_INTERVAL_SECONDS", 0.05)
+    username = _unique("alice")
+    _register(ws_client, username=username)
+    room = ws_client.post("/api/rooms", json={"name": _unique("general")}).json()
+
+    with ws_client.websocket_connect("/ws/chat") as ws:
+        assert ws.receive_json() == {"type": "ping"}
+        ws.send_json({"type": "pong"})
+        ws.send_json({"type": "join", "room_id": room["id"]})
+        assert ws.receive_json() == {"type": "joined", "room_id": room["id"]}
+
+
 def test_ws_message_without_join_errors(ws_client):
     _register(ws_client, username=_unique("alice"))
     room = ws_client.post("/api/rooms", json={"name": _unique("general")}).json()

deploy/upgrade.sh

@@ -16,9 +16,18 @@ BACKEND_DIR="${REPO_DIR}/backend"
 FRONTEND_DIR="${REPO_DIR}/frontend"
 ENV_FILE="/etc/ds-chat/env"
 
-echo "==> Pulling latest code"
+echo "==> Fetching latest release"
 cd "$REPO_DIR"
-git pull --ff-only
+git fetch --tags --force
+LATEST_TAG="$(git tag --sort=-creatordate | head -n1)"
+if [[ -z "$LATEST_TAG" ]]; then
+  echo "No tags found -- nothing to deploy" >&2
+  exit 1
+fi
+echo "Deploying $LATEST_TAG"
+# Detached HEAD, not a branch checkout -- this directory only ever runs a
+# tagged release, never whatever the default branch's tip happens to be.
+git checkout --quiet --detach "$LATEST_TAG"
 
 echo "==> Installing backend dependencies"
 cd "$BACKEND_DIR"

frontend/.env.example

@@ -0,0 +1,7 @@
+# Shown as the "Source code" link in the app's About screen -- required
+# for AGPL-3.0 section 13 compliance once you deploy this (a link so users
+# interacting with the app over the network can get the actual source,
+# including any modifications you've made). The default below points at
+# the upstream project -- fine if you're running it unmodified, but if
+# you've forked or patched the code, point this at *your* copy instead.
+VITE_SOURCE_URL=https://github.com/ds-ksmith/DS-Chat

frontend/README.md

@@ -112,7 +112,7 @@ src/
     LoginPage.tsx, SignupPage.tsx, ForgotPasswordPage.tsx, ResetPasswordPage.tsx
     ChatShellPage.tsx, AdminPage.tsx, HelpPage.tsx
 
-  styles/tokens.css          design tokens (DarkSingularity theme: colors, spacing, etc.)
+  styles/tokens.css          design tokens (default theme: colors, spacing, etc.)
   sw.ts                      custom service worker (injectManifest): app-shell
                                precache + NetworkFirst runtime caching, push/notificationclick
                                handlers, SKIP_WAITING messaging for the update-prompt flow

frontend/package.json

@@ -1,14 +1,15 @@
 {
   "name": "frontend",
   "private": true,
-  "version": "1.1.0",
+  "version": "2026.9.28",
   "license": "AGPL-3.0-or-later",
   "type": "module",
   "scripts": {
     "dev": "vite",
     "build": "tsc -b && vite build",
     "lint": "oxlint",
-    "preview": "vite preview"
+    "preview": "vite preview",
+    "test": "vitest run"
   },
   "dependencies": {
     "markdown-to-jsx": "^9.10.2",
@@ -18,13 +19,16 @@
     "vite-plugin-pwa": "^1.3.0"
   },
   "devDependencies": {
+    "@testing-library/react": "^16.3.3",
     "@types/node": "^24.13.3",
     "@types/react": "^19.2.17",
     "@types/react-dom": "^19.2.3",
     "@vitejs/plugin-react": "^6.0.4",
+    "jsdom": "^30.1.1",
     "oxlint": "^1.75.0",
     "typescript": "~6.0.2",
     "vite": "^8.2.0",
+    "vitest": "^5.0.2",
     "workbox-cacheable-response": "^7.4.1",
     "workbox-expiration": "^7.4.1",
     "workbox-precaching": "^7.4.1",

frontend/src/api/auth.ts

@@ -1,5 +1,5 @@
 import { apiFetch, ApiError, NetworkError } from './client'
-import type { User, UserSession } from '../types'
+import type { EmojiScale, TextScale, User, UserSession } from '../types'
 
 // No register() here: this is an invite-only site. Accounts are created by
 // an operator via the backend CLI (`python -m app.cli create-user`), not
@@ -50,6 +50,23 @@ export function updateTheme(theme: 'dark' | 'light' | 'midnight' | 'sunset'): Pr
   })
 }
 
+// #71: its own call, same reasoning as updateTheme above -- the backend
+// only applies fields actually present in the request body, so this can't
+// clobber theme (or vice versa).
+export function updateTextScale(textScale: TextScale): Promise<User> {
+  return apiFetch<User>('/api/auth/me', {
+    method: 'PATCH',
+    body: JSON.stringify({ text_scale: textScale }),
+  })
+}
+
+export function updateEmojiScale(emojiScale: EmojiScale): Promise<User> {
+  return apiFetch<User>('/api/auth/me', {
+    method: 'PATCH',
+    body: JSON.stringify({ emoji_scale: emojiScale }),
+  })
+}
+
 export function removeAvatar(): Promise<User> {
   return apiFetch<User>('/api/auth/me/avatar', { method: 'DELETE' })
 }

frontend/src/components/AboutModal.tsx

@@ -10,8 +10,10 @@ interface AboutModalProps {
 // users remotely through a computer network, you should also make sure
 // that it provides a way for users to get its source... its interface
 // could display a 'Source' link" -- this modal is that link, not just a
-// courtesy credits screen.
+// courtesy credits screen. Deliberately not a hardcoded URL: whoever
-const SOURCE_URL = 'https://git.darksingularity.org/DarkSingularity/ds-chat'
+// deploys this needs to point it at *their* copy of the repo (including
+// any modifications), not the upstream project -- see frontend/.env.example.
+const SOURCE_URL = import.meta.env.VITE_SOURCE_URL as string | undefined
 
 export function AboutModal({ onClose }: AboutModalProps) {
   return (
@@ -39,11 +41,13 @@ export function AboutModal({ onClose }: AboutModalProps) {
           </a>
           .
         </p>
+        {SOURCE_URL && (
           <p className="about-modal-line">
             <a href={SOURCE_URL} target="_blank" rel="noopener noreferrer">
               Source code
             </a>
           </p>
+        )}
 
         <div className="modal-actions" style={{ marginTop: '1rem' }}>
           <button type="button" className="btn-secondary" onClick={onClose}>

frontend/src/components/CustomEmojiManageModal.tsx

@@ -0,0 +1,99 @@
+import { useState } from 'react'
+import { deleteCustomEmoji } from '../api/customEmoji'
+import { useAuth } from '../context/AuthContext'
+import { useCustomEmoji } from '../context/CustomEmojiContext'
+import type { CustomEmoji } from '../types'
+import { CustomEmojiUploadModal } from './CustomEmojiUploadModal'
+import { EmojiGlyph } from './MessageContent'
+import './Modal.css'
+
+interface CustomEmojiManageModalProps {
+  onClose: () => void
+}
+
+// Moved out of the reaction/composer emoji picker -- that grid packs items
+// 9-to-a-row with a delete "x" overlapping the glyph itself, which on a
+// touch screen is far too easy to hit by accident while just trying to
+// react. A dedicated list with a normal-sized "Delete" button (plus the
+// same confirm() every other destructive action in this app uses) needs a
+// deliberate tap to actually delete something.
+export function CustomEmojiManageModal({ onClose }: CustomEmojiManageModalProps) {
+  const { user } = useAuth()
+  const { list, refresh } = useCustomEmoji()
+  const [uploadOpen, setUploadOpen] = useState(false)
+  const [deletingId, setDeletingId] = useState<string | null>(null)
+
+  async function handleDelete(emoji: CustomEmoji) {
+    if (!confirm(`Delete :${emoji.shortcode}:? This can't be undone.`)) return
+    setDeletingId(emoji.id)
+    try {
+      await deleteCustomEmoji(emoji.id)
+      await refresh()
+    } finally {
+      setDeletingId(null)
+    }
+  }
+
+  return (
+    <div className="modal-scrim" onClick={onClose}>
+      <div className="modal" onClick={(e) => e.stopPropagation()}>
+        <div className="modal-header">
+          <h2>Custom emoji</h2>
+          <button type="button" className="modal-close" onClick={onClose} aria-label="Close">
+            &times;
+          </button>
+        </div>
+
+        <div className="modal-field-label">Site emoji</div>
+        {list.length === 0 ? (
+          <p className="modal-empty">No custom emoji yet.</p>
+        ) : (
+          list.map((emoji) => {
+            const canDelete = user?.id === emoji.uploaded_by || user?.is_site_admin
+            return (
+              <div key={emoji.id} className="modal-list-row">
+                <div className="modal-list-row-body">
+                  <div className="modal-list-row-title">
+                    <EmojiGlyph value={`:${emoji.shortcode}:`} /> :{emoji.shortcode}:
+                  </div>
+                  <div className="modal-list-row-sub">
+                    Added {new Date(emoji.created_at).toLocaleDateString()}
+                  </div>
+                </div>
+                {canDelete && (
+                  <button
+                    type="button"
+                    className="modal-list-row-action"
+                    disabled={deletingId === emoji.id}
+                    onClick={() => handleDelete(emoji)}
+                  >
+                    {deletingId === emoji.id ? 'Deleting…' : 'Delete'}
+                  </button>
+                )}
+              </div>
+            )
+          })
+        )}
+
+        <div className="modal-actions">
+          <button type="button" className="btn-secondary" onClick={onClose}>
+            Close
+          </button>
+          <button type="button" className="btn-primary" onClick={() => setUploadOpen(true)}>
+            Add emoji
+          </button>
+        </div>
+      </div>
+
+      {uploadOpen && (
+        <CustomEmojiUploadModal
+          onClose={() => setUploadOpen(false)}
+          onUploaded={() => {
+            refresh()
+            setUploadOpen(false)
+          }}
+        />
+      )}
+    </div>
+  )
+}

frontend/src/components/EmojiPicker.css

@@ -69,26 +69,6 @@
   padding: 4px 4px 2px;
 }
 
-.emoji-picker-category-label-row {
-  display: flex;
-  align-items: center;
-  justify-content: space-between;
-}
-
-.emoji-picker-add-custom {
-  background: transparent;
-  border: none;
-  color: var(--ds-accent);
-  font-size: 0.7rem;
-  font-weight: 600;
-  cursor: pointer;
-  padding: 2px 4px;
-}
-
-.emoji-picker-add-custom:hover {
-  text-decoration: underline;
-}
-
 .emoji-picker-grid {
   display: grid;
   grid-template-columns: repeat(9, 1fr);
@@ -109,32 +89,6 @@
   background: var(--ds-surface-2);
 }
 
-.emoji-picker-item-custom {
-  position: relative;
-}
-
-.emoji-picker-item-remove {
-  position: absolute;
-  top: -2px;
-  right: -2px;
-  display: flex;
-  align-items: center;
-  justify-content: center;
-  width: 14px;
-  height: 14px;
-  border-radius: 50%;
-  background: var(--ds-danger);
-  color: white;
-  font-size: 0.65rem;
-  line-height: 1;
-  opacity: 0;
-  cursor: pointer;
-}
-
-.emoji-picker-item-custom:hover .emoji-picker-item-remove {
-  opacity: 1;
-}
-
 /* The picker is positioned absolutely relative to its trigger button, which
    can sit close enough to a narrow viewport's edge that the full 320px
    width runs off-screen (e.g. the composer's emoji trigger, near the left

frontend/src/components/EmojiPicker.tsx

@@ -1,12 +1,9 @@
-import { useMemo, useState, type MouseEvent } from 'react'
+import { useMemo, useState } from 'react'
-import { deleteCustomEmoji } from '../api/customEmoji'
-import { useAuth } from '../context/AuthContext'
 import { useCustomEmoji } from '../context/CustomEmojiContext'
 import { useEscapeKey } from '../hooks/useEscapeKey'
 import { ALL_EMOJI, EMOJI_CATEGORIES } from '../lib/emoji'
 import { EMOJI_NAMES } from '../lib/emojiNames'
 import { getRecentEmoji, recordEmojiUsed } from '../lib/recentEmoji'
-import { CustomEmojiUploadModal } from './CustomEmojiUploadModal'
 import { EmojiGlyph } from './MessageContent'
 import './EmojiPicker.css'
 
@@ -55,11 +52,8 @@ function searchEmoji(query: string, customShortcodes: string[]): string[] {
 
 export function EmojiPicker({ onPick, onClose, placement = 'below', align = 'left' }: EmojiPickerProps) {
   useEscapeKey(onClose)
-  const { user } = useAuth()
+  const { list: customEmoji } = useCustomEmoji()
-  const { list: customEmoji, refresh: refreshCustomEmoji } = useCustomEmoji()
   const [query, setQuery] = useState('')
-  const [uploadOpen, setUploadOpen] = useState(false)
-  const [deletingId, setDeletingId] = useState<string | null>(null)
   const customShortcodes = useMemo(() => customEmoji.map((e) => e.shortcode), [customEmoji])
   const searchResults = useMemo(() => searchEmoji(query, customShortcodes), [query, customShortcodes])
   const searching = query.trim().length > 0
@@ -74,18 +68,6 @@ export function EmojiPicker({ onPick, onClose, placement = 'below', align = 'lef
     onPick(emoji)
   }
 
-  async function handleDeleteCustomEmoji(e: MouseEvent, emojiId: string) {
-    // Delete, not pick -- must never bubble to the button's own onClick.
-    e.stopPropagation()
-    setDeletingId(emojiId)
-    try {
-      await deleteCustomEmoji(emojiId)
-      await refreshCustomEmoji()
-    } finally {
-      setDeletingId(null)
-    }
-  }
-
   return (
     <>
       <div className="emoji-picker-scrim" onClick={onClose} />
@@ -122,48 +104,25 @@ export function EmojiPicker({ onPick, onClose, placement = 'below', align = 'lef
           )
         ) : (
           <>
-            <div className="emoji-picker-category">
-              <div className="emoji-picker-category-label-row">
-                <div className="emoji-picker-category-label">Custom</div>
-                <button
-                  type="button"
-                  className="emoji-picker-add-custom"
-                  onClick={() => setUploadOpen(true)}
-                >
-                  + Add
-                </button>
-              </div>
             {customEmoji.length > 0 && (
+              <div className="emoji-picker-category">
+                <div className="emoji-picker-category-label">Custom</div>
                 <div className="emoji-picker-grid">
-                  {customEmoji.map((e) => {
+                  {customEmoji.map((e) => (
-                    const canDelete = user?.id === e.uploaded_by || user?.is_site_admin
-                    return (
                     <button
                       key={e.id}
                       type="button"
                       role="menuitem"
-                        className="emoji-picker-item emoji-picker-item-custom"
+                      className="emoji-picker-item"
                       title={`:${e.shortcode}:`}
                       onClick={() => pick(`:${e.shortcode}:`)}
                     >
                       <EmojiGlyph value={`:${e.shortcode}:`} />
-                        {canDelete && (
-                          <span
-                            role="button"
-                            aria-label={`Remove :${e.shortcode}:`}
-                            className="emoji-picker-item-remove"
-                            onClick={(ev) => handleDeleteCustomEmoji(ev, e.id)}
-                            style={deletingId === e.id ? { opacity: 0.5, pointerEvents: 'none' } : undefined}
-                          >
-                            ×
-                          </span>
-                        )}
                     </button>
-                    )
+                  ))}
-                  })}
+                </div>
               </div>
             )}
-            </div>
             {recent.length > 0 && (
               <div className="emoji-picker-category">
                 <div className="emoji-picker-category-label">Recently used</div>
@@ -205,15 +164,6 @@ export function EmojiPicker({ onPick, onClose, placement = 'below', align = 'lef
           </>
         )}
       </div>
-      {uploadOpen && (
-        <CustomEmojiUploadModal
-          onClose={() => setUploadOpen(false)}
-          onUploaded={() => {
-            refreshCustomEmoji()
-            setUploadOpen(false)
-          }}
-        />
-      )}
     </>
   )
 }

frontend/src/components/MessageContent.css

@@ -4,10 +4,25 @@
    own font-size and this just tracks it. Kept in this file (imported
    directly by MessageContent.tsx) rather than MessageList.css so it's
    loaded wherever MessageContent renders -- FilePreviewModal and HelpPage
-   included, not just the message list. */
+   included, not just the message list.
+
+   #71: also multiplied by --emoji-scale, the manual "make emoji bigger"
+   preference -- but that variable is only ever set on MessageContent's own
+   wrapper div (inline style, scoped to that element and its descendants),
+   never at :root, so var(..., 1) correctly falls back to a no-op multiplier
+   everywhere else this class is reused (the picker's grid, reaction pills)
+   instead of also inflating those and breaking their fixed-size layout. */
 .message-custom-emoji {
-  height: 1.2em;
+  height: calc(1.2em * var(--emoji-scale, 1));
-  width: 1.2em;
+  width: calc(1.2em * var(--emoji-scale, 1));
   object-fit: contain;
   vertical-align: -0.25em;
 }
+
+/* #71: a raw unicode emoji wrapped by wrapEmojiGlyphs -- same --emoji-scale
+   multiplier as the custom-emoji image above, so "make emoji bigger"
+   applies uniformly regardless of which kind of emoji it is. */
+.inline-emoji {
+  display: inline-block;
+  font-size: calc(1em * var(--emoji-scale, 1));
+}

frontend/src/components/MessageContent.tsx

@@ -1,7 +1,8 @@
 import Markdown from 'markdown-to-jsx'
-import type { ReactNode } from 'react'
+import type { CSSProperties, ReactNode } from 'react'
 import { Link } from 'react-router-dom'
 import { getCustomEmojiUrl } from '../api/customEmoji'
+import { useAuth } from '../context/AuthContext'
 import { useCustomEmoji } from '../context/CustomEmojiContext'
 import { EMOJI_SHORTCODES } from '../lib/emojiShortcodes'
 import './MessageContent.css'
@@ -85,6 +86,14 @@ function MarkdownLink({ href, children }: MarkdownLinkProps) {
       />
     )
   }
+  // #71: a raw unicode emoji has no element of its own to size independently
+  // of the surrounding text -- it's just characters in a string. Wrapping
+  // each one individually (see wrapEmojiGlyphs below) gives it one, purely
+  // so the emoji-size preference can scale it via CSS the same way it
+  // already scales a custom emoji's <img>.
+  if (href === 'glyph:') {
+    return <span className="inline-emoji">{children}</span>
+  }
   return (
     <a href={href} target="_blank" rel="noopener noreferrer">
       {children}
@@ -247,7 +256,12 @@ export function EmojiGlyph({ value }: EmojiGlyphProps) {
       />
     )
   }
-  return <>{value}</>
+  // Wrapped the same way wrapEmojiGlyphs wraps a raw emoji in message text
+  // (see .inline-emoji), so a --emoji-scale set on an ancestor (the
+  // reaction pill's own span in MessageList.tsx) scales this the same way
+  // it scales the .message-custom-emoji img above -- and falls back to a
+  // no-op 1x everywhere else (the picker) with no --emoji-scale set at all.
+  return <span className="inline-emoji">{value}</span>
 }
 
 const MENTION_PATTERN = /@([a-zA-Z0-9_.-]+)/g
@@ -370,14 +384,66 @@ export function preprocessMarkdown(text: string): { text: string; headingIds: Ma
   return extractHeadingIds(convertSubSuperscript(text))
 }
 
+// #71: gives every individual unicode emoji its own element (see
+// MarkdownLink's `glyph:` branch) purely so the emoji-size preference can
+// scale it independently of the surrounding text -- a raw emoji is just
+// characters in a string otherwise, with nothing CSS can address on its
+// own. Runs after convertShortcodes so a built-in `:name:` that just
+// became a glyph is wrapped too ("all emoji", not just ones typed as
+// literal unicode); same fence/code-span skip convention as every other
+// converter here.
+const EMOJI_GLYPH_PATTERN = /\p{Extended_Pictographic}(?:\p{Emoji_Modifier}|\u200D\p{Extended_Pictographic}|\uFE0F)*/gu
+
+function wrapEmojiGlyphs(text: string): string {
+  const lines = text.split('\n')
+  let inFence = false
+  return lines
+    .map((line) => {
+      if (/^\s*```/.test(line)) {
+        inFence = !inFence
+        return line
+      }
+      if (inFence) return line
+      return line
+        .split(/(`+[^`]*`+)/g)
+        .map((part, i) => (i % 2 === 0 ? part.replace(EMOJI_GLYPH_PATTERN, (match) => `[${match}](glyph:)`) : part))
+        .join('')
+    })
+    .join('\n')
+}
+
+// Exported so MessageList's reaction pills can apply the same viewer
+// preference to their own EmojiGlyph -- reactions render outside the
+// markdown pipeline entirely (see EmojiGlyph's own comment above), so they
+// need this looked up independently rather than inheriting --emoji-scale
+// from this component's wrapper div.
+export const EMOJI_SCALE_MULTIPLIER: Record<string, number> = {
+  small: 0.8,
+  normal: 1,
+  large: 1.5,
+  xlarge: 2,
+}
+
 export function MessageContent({ content, memberUsernames, myRooms }: MessageContentProps) {
+  const { user } = useAuth()
   const { byShortcode } = useCustomEmoji()
+  const customShortcodes = new Set(byShortcode.keys())
   const withMentions = memberUsernames ? highlightMentions(content, memberUsernames) : content
   const withRoomRefs = myRooms ? highlightRoomReferences(withMentions, myRooms) : withMentions
-  const withCustomEmoji = convertCustomEmojiShortcodes(
+  const withCustomEmoji = convertCustomEmojiShortcodes(convertShortcodes(withRoomRefs), customShortcodes)
-    convertShortcodes(withRoomRefs),
+  const withEmojiGlyphs = wrapEmojiGlyphs(withCustomEmoji)
-    new Set(byShortcode.keys()),
+  const { text, headingIds } = preprocessMarkdown(withEmojiGlyphs)
+  // #71: scoped to this element (not a :root-level variable) so it only
+  // ever affects emoji rendered in message text -- not the same
+  // .message-custom-emoji/EmojiGlyph markup reused by the emoji picker's
+  // grid, where a bigger image would just break its fixed-size layout
+  // instead of doing anything useful. Reaction pills DO scale too, but via
+  // their own inline --emoji-scale in MessageList.tsx, not by inheriting
+  // this one -- a pill isn't a descendant of this wrapper div.
+  const emojiScale = EMOJI_SCALE_MULTIPLIER[user?.emoji_scale ?? 'normal']
+  return (
+    <div style={{ '--emoji-scale': emojiScale } as CSSProperties}>
+      <Markdown options={createMarkdownOptions(headingIds)}>{preserveLineBreaks(text)}</Markdown>
+    </div>
   )
-  const { text, headingIds } = preprocessMarkdown(withCustomEmoji)
-  return <Markdown options={createMarkdownOptions(headingIds)}>{preserveLineBreaks(text)}</Markdown>
 }

frontend/src/components/MessageList.css

@@ -53,8 +53,13 @@
 
 .message-image {
   display: block;
-  max-width: min(320px, 100%);
+  /* #71: rem, not px -- scales with the text-size setting (see
-  max-height: 240px;
+     lib/theme.ts's applyTextScale), same as every other size in this app.
+     min(...) still caps against the viewport in absolute px, since a
+     percentage-of-viewport constraint isn't something a root font-size
+     change should affect. */
+  max-width: min(20rem, 100%);
+  max-height: 15rem;
   object-fit: contain;
   border-radius: var(--radius);
   border: 1px solid var(--ds-border);
@@ -65,14 +70,14 @@
 .message-video-wrap {
   position: relative;
   display: inline-block;
-  max-width: min(320px, 100%);
+  max-width: min(20rem, 100%);
   margin-bottom: 4px;
 }
 
 .message-video {
   display: block;
   width: 100%;
-  max-height: 240px;
+  max-height: 15rem;
   border-radius: var(--radius);
   border: 1px solid var(--ds-border);
   background: var(--ds-void);
@@ -111,7 +116,7 @@
   margin-bottom: 4px;
   color: var(--ds-text);
   text-decoration: none;
-  max-width: min(320px, 100%);
+  max-width: min(20rem, 100%);
 }
 
 .message-file-attachment:hover {

frontend/src/components/MessageList.tsx

@@ -1,3 +1,4 @@
+import type { CSSProperties } from 'react'
 import { useEffect, useMemo, useRef, useState } from 'react'
 import { getRoomFileUrl, getRoomImageUrl } from '../api/rooms'
 import { useAuth } from '../context/AuthContext'
@@ -8,7 +9,7 @@ import { EMOJI_PICKER_MAX_HEIGHT, EmojiPicker } from './EmojiPicker'
 import { FilePreviewModal, getPreviewKind } from './FilePreviewModal'
 import { ImageLightbox } from './ImageLightbox'
 import { LinkPreviewCard } from './LinkPreviewCard'
-import { EmojiGlyph, MessageContent } from './MessageContent'
+import { EMOJI_SCALE_MULTIPLIER, EmojiGlyph, MessageContent } from './MessageContent'
 import { UserAvatar } from './UserAvatar'
 import { VideoLightbox } from './VideoLightbox'
 import './MessageList.css'
@@ -68,6 +69,13 @@ function FileAttachmentCard({ file, roomId, onPreview }: FileAttachmentCardProps
 // trigger a download the moment the browser tries to fetch it).
 const PLAYABLE_VIDEO_CONTENT_TYPES = new Set(['video/mp4', 'video/webm', 'video/ogg'])
 
+// #73: Slack's own threshold for the same "still grouped, but it's been a
+// while" call -- past this gap a same-sender message starts a new group
+// (its own avatar/name/timestamp) even with nobody else posting in
+// between, so a message sent minutes later doesn't hide under a stale
+// timestamp from the start of the run.
+const GROUP_BREAK_MS = 5 * 60 * 1000
+
 interface VideoAttachmentProps {
   file: MessageFileInfo
   roomId: string
@@ -126,6 +134,10 @@ export function MessageList({
   onDelete,
 }: MessageListProps) {
   const { user } = useAuth()
+  // #71: same viewer preference MessageContent applies to in-text emoji,
+  // looked up separately here since a reaction pill isn't a descendant of
+  // that component's wrapper div (see EmojiGlyph's own comment).
+  const emojiScale = EMOJI_SCALE_MULTIPLIER[user?.emoji_scale ?? 'normal']
   const containerRef = useRef<HTMLDivElement>(null)
   const bottomRef = useRef<HTMLDivElement>(null)
   // Whether the view should be pinned to the latest message -- true right
@@ -205,8 +217,13 @@ export function MessageList({
         // Mattermost-style grouping: every message shows who sent it, but
         // consecutive messages from the same sender only repeat the
         // avatar/name/timestamp header on the first one in the run --
-        // applies uniformly, including to your own messages.
+        // applies uniformly, including to your own messages. Also breaks on
-        const isGroupStart = !prev || prev.user_id !== msg.user_id
+        // a long gap (see GROUP_BREAK_MS) so a message sent well after the
+        // rest of the run still gets its own visible timestamp.
+        const isGroupStart =
+          !prev ||
+          prev.user_id !== msg.user_id ||
+          new Date(msg.created_at).getTime() - new Date(prev.created_at).getTime() > GROUP_BREAK_MS
         const editing = editingId === msg.id
         const deleted = !!msg.deleted_at
 
@@ -302,7 +319,10 @@ export function MessageList({
                             title={r.user_ids.map(displayNameForUserId).join(', ')}
                             onClick={() => onReact(msg.id, r.emoji)}
                           >
-                            <span>
+                            {/* No .inline-emoji here -- EmojiGlyph's own fallback branch
+                                already applies it, and stacking it here too would double
+                                the font-size multiplication for a custom-emoji img. */}
+                            <span style={{ '--emoji-scale': emojiScale } as CSSProperties}>
                               <EmojiGlyph value={r.emoji} />
                             </span>
                             <span>{r.count}</span>

frontend/src/components/Modal.css

@@ -230,6 +230,45 @@
   color: var(--ds-muted);
 }
 
+.text-scale-options {
+  display: grid;
+  grid-template-columns: repeat(4, 1fr);
+  gap: var(--sp-2);
+  margin-bottom: var(--sp-4);
+}
+
+.text-scale-option {
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  gap: 4px;
+  background: var(--ds-surface-2);
+  border: 1px solid var(--ds-border);
+  border-radius: var(--radius);
+  padding: 10px 4px;
+  cursor: pointer;
+}
+
+.text-scale-option:hover {
+  border-color: var(--ds-accent);
+}
+
+.text-scale-option-selected {
+  border-color: var(--ds-accent);
+  box-shadow: 0 0 0 1px var(--ds-accent);
+}
+
+.text-scale-option-preview {
+  font-weight: 700;
+  color: var(--ds-text);
+  line-height: 1;
+}
+
+.text-scale-option-label {
+  font-size: 0.7rem;
+  color: var(--ds-muted);
+}
+
 .theme-swatch-preview-new {
   background: transparent;
   border-style: dashed;

frontend/src/components/ProfileModal.tsx

@@ -5,7 +5,9 @@ import {
   me,
   removeAvatar,
   revokeSession,
+  updateEmojiScale,
   updateProfile,
+  updateTextScale,
   updateTheme,
   uploadAvatar,
 } from '../api/auth'
@@ -20,8 +22,8 @@ import {
 import { getUserAvatarUrl } from '../api/users'
 import { useAuth } from '../context/AuthContext'
 import { hashIndex } from '../lib/avatar'
-import { applyTheme, DEFAULT_CUSTOM_COLORS } from '../lib/theme'
+import { applyTextScale, applyTheme, DEFAULT_CUSTOM_COLORS } from '../lib/theme'
-import type { CustomTheme, CustomThemeColors, UserSession } from '../types'
+import type { CustomTheme, CustomThemeColors, EmojiScale, TextScale, UserSession } from '../types'
 import { ThemeBuilderModal } from './ThemeBuilderModal'
 import { UserAvatar } from './UserAvatar'
 import './Modal.css'
@@ -33,6 +35,26 @@ const THEME_OPTIONS: { name: 'dark' | 'light' | 'midnight' | 'sunset'; label: st
   { name: 'sunset', label: 'Sunset' },
 ]
 
+// #71: the "Aa" preview scales with each option's own size, the standard
+// way a text-size picker shows what it does without a separate demo area.
+const TEXT_SCALE_OPTIONS: { name: TextScale; label: string; previewSize: string }[] = [
+  { name: 'small', label: 'Small', previewSize: '0.8rem' },
+  { name: 'normal', label: 'Normal', previewSize: '1rem' },
+  { name: 'large', label: 'Large', previewSize: '1.25rem' },
+  { name: 'xlarge', label: 'Extra large', previewSize: '1.5rem' },
+]
+
+// #71: independent of text size -- only scales emoji rendered in message
+// text (see MessageContent.tsx's --emoji-scale). The preview uses an
+// actual emoji so it demonstrates itself the same way the text-size
+// options do with "Aa".
+const EMOJI_SCALE_OPTIONS: { name: EmojiScale; label: string; previewSize: string }[] = [
+  { name: 'small', label: 'Small', previewSize: '1rem' },
+  { name: 'normal', label: 'Normal', previewSize: '1.25rem' },
+  { name: 'large', label: 'Large', previewSize: '1.6rem' },
+  { name: 'xlarge', label: 'Extra large', previewSize: '2rem' },
+]
+
 const CUSTOM_COLOR_FIELDS: { key: keyof Omit<CustomThemeColors, 'color_scheme'>; label: string }[] = [
   { key: 'void', label: 'Background' },
   { key: 'void_2', label: 'Sidebar background' },
@@ -60,6 +82,8 @@ export function ProfileModal({ onClose }: ProfileModalProps) {
   const [uploadingAvatar, setUploadingAvatar] = useState(false)
   const fileInputRef = useRef<HTMLInputElement>(null)
   const [themeError, setThemeError] = useState<string | null>(null)
+  const [textScaleError, setTextScaleError] = useState<string | null>(null)
+  const [emojiScaleError, setEmojiScaleError] = useState<string | null>(null)
 
   const [customThemes, setCustomThemes] = useState<CustomTheme[]>([])
   const [editingThemeId, setEditingThemeId] = useState<string | null>(null)
@@ -150,6 +174,32 @@ export function ProfileModal({ onClose }: ProfileModalProps) {
     }
   }
 
+  async function handleSelectTextScale(scale: TextScale) {
+    // Same instant-apply-then-persist pattern as handleSelectPreset above.
+    applyTextScale(scale)
+    setTextScaleError(null)
+    try {
+      const updated = await updateTextScale(scale)
+      updateUser(updated)
+    } catch (err) {
+      applyTextScale(user?.text_scale ?? null)
+      setTextScaleError(err instanceof ApiError ? err.message : String(err))
+    }
+  }
+
+  async function handleSelectEmojiScale(scale: EmojiScale) {
+    // No instant-apply DOM mutation here (unlike theme/text scale) -- it's
+    // just a value MessageContent reads from `user` on its next render, so
+    // persisting and updating that is the whole job.
+    setEmojiScaleError(null)
+    try {
+      const updated = await updateEmojiScale(scale)
+      updateUser(updated)
+    } catch (err) {
+      setEmojiScaleError(err instanceof ApiError ? err.message : String(err))
+    }
+  }
+
   async function handleActivateCustomTheme(theme: CustomTheme) {
     applyTheme('custom', theme.colors)
     setThemeError(null)
@@ -360,6 +410,48 @@ export function ProfileModal({ onClose }: ProfileModalProps) {
         </div>
         {themeError && <p className="modal-error">{themeError}</p>}
 
+        <div className="modal-field-label">Text size</div>
+        <div className="text-scale-options">
+          {TEXT_SCALE_OPTIONS.map((option) => (
+            <button
+              key={option.name}
+              type="button"
+              className={`text-scale-option${
+                (user.text_scale ?? 'normal') === option.name ? ' text-scale-option-selected' : ''
+              }`}
+              onClick={() => handleSelectTextScale(option.name)}
+              aria-pressed={(user.text_scale ?? 'normal') === option.name}
+            >
+              <span className="text-scale-option-preview" style={{ fontSize: option.previewSize }}>
+                Aa
+              </span>
+              <span className="text-scale-option-label">{option.label}</span>
+            </button>
+          ))}
+        </div>
+        {textScaleError && <p className="modal-error">{textScaleError}</p>}
+
+        <div className="modal-field-label">Emoji size</div>
+        <div className="text-scale-options">
+          {EMOJI_SCALE_OPTIONS.map((option) => (
+            <button
+              key={option.name}
+              type="button"
+              className={`text-scale-option${
+                (user.emoji_scale ?? 'normal') === option.name ? ' text-scale-option-selected' : ''
+              }`}
+              onClick={() => handleSelectEmojiScale(option.name)}
+              aria-pressed={(user.emoji_scale ?? 'normal') === option.name}
+            >
+              <span className="text-scale-option-preview" style={{ fontSize: option.previewSize }}>
+                🎉
+              </span>
+              <span className="text-scale-option-label">{option.label}</span>
+            </button>
+          ))}
+        </div>
+        {emojiScaleError && <p className="modal-error">{emojiScaleError}</p>}
+
         <div className="modal-field-label">My custom themes</div>
         <div className="theme-swatch-grid">
           {customThemes.map((theme) => {

frontend/src/components/TopBar.tsx

@@ -13,6 +13,7 @@ import {
 } from '../lib/desktopBridge'
 import { getPushSubscriptionStatus, isPushSupported, subscribeToPush, unsubscribeFromPush } from '../lib/push'
 import { AboutModal } from './AboutModal'
+import { CustomEmojiManageModal } from './CustomEmojiManageModal'
 import { ProfileModal } from './ProfileModal'
 import { UserAvatar } from './UserAvatar'
 import './TopBar.css'
@@ -28,6 +29,7 @@ export function TopBar() {
   const navigate = useNavigate()
   const [menuOpen, setMenuOpen] = useState(false)
   const [profileModalOpen, setProfileModalOpen] = useState(false)
+  const [customEmojiModalOpen, setCustomEmojiModalOpen] = useState(false)
   const [aboutModalOpen, setAboutModalOpen] = useState(false)
   const [pushSubscribed, setPushSubscribed] = useState(false)
   const [pushBusy, setPushBusy] = useState(false)
@@ -124,6 +126,16 @@ export function TopBar() {
               >
                 Profile settings
               </button>
+              <button
+                type="button"
+                role="menuitem"
+                onClick={() => {
+                  setMenuOpen(false)
+                  setCustomEmojiModalOpen(true)
+                }}
+              >
+                Custom emoji
+              </button>
               <button
                 type="button"
                 role="menuitem"
@@ -190,6 +202,7 @@ export function TopBar() {
         )}
       </div>
       {profileModalOpen && <ProfileModal onClose={() => setProfileModalOpen(false)} />}
+      {customEmojiModalOpen && <CustomEmojiManageModal onClose={() => setCustomEmojiModalOpen(false)} />}
       {aboutModalOpen && <AboutModal onClose={() => setAboutModalOpen(false)} />}
     </header>
   )

frontend/src/context/AuthContext.tsx

@@ -3,7 +3,7 @@ import * as authApi from '../api/auth'
 import { ApiError, NetworkError } from '../api/client'
 import { clearLastUser, loadLastUser, saveLastUser } from '../lib/lastUser'
 import { unsubscribeFromPush } from '../lib/push'
-import { applyTheme } from '../lib/theme'
+import { applyTextScale, applyTheme } from '../lib/theme'
 import type { User } from '../types'
 
 interface AuthContextValue {
@@ -26,6 +26,10 @@ export function AuthProvider({ children }: { children: ReactNode }) {
     applyTheme(user?.theme ?? null, user?.active_custom_theme?.colors ?? null)
   }, [user?.theme, user?.active_custom_theme])
 
+  useEffect(() => {
+    applyTextScale(user?.text_scale ?? null)
+  }, [user?.text_scale])
+
   useEffect(() => {
     authApi
       .me()

frontend/src/hooks/useDesktopRoomBridge.test.ts

@@ -0,0 +1,68 @@
+import { renderHook } from '@testing-library/react'
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
+import * as desktopBridge from '../lib/desktopBridge'
+import { useDesktopRoomBridge } from './useDesktopRoomBridge'
+
+describe('useDesktopRoomBridge', () => {
+  beforeEach(() => {
+    vi.spyOn(desktopBridge, 'notifyRoomOpened')
+    vi.spyOn(desktopBridge, 'notifyUnreadCount')
+  })
+
+  afterEach(() => {
+    vi.restoreAllMocks()
+  })
+
+  it('calls roomOpened once when entering a room', () => {
+    renderHook(({ roomId, count }) => useDesktopRoomBridge(roomId, count), {
+      initialProps: { roomId: 'room-1', count: 0 },
+    })
+    expect(desktopBridge.notifyRoomOpened).toHaveBeenCalledTimes(1)
+  })
+
+  it('does not call roomOpened again on a rerender with the same roomId', () => {
+    const { rerender } = renderHook(({ roomId, count }) => useDesktopRoomBridge(roomId, count), {
+      initialProps: { roomId: 'room-1', count: 0 },
+    })
+    expect(desktopBridge.notifyRoomOpened).toHaveBeenCalledTimes(1)
+
+    // Unrelated rerender -- unreadRoomCount changed, roomId did not.
+    rerender({ roomId: 'room-1', count: 3 })
+    rerender({ roomId: 'room-1', count: 3 })
+    expect(desktopBridge.notifyRoomOpened).toHaveBeenCalledTimes(1)
+  })
+
+  it('calls roomOpened again on a genuine transition to a different room', () => {
+    const { rerender } = renderHook(({ roomId, count }) => useDesktopRoomBridge(roomId, count), {
+      initialProps: { roomId: 'room-1', count: 0 },
+    })
+    rerender({ roomId: 'room-2', count: 0 })
+    expect(desktopBridge.notifyRoomOpened).toHaveBeenCalledTimes(2)
+  })
+
+  it('does not call roomOpened when there is no room id', () => {
+    renderHook(({ roomId, count }) => useDesktopRoomBridge(roomId, count), {
+      initialProps: { roomId: undefined as string | undefined, count: 0 },
+    })
+    expect(desktopBridge.notifyRoomOpened).not.toHaveBeenCalled()
+  })
+
+  it('sends the unread count whenever it changes, independent of roomOpened', () => {
+    const { rerender } = renderHook(({ roomId, count }) => useDesktopRoomBridge(roomId, count), {
+      initialProps: { roomId: undefined as string | undefined, count: 0 },
+    })
+    expect(desktopBridge.notifyUnreadCount).toHaveBeenCalledWith(0)
+
+    rerender({ roomId: undefined, count: 2 })
+    expect(desktopBridge.notifyUnreadCount).toHaveBeenCalledWith(2)
+    expect(desktopBridge.notifyRoomOpened).not.toHaveBeenCalled()
+  })
+})
+
+describe('desktop bridge functions with no window.dsDesktop', () => {
+  it('remain optional -- calling them does not throw when the bridge is absent', () => {
+    expect(window.dsDesktop).toBeUndefined()
+    expect(() => desktopBridge.notifyRoomOpened()).not.toThrow()
+    expect(() => desktopBridge.notifyUnreadCount(5)).not.toThrow()
+  })
+})

frontend/src/hooks/useDesktopRoomBridge.ts

@@ -0,0 +1,26 @@
+import { useEffect } from 'react'
+import { notifyRoomOpened, notifyUnreadCount } from '../lib/desktopBridge'
+
+// Tells the Electron desktop wrapper (a) when the user has genuinely
+// entered a room, so it can clear that room's tray indicator, and (b) the
+// current unread-room count, so the tray badge stays accurate. Both are
+// no-ops in every browser build (see desktopBridge.ts).
+//
+// roomOpened must fire only on a genuine transition to a real room id --
+// never merely because the window was shown, focused, restored, or opened
+// from the tray, and never when no room is open at all. Depending on just
+// the id (a primitive, not the room object) is what makes this safe to
+// call from a component that rerenders often: an unrelated rerender that
+// leaves activeRoomId unchanged doesn't rerun the effect, so there's no
+// separate de-duplication to get wrong here -- it falls out of React's own
+// effect-dependency comparison.
+export function useDesktopRoomBridge(activeRoomId: string | undefined, unreadRoomCount: number): void {
+  useEffect(() => {
+    if (!activeRoomId) return
+    notifyRoomOpened()
+  }, [activeRoomId])
+
+  useEffect(() => {
+    notifyUnreadCount(unreadRoomCount)
+  }, [unreadRoomCount])
+}

frontend/src/lib/desktopBridge.ts

@@ -15,6 +15,12 @@ declare global {
   interface Window {
     dsDesktop?: {
       setUnreadCount?(unreadRoomCount: number): void
+      // Tells the tray/dock icon the user has genuinely entered a room --
+      // the desktop wrapper's own signal to clear that room's tray
+      // indicator. Deliberately not fired just because the window was
+      // shown/focused/restored, or opened from the tray: none of those mean
+      // the user actually looked at the room the unread message is in.
+      roomOpened?(): void
       showNotification?(notification: DesktopNotificationRequest): void
       onNotificationClick?(callback: (roomId: string) => void): () => void
     }
@@ -38,6 +44,16 @@ export function isDesktopClickListenerSupported(): boolean {
   return typeof window.dsDesktop?.onNotificationClick === 'function'
 }
 
+// Both no-op silently if the bridge or this specific method isn't present,
+// same as showDesktopNotification below -- callers never need to guard.
+export function notifyRoomOpened(): void {
+  window.dsDesktop?.roomOpened?.()
+}
+
+export function notifyUnreadCount(unreadRoomCount: number): void {
+  window.dsDesktop?.setUnreadCount?.(unreadRoomCount)
+}
+
 // No-ops silently if the bridge or this specific method isn't present --
 // callers don't need to guard, matching the rest of this module's
 // capability-detect-per-method philosophy (see #49's "feature-test each

frontend/src/lib/roomIndicators.test.ts

@@ -0,0 +1,56 @@
+import { describe, expect, it } from 'vitest'
+import type { MyRoomItem } from '../types'
+import { clearRoomIndicators } from './roomIndicators'
+
+function makeRoom(overrides: Partial<MyRoomItem> & { id: string }): MyRoomItem {
+  return {
+    name: overrides.id,
+    description: null,
+    is_private: false,
+    is_dm: false,
+    is_archived: false,
+    owner_id: 'owner',
+    created_at: '2026-01-01T00:00:00Z',
+    role: 'member',
+    has_unread: false,
+    has_mention: false,
+    dm_partner: null,
+    email_notifications: false,
+    ...overrides,
+  }
+}
+
+describe('clearRoomIndicators', () => {
+  it('clears has_unread and has_mention on the matching room', () => {
+    const rooms = [makeRoom({ id: 'room-1', has_unread: true, has_mention: true })]
+    const result = clearRoomIndicators(rooms, 'room-1')
+    expect(result[0].has_unread).toBe(false)
+    expect(result[0].has_mention).toBe(false)
+  })
+
+  it('preserves unread state on every other room', () => {
+    const rooms = [
+      makeRoom({ id: 'room-1', has_unread: true, has_mention: true }),
+      makeRoom({ id: 'room-2', has_unread: true, has_mention: false }),
+      makeRoom({ id: 'room-3', has_unread: false, has_mention: false }),
+    ]
+    const result = clearRoomIndicators(rooms, 'room-1')
+    const room2 = result.find((r) => r.id === 'room-2')
+    const room3 = result.find((r) => r.id === 'room-3')
+    expect(room2?.has_unread).toBe(true)
+    expect(room3?.has_unread).toBe(false)
+  })
+
+  it('leaves untouched rooms as the same object reference', () => {
+    const room2 = makeRoom({ id: 'room-2', has_unread: true })
+    const rooms = [makeRoom({ id: 'room-1', has_unread: true }), room2]
+    const result = clearRoomIndicators(rooms, 'room-1')
+    expect(result[1]).toBe(room2)
+  })
+
+  it('is a no-op if the room id is not found', () => {
+    const rooms = [makeRoom({ id: 'room-1', has_unread: true })]
+    const result = clearRoomIndicators(rooms, 'does-not-exist')
+    expect(result[0].has_unread).toBe(true)
+  })
+})

frontend/src/lib/roomIndicators.ts

@@ -0,0 +1,9 @@
+import type { MyRoomItem } from '../types'
+
+// Pulled out of ChatShellPage as a pure function purely so it's directly
+// unit-testable without rendering the whole page -- behavior is unchanged.
+// Only touches the matching room; every other room's object reference (and
+// therefore its unread state) is left untouched.
+export function clearRoomIndicators(rooms: MyRoomItem[], roomId: string): MyRoomItem[] {
+  return rooms.map((r) => (r.id === roomId ? { ...r, has_unread: false, has_mention: false } : r))
+}

frontend/src/lib/theme.ts

@@ -1,4 +1,4 @@
-import type { CustomThemeColors, ThemeName } from '../types'
+import type { CustomThemeColors, TextScale, ThemeName } from '../types'
 
 // The inline custom properties a custom theme sets on :root -- must be
 // removed explicitly when switching to a preset, since an inline style
@@ -76,3 +76,21 @@ export function applyTheme(theme: ThemeName | null, customColors: CustomThemeCol
   for (const varName of CUSTOM_THEME_VARS) root.style.removeProperty(varName)
   root.style.removeProperty('color-scheme')
 }
+
+// #71: percentages, not fixed px -- stacks on top of the browser/OS's own
+// zoom or accessibility text-size setting instead of overriding it. Every
+// component in this app already sizes itself in rem (see tokens.css),
+// which is relative to this root value, so setting it here is the one
+// change that scales text *and* the message-image/video max-size caps
+// (also converted to rem -- see MessageList.css) uniformly, with no
+// per-component work.
+const TEXT_SCALE_PERCENT: Record<TextScale, string> = {
+  small: '87.5%',
+  normal: '100%',
+  large: '112.5%',
+  xlarge: '125%',
+}
+
+export function applyTextScale(scale: TextScale | null): void {
+  document.documentElement.style.fontSize = TEXT_SCALE_PERCENT[scale ?? 'normal']
+}

frontend/src/pages/ChatShellPage.tsx

@@ -1,4 +1,4 @@
-import { useCallback, useEffect, useState } from 'react'
+import { useCallback, useEffect, useMemo, useState } from 'react'
 import { useNavigate, useParams } from 'react-router-dom'
 import { NetworkError } from '../api/client'
 import { listMyRooms, listRoomMembers } from '../api/rooms'
@@ -12,7 +12,9 @@ import { Sidebar } from '../components/Sidebar'
 import { TopBar } from '../components/TopBar'
 import { useAuth } from '../context/AuthContext'
 import { useChatSocketContext } from '../context/ChatSocketContext'
+import { useDesktopRoomBridge } from '../hooks/useDesktopRoomBridge'
 import { MOBILE_BREAKPOINT, useWindowWidth } from '../hooks/useWindowWidth'
+import { clearRoomIndicators } from '../lib/roomIndicators'
 import type { MyRoomItem, RoomMember } from '../types'
 import './ChatShellPage.css'
 
@@ -60,10 +62,13 @@ export function ChatShellPage() {
 
   const socket = useChatSocketContext()
 
-  const clearRoomIndicators = useCallback((id: string) => {
+  const handleRoomRead = useCallback((id: string) => {
-    setRooms((prev) => prev.map((r) => (r.id === id ? { ...r, has_unread: false, has_mention: false } : r)))
+    setRooms((prev) => clearRoomIndicators(prev, id))
   }, [])
 
+  const unreadRoomCount = useMemo(() => rooms.filter((r) => r.has_unread).length, [rooms])
+  useDesktopRoomBridge(activeRoom?.id, unreadRoomCount)
+
   useEffect(
     () =>
       socket.subscribe((envelope) => {
@@ -134,7 +139,7 @@ export function ChatShellPage() {
               onToggleInfo={() => setInfoOpen((v) => !v)}
               infoOpen={infoOpen}
               socket={socket}
-              onRoomRead={clearRoomIndicators}
+              onRoomRead={handleRoomRead}
             />
           ) : (
             !isMobile && (

frontend/src/types.ts

@@ -1,5 +1,12 @@
 export type ThemeName = 'dark' | 'light' | 'midnight' | 'sunset' | 'custom'
 
+// #71: null means "normal" -- see lib/theme.ts's TEXT_SCALE_PERCENT map.
+export type TextScale = 'small' | 'normal' | 'large' | 'xlarge'
+
+// #71: independent of TextScale -- see MessageContent.tsx's
+// EMOJI_SCALE_MULTIPLIER map. Same preset shape for UI consistency.
+export type EmojiScale = 'small' | 'normal' | 'large' | 'xlarge'
+
 // Matches exactly the CSS custom properties frontend/src/styles/themes.css
 // overrides per built-in preset -- kept in sync with
 // backend/app/schemas/custom_theme.py's CustomThemeColors.
@@ -45,6 +52,8 @@ export interface User {
   // Only non-null when theme === 'custom' -- see UserRead's model_validator
   // in backend/app/schemas/user.py.
   active_custom_theme: CustomTheme | null
+  text_scale: TextScale | null
+  emoji_scale: EmojiScale | null
   avatar_filename: string | null
   appear_offline: boolean
   created_at: string

frontend/src/ws/useChatSocket.ts

@@ -10,6 +10,13 @@ interface UseChatSocketOptions {
 const RECONNECT_BASE_DELAY_MS = 1000
 const RECONNECT_MAX_DELAY_MS = 30000
 
+// #76: the backend pings every 30s (WS_PING_INTERVAL_SECONDS in chat.py) --
+// this needs to comfortably tolerate one missed ping plus ordinary network
+// jitter before declaring the connection dead, not fire on the very first
+// late beat.
+const ZOMBIE_TIMEOUT_MS = 65000
+const ZOMBIE_CHECK_INTERVAL_MS = 10000
+
 const desktopMode = isDesktopNotificationsSupported()
 
 // Room join/leave (live message delivery) depends only on visibility --
@@ -51,6 +58,16 @@ export function useChatSocket({ onUnauthenticated }: UseChatSocketOptions) {
   // live WebSocket delivery the user can't actually see will do the job.
   const desiredRoomsRef = useRef(new Set<string>())
   const isVisibleRef = useRef(isVisible())
+  // #76: last time *anything* arrived on the socket (a real message, a
+  // "ping", doesn't matter which) -- the only signal that a connection
+  // sitting at readyState OPEN is actually still alive rather than a
+  // zombie. Without this, a connection that dies without ever sending a
+  // close frame (laptop sleep, a NAT silently dropping an idle mapping)
+  // looks identical to a healthy-but-quiet one: onclose never fires, so the
+  // already-correct reconnect+rejoin logic below never runs, and new
+  // messages just stop arriving until something unrelated (e.g. switching
+  // rooms, which refetches history over plain REST) papers over it.
+  const lastActivityRef = useRef(Date.now())
 
   const sendRoomFrame = useCallback((type: 'join' | 'leave', roomId: string) => {
     const ws = socketRef.current
@@ -99,6 +116,7 @@ export function useChatSocket({ onUnauthenticated }: UseChatSocketOptions) {
         if (hasConnectedBefore) checkForUpdate()
         hasConnectedBefore = true
         reconnectDelay = RECONNECT_BASE_DELAY_MS
+        lastActivityRef.current = Date.now()
         setConnected(true)
         // Re-join whatever rooms were joined before a reconnect -- the
         // server has no memory of a dropped connection's prior state. Only
@@ -122,7 +140,17 @@ export function useChatSocket({ onUnauthenticated }: UseChatSocketOptions) {
 
       ws.onmessage = (event) => {
         if (socketRef.current !== ws) return
-        const envelope = JSON.parse(event.data) as ServerEnvelope
+        lastActivityRef.current = Date.now()
+        const parsed = JSON.parse(event.data)
+        // #76: the server's heartbeat -- purely transport-level, not a
+        // domain event, so it's answered and swallowed right here instead
+        // of being forwarded to subscribers (ServerEnvelope's own type
+        // doesn't include it for exactly that reason).
+        if (parsed.type === 'ping') {
+          ws.send(JSON.stringify({ type: 'pong' }))
+          return
+        }
+        const envelope = parsed as ServerEnvelope
         for (const handler of subscribersRef.current) handler(envelope)
       }
 
@@ -143,10 +171,9 @@ export function useChatSocket({ onUnauthenticated }: UseChatSocketOptions) {
         if (stopped) return
 
         // Unexpected close -- a deploy restarting the app server, a brief
-        // network blip, or (absent any app-level ping/pong) an idle
+        // network blip, or an idle connection getting recycled by a reverse
-        // connection getting recycled by a reverse proxy in front of it.
+        // proxy in front of it. Retry with exponential backoff instead of
-        // Retry with exponential backoff instead of leaving the chat
+        // leaving the chat silently dead until the user manually reloads.
-        // silently dead until the user manually reloads.
         reconnectTimer = setTimeout(connect, reconnectDelay)
         reconnectDelay = Math.min(reconnectDelay * 2, RECONNECT_MAX_DELAY_MS)
       }
@@ -154,9 +181,21 @@ export function useChatSocket({ onUnauthenticated }: UseChatSocketOptions) {
 
     connect()
 
+    // #76: catches the case onclose can't -- a connection that dies without
+    // ever sending a close frame at all (readyState stays OPEN forever) is
+    // otherwise invisible to this hook. Forcing a close here just routes it
+    // through the exact same onclose/reconnect/rejoin path as a normal
+    // disconnect, rather than needing separate recovery logic of its own.
+    const zombieCheckTimer = setInterval(() => {
+      if (Date.now() - lastActivityRef.current > ZOMBIE_TIMEOUT_MS) {
+        socketRef.current?.close()
+      }
+    }, ZOMBIE_CHECK_INTERVAL_MS)
+
     return () => {
       stopped = true
       if (reconnectTimer) clearTimeout(reconnectTimer)
+      clearInterval(zombieCheckTimer)
       socketRef.current?.close()
       socketRef.current = null
     }

frontend/vitest.config.ts

@@ -0,0 +1,13 @@
+import { defineConfig } from 'vitest/config'
+import react from '@vitejs/plugin-react'
+
+// Deliberately separate from vite.config.ts, not an extension of it -- that
+// file's VitePWA plugin does real build-time work (service worker
+// injection, manifest generation) that has nothing to do with running unit
+// tests and would just slow every test run down for no benefit.
+export default defineConfig({
+  plugins: [react()],
+  test: {
+    environment: 'jsdom',
+  },
+})