stevezau/shortlist v1.11.0

on GitHub

Back up /config before upgrading: this release adds database migrations 0102 to 0111. No settings need
changing, and assistant access stays off until you opt in.

Added

• Assistant access (MCP, #141). Optional, and off until SHORTLIST_MCP_URL is set. Connect a local or
  hosted AI assistant that can view Shortlist or manage its supported settings, rows and saved
  configuration. You approve each named connection yourself, and the assistant never sees the owner API
  token. Migrations 0103 to 0108 add the tables for its connections, approvals, budgets and request
  history. See Connect an assistant.
• Hold requests by genre or TMDB tag. Settings, Requests has a new "Don't request these
  automatically" filter: movies with a chosen genre or TMDB tag wait in the inbox for you to approve instead of being
  sent on their own.
• More seasonal presets, and a simpler way to add a season to a row.
• Requests has its own Settings tab. Settings now has four tabs. Changed settings are marked with their
  default and a reset.
• Warnings view in Activity. The log can be filtered to warnings, with context, and warning lines are
  tinted.
• A per-user Kometa guide on the docs site.
• A webhook event for a skipped scheduled job (job.skipped). Webhook events can now be chosen before an
  address is saved; only the on switch waits for one.
• Pick reasons name what the picks share. "Because you watched Dune — more sci-fi and action", and the
  person page groups a person's picks by reason.
• Dashboard. "Last run" sums the night's scheduled runs, Finished says how many are still going, and both
  bar charts say which shade is finished and which is still going. Durations of an hour or more read as hours
  and minutes.
• First-run wizard has a TMDB key walkthrough, and its last step shows the rows it built.

Changed

• Pause all users moved from the Danger zone to Settings, System, Run speed. Only the full uninstall
  stays in the Danger zone.
• Assistant consent names the read-only scopes it asks for, and access is a simple choice of view or
  manage.
• Numbers agree across pages. People watching counts enabled people only, a person's header counts
  shared-row watches like the dashboard does, and shared rows say they are one copy for everyone. The Requests
  figures count only the titles that really reached the inbox.
• Redesigned screens. The dashboard, rows list, row editor, Add a row (now its own page), Users and
  person page, Privacy, Runs and run detail, Activity, Settings and the first-run wizard were reworked.
  Runs show one verdict ("OK · N warnings") everywhere, and the Privacy page leads with its enforcement
  check and a who-sees-what grid. Breadcrumb headers, plain-word row names and a 12px text floor apply
  throughout.
• Requests page. Each waiting request has one Send action, with Reject and Dismiss in its menu. Waiting
  requests sit in one divided card, and the bulk bar shows how many are selected.
• Jobs page shows each job's details in plain English.
• The Plex card in Settings, Connections has a "Find your token" link.
• Linking a Plex server checks the address. A Plex URL that is not http or https, or that points at a
  cloud-metadata address, is refused.
• The uninstall preview is grouped by library, person and row, and names a shared row "Shared row".
• Confirmed row membership is kept separately from run logs (migration 0102), so clearing run history
  no longer loses what a row delivered or where a watch is credited.
• Deleted row ids are never reused (migration 0109). It seeds the counter above every id that grants and
  history still refer to, so a new row can never inherit an old one's access.

Fixed

• A watch is credited to the verified Plex owner, for both live watches and completed history.
• Shared rows render their configured poster, and retried rows keep their delivery evidence.
• Row names fill {user} instead of stripping it.
• The requests-row check redacts credentials in a source failure before logging or returning it.
• Privacy never shows a green all-clear above an exposure: the Off banner, the exposed-rows badge, the
  first-run finish and the person page all follow the engine's reading, including during a plex.tv outage.
• A run that could not save or check someone's hide rules reads as a warning.
• The Most watched shelf stays inside its scroll area on a phone.
• A schedule rebuild no longer drops that night's past-due row run.
• A scheduled job that is skipped is reported (webhook and bell) instead of vanishing.
• Watch sessions left open by a restart close as orphaned instead of timeout.
• Web search no longer caches a SearXNG answer that was not JSON. An unusable Exa answer is cached for an
  hour, and the run's trace lists the seeds that failed.
• Retention pruning also drops the cached dashboard report, so the dashboard cannot show pruned data.
• The log view widens its window once when a stretch of debug lines pushed every info line out.
• Tab strips keep the active tab in view, small controls have larger touch targets, and the rows, Users and
  Privacy pages read cleanly from 320px wide up.

Removed

• GET /api/report/engagement no longer returns losing, stop_points or observed. Nothing in the
  app read them; people is unchanged.
• GET /api/system/syncs is gone. Nothing in the app used it.