Settings gets "Notifications on this device" (PushEditor): personas, alerts in four groups, policy, a test
notification, after a one-time notice of what the push service learns. Each persona ticked gets its own service
worker registration (push-worker.js at push/{slot}/), so its own endpoint and keys, subscribed with its own token
(IHttpService's *As); nothing PrivaPub stores ties two personas together and the worker never reads the payload's
token. PushService follows the tokens (re-subscribing after a renewal), reconciles at every session start, turns
off before sign-out and drops deleted or suspended personas. A click opens that persona's notifications; the app
badge counts the unread ones. Exchanges now ask for the push scope.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WpVCoDeM1qUTRp4fXQdfof
A profile shows the hashtags an account features as chips opening their feed; under another account's header is what the
persona keeps of it for itself alone: for one it follows, a bell (follow with notify), the languages of its posts that
reach the home feed (follow with languages[]) and Recommend (endorsements, private here), and for any account a private
note saved on blur or with Save. Settings edits the persona's featured hashtags (added by name or from those used lately,
10 at most) and its reading preferences (/clientapi/avatar/private/preferences), which CascadingState keeps per persona
and the card applies to sensitive media and content warnings. A hashtag's tab features it on the profile. Explore has Who
to follow (/api/v2/suggestions, dismissed for good after asking); Administration has Staff picks. The Direct tab follows
the direct stream's conversation events; status notifications show the post. The follows CSV carries notify and languages.
Verified: Release build, 0 errors (the 3 warnings were there before). Not run against PrivaPub or in the pasture.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WpVCoDeM1qUTRp4fXQdfof
Administration's Pleroma placeholders give way to real sections: Instance (title, descriptions, contact, thumbnail,
privacy policy, terms, the weekly activity), Rules, Announcements, Relays, Media limits and the Moderation log, next to
Reports, blocked servers (now published or obfuscated), statistics and backups; moderators see Reports and the log. Every
save sends the whole settings form built from the latest read, since PrivaPub replaces what it carries. /about shows the
server's documents to anyone. The report dialog picks the rules a violation breaks, and reports show them. Announcements
show above the feed tabs, dismissed and reacted to, live from the user stream. Login names the server.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WpVCoDeM1qUTRp4fXQdfof
The composer uploads each file as soon as it is attached (streamed from the browser file, under its own ten-minute
timeout), shows it uploading, processing (a 202 followed until ready), ready or failed with Retry, and cancels one
removed; publishing sends the ids, so a failed publish or a retried upload sends nothing twice. Size limits and types
come from /api/v2/instance, a file without a browser type gets one from its extension, and each refusal names the file
and why. Previews are 320 px thumbnails, not whole files in memory. A picture without a description is asked about once
("Publish anyway"; a setting turns it off), descriptions are sent with PUT, and an edit keeps the post's media, language
and sensitive flag.
Profile pictures are checked against the server's types and can be removed. A card shows "Media unavailable" with the
original's link when a picture fails, the blurhash behind a sensitive placeholder, and plays a GIF only on tap under
reduced motion. A failed export says so instead of downloading part of a file.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw