v0.3.0

A campaign is more than a graph of names. This release gives entities a
sheet — a character sheet you design yourself, fill in and print — and
gives your players a way in: a link that shows them exactly the cards you
chose to reveal, on your machine, over the Wi-Fi, or over the internet.

Added

• Entity templates and character sheets. An entity can now be bound to a
  template: a set of fields plus a layout for them. One layout renders three
  ways — a compact sheet in the graph drawer, a full one in the editor, and a
  print version through your browser's own print dialog — so a character is
  readable at the table and on paper without maintaining it twice. Four
  built-in templates ship (Character, NPC, Location, Faction); the Character
  one is a full D&D 5e sheet whose saves, skills and passive scores are
  computed, not typed.

  A template stays a preset, never a schema: fields it creates are yours
  to edit, add to and delete afterwards, and a field the layout does not
  mention is shown under "Other fields" rather than hidden — a sheet can never
  swallow data. An entity with no template keeps the plain field list it
  always had.

• A designer for those templates, under Project settings → Templates.
  Build a sheet out of regions (a flat band, side-by-side columns, or tabs),
  sections and blocks, by dragging; a built-in can be duplicated and rebuilt
  for your own system. Ten widgets draw the same stored value different ways —
  a number can be a plain input, an ability-score box with its modifier, a row
  of rating dots, or a current/max tracker — because how a value is drawn is
  a property of your system, not of the store. The modifier next to a stat
  carries its own formula (D&D halves the score; a system with no such notion
  sets it to nothing), and the whole sheet is keyboard-navigable, tabs and
  rating dots included.

• Computed values. A block can show a value derived from other fields —
  an ability modifier, a skill bonus that accounts for proficiency, a carrying
  capacity — written in a small expression language of its own, with no eval
  and no reach beyond the current entity's fields. Proficiency toggles sit
  right in the row, so ticking "Stealth" recomputes the bonus in place.
  Whether the result carries a leading "+" is per block: "+3" says "add this
  to a roll", which a saving throw means and a passive score does not.
  A bad formula is refused when the template is saved, with the reason shown
  in the designer, rather than surfacing later as a wrong number on a sheet.

• Sheet presets. A section you built — an ability with its skills, a
  resource tracker, a checklist — can be saved and dropped whole into another
  template. Three ship built in.

• Limited player access. You can now let your players see the world you
  choose to show them. Reveal a card (one click from the board, or the "Player
  access" panel on any entity), write a separate text meant for players, and
  tick which of its fields they may see — everything else stays hidden. Each
  player gets an invite link; opening it shows only the revealed cards and a
  read-only board, and lets them keep their own notes, public to the party or
  private (you see all of them). Manage players under Project settings →
  Players: invite, rotate a leaked link, revoke (keeps their notes), or
  delete (removes them). Nothing is shown to anyone until you reveal it.

• Play over the local network, or over the internet. start.bat --lan opens
  the app to your Wi-Fi so players connect from their own devices;
  start.bat --internet additionally asks the router, over UPnP, to forward the
  port, and invite links switch to your public address — players anywhere just
  open a link, with nothing to install. The mapping is removed when the app
  closes. Without a flag nothing changes: the app stays on this machine.
  When internet access can't work, the app says which of the causes it is —
  a shared provider address (CGNAT), UPnP switched off, or a router that
  refused — because each needs a different fix, and the local network keeps
  working meanwhile. Shown both in the launcher and under Project settings →
  Players.

• Optional HTTPS. Point CAMPAIGN_SSL_CERTFILE and CAMPAIGN_SSL_KEYFILE
  at a certificate and its key and everything, invite links included, moves to
  https. Off by default; half a pair is refused at startup rather than serving
  plain HTTP while looking secure.

• One process, one port. The app and its API now share port 8000 — one
  firewall rule, one port to forward, and no CORS. The launcher builds the
  interface at startup instead of running a second dev server.

• Updates ask before they happen. When a git-cloned copy finds a newer
  version on start, the launcher shows what version it is and the changelog for
  it, then offers update now, later, or skip this version — and remembers
  the choice (ask / automatic / never) in backend/data/update.conf. The
  in-app preferences popover shows the current version, any waiting update, and
  lets you change the mode. Double-clicking start.bat stays unattended: an
  unanswered prompt just continues without updating after 30 seconds.

• A real editor toolbar for rich text fields. Block format (paragraph,
  four heading levels, quote, code block), font family and size, text colour
  and highlighter, alignment, checklists, indent/outdent, links, tables and a
  clear-formatting button, next to the bold/italic/list controls that were
  there before. Everything is one click away with a tooltip; nothing changed
  about how field values are stored.

• Tables and checklists survive Markdown export/import. They round-trip as
  GFM tables and - [x] items, and ==highlighted== text keeps its
  highlight. Text colour, font and alignment are presentation-only and are
  dropped on Markdown export — the text itself always survives.

• A live demo in the browser, at
  hadesxgod1337.github.io/loregraph
  — a sample campaign with its graph, its sheets and a scripted assistant,
  running entirely in your browser with no backend, no install and no API key.
  Nothing you do there leaves the tab, and a reload restores the world.

• Import a whole party's character sheets in one go. "Импорт листов
  персонажей" now sits on the Entities page, where the party is, instead of
  behind a connection you first had to create in project settings — the
  button makes that connection itself. Drag in every .json the players
  exported from longstoryshort.app at once; the dialog leads with the file
  drop (the path that always works) and folds the share link — which needs
  the site to answer, and it does not — into "other ways". The result names
  every character that came in, and a file that fails to parse costs that one
  character, not the run.

Changed

• Integrations are their own section in the left rail. They were a tab
  inside project settings — a place you had to remember the way to, for
  something you use during a session rather than configure once.
• An imported character arrives as a character sheet. Import binds the
  party-member template — your own if the project has one, the built-in
  otherwise — so a freshly imported party is readable and printable without
  binding a template by hand, once per character. Characters imported earlier
  get theirs on their next refresh.
• The two switches on a connection say what they do, and only appear where
  they do something. "Use for grounding" is now "Let the assistant read this
  source" and "Auto-push after commit" is "Export automatically after
  approval", each with a line under it explaining the consequence. Both used
  to be offered on every connection, including ones that can neither be read
  by the assistant nor export anything — as did the Export and Import buttons,
  which answered 422 on connections with no such capability. Every action and
  switch is now gated on what the connector actually supports.

Fixed

• Grounding through LongStoryShort no longer silently contributes nothing.
  LSS publishes no JSON endpoint we may call, so the live fetch fails for
  everyone — and the source answered with an empty list, making the switch a
  no-op. It falls back to the sheet as last imported, labelled as such so
  stale HP can't read as current.
• The entity list refreshes after an import. It kept showing the party as
  it was before.
• Toolbar buttons show what is actually active. Bold, lists, headings and
  the rest never lit up while typing and never turned off again — the toolbar
  was frozen at whatever the document looked like when the editor opened, so
  the only way to see a state change was to select text first. It now follows
  the cursor.

Migration

• The database gains four tables (templates, sheet presets, players, player
  notes) and three entity columns, added automatically on first start —
  nothing to do. Every entity starts hidden: players see nothing until you
  reveal it. Nothing you already have is bound to a template; existing
  entities keep their plain field list until you pick one.
• Project export now carries your templates and sheet presets, each entity's
  template binding, and its reveal state, player text and field whitelist.
  Files exported before this version import cleanly (no templates, all
  hidden), and a file written now still opens in 0.2.0 — the older build just
  ignores what it does not know. Players and their notes are not part of
  an export — invite tokens and personal notes never travel in a file;
  re-invite players after importing a project elsewhere.
• Serving attachments moved from a static mount to an access-checked route.
  URLs are unchanged; there is nothing to migrate. If you run behind a reverse
  proxy, set CAMPAIGN_TRUST_LOOPBACK=false and put real authentication in
  front — loopback trust assumes the peer address is the real client.
• The app now lives on port 8000 only. Port 5173 is no longer used by the
  launcher, so a firewall rule for it can be removed. Open http://localhost:8000
  and update any bookmark. Running the frontend dev server by hand still works
  for development and still uses 5173.