- Add logic to detect the best LAN IPv4 address for easier network
access
- Display both localhost and LAN access URLs in the CLI
- Remove file path from journal input autocomplete labels
Introduce `useNavigateWithParams` to ensure that existing URL search
parameters are maintained when navigating between pages. This is
applied to the Article component, FileTree, and journal links to
prevent losing state (like session or player info) during client-side
navigation.
Also intercepts markdown anchor links in the Article component to use
client-side navigation.
Migrate GM action button injection from imperative DOM manipulation to
a reactive Solid component (`RevealManager`) using `Portal`. This
improves cleanup reliability and ensures buttons are proper Solid
components rather than raw HTML strings injected into the DOM.
Also adds URI encoding for link paths and simplifies the reveal
store logic.
Decouple the reveal logic from the `Article` component by introducing
a `RevealManager` component and an `ArticleDomCtx`. This allows the
reveal logic to reactively manage DOM injections and classes without
polluting the `Article` component's props or lifecycle.
Additionally, implements a robust `cleanupInjections` mechanism to
ensure that injected buttons and styling artifacts are properly removed
during navigation, role changes, or disconnections.
Update the spark button injection to use `data-spark` slugs on
`md-table` elements instead of manual table parsing. This improves
reliability by leveraging the slug generated during markdown
rendering.
Introduces the ability for GMs to inject spark roll buttons into
tables that match existing spark completions.
- Replaces `linkPrefill` with a more flexible `actionPrefill` system
to support multiple command types (e.g., `/link`, `/spark`).
- Implements `injectSparkButtons` to identify spark tables in the
DOM and insert action buttons.
- Updates `JournalInput` to handle structured prefill actions.
Replace the heuristic-based slug parsing with a more robust
combined-slug lookup to correctly handle page names containing
hyphens. Added `scanSparkTables` to provide better error messages
when a table is not found.
Implement "spark tables" functionality, which allows users to roll
dice against markdown tables to retrieve specific values.
- Add `sparkTablesSource` to scan markdown files for tables starting
with a dice notation (e.g., d6, d20).
- Implement `/spark` command in the journal to resolve and roll
spark tables.
- Add a new message type `spark` with a dedicated UI component to
render the results.
- Update completions API to include spark table metadata.
Move localStorage and URL parameter management from `journalStream`
to a dedicated `persistence` store. Relocate DOM reveal and heading
injection logic to a new `reveal` store to improve separation of
concerns.
Rewrite `addRevealedClasses` to use a two-pass approach that correctly
cascades revealed status from child headings to parent headings. This
ensures that if a sub-heading is revealed, its parent headings are
also treated as revealed, maintaining proper visibility hierarchy.
Add `addRevealedClasses` to apply `revealed` or `concealed` CSS classes
to article elements based on the current heading hierarchy and
revealed paths. Update `Article` to expose its DOM element via an
`onDom` callback.
Add support for hiding file tree nodes and headings based on path
revelation status. This allows the sidebar to dynamically show only
the relevant parts of the file structure and table of contents.
Introduce the ability to reveal specific sections of a document
rather than entire files. This includes:
- Updating `revealedPaths` to map normalized file paths to sets of
revealed section slugs.
- Adding `isPathRevealed` to handle visibility logic for connected
clients and GMs.
- Enhancing `extractHeadings` to calculate `startLine` and `endLine`
for each TOC node to support precise section identification.
Implement a system to invite players via generated URLs that
automatically configure their name and role.
- Add `InviteDialog` to allow users to generate and copy invite links
- Implement autojoin logic in `ConnectDialog` based on URL parameters
- Add an "+ invite" button to the `JournalPanel` player list
- Ensure `autojoin` parameter is stripped on disconnect to allow
manual connection later
Introduces `plainDetail` to the dice engine to provide a non-HTML
version of the roll results. This allows components to display
the roll breakdown without needing to parse or render HTML.
Refactor journal message types to use flatter, more intuitive names:
- `narrative` becomes `chat`
- `article.reveal` becomes `reveal`
- `roll.request`/`roll.result` becomes a single `roll` type
Implement client-side dice roll resolution for GMs. When a GM uses
the `/roll` command, the result is now calculated locally using
`resolveRollPayload` before being sent to the stream, rather than
requesting a roll from the server.
Introduce 'gm', 'player', and 'observer' roles to the journal system.
This includes:
- Restricting command usage and completions to the GM role.
- Implementing role-based message emission permissions.
- Adding a player list and role indicators in the UI.
- Persisting the user's role in localStorage.
- Updating the connection logic to include the role in the client ID.
- Move the revert functionality from the journal input to individual
message cards for better UX.
- Implement `RevealLink` component for article reveals to allow
navigation without losing URL parameters.
- Add helper functions to format article paths and slugs into
human-readable titles.
- Scroll the selected completion item into view
- Filter commands based on prefix when typing a slash command
- Keep completion menu open while typing a command
- Add data attributes to track completion indices for scrolling
Refactor the completions system to support both CLI mode (via
`/__COMPLETIONS.json`) and a client-side fallback scan for dev mode.
- Implement client-side scanning of the file index for dice and headings
- Add `ensureCompletions` to allow components to await data readiness
- Update `JournalInput` to handle "no results" states in the dropdown
- Improve keyboard navigation (Tab to accept, Enter to select)
- Update dice regex to support `:md-dice[...]` syntax
Implement a new completions system that scans markdown files for
dice notation and headings. The server now exposes a
`/__COMPLETIONS.json`
endpoint and automatically recomputes the completion index when
files are added, updated, or deleted.
Extract the header logic into a standalone `JournalHeader` component.
This new component manages the connection status display, session
dropdown for switching between sessions, and the ability to create
new sessions or disconnect.
Implement URL search parameter synchronization for `session` and
`player`
to allow for bookmarkable links. Added `sessionName` to the journal
stream state to support displaying human-readable session names in the
UI. Improved layout transitions and updated the JournalPanel to show
more detailed connection information.
Remove the `websocket-stream` dependency in favor of
`createWebSocketStream`
from the `ws` package. This simplifies the MQTT broker setup in the
journal server and removes unnecessary transitive dependencies.
Also improve the JournalPanel UI with smoother transitions and better
visibility handling.
Integrate `websocket-stream` to handle MQTT broker upgrades in the
CLI and add granular connection status tracking to the UI.
- Add `websocket-stream` dependency
- Update CLI to use `websocket-stream` for socket upgrades
- Enhance `journalStream` store with `connectionStatus` and
`connectionError` states
- Add visual connection status indicators to `App` and `JournalPanel`
Refactor the journal server to run over WebSockets on the existing
HTTP server instead of a separate MQTT TCP port. This removes the need
for a separate port configuration and simplifies the connection flow
by auto-connecting to the current host.
- Remove `--mqtt-port` CLI option
- Attach Aedes broker to the HTTP server via WebSocket upgrade
- Update UI to remove manual broker URL and player name inputs
- Add "Content Source" access from the sidebar
- Update JournalPanel to auto-connect to the local server
Introduce a complete set of journal components including:
- JournalPanel: main container with mobile/desktop support
- ConnectionBar: MQTT connection and session management
- StreamView and StreamMessage: message log and individual message cards
- ComposePanel: message composition with type selection
- DynamicForm: recursive form generator based on Zod schemas
Implement a robust journal system featuring a type-safe message
registry and an MQTT-backed stream store.
- Add `registry.ts` to manage message type definitions, Zod schemas,
and custom reducers.
- Implement `journalStream.ts` to manage MQTT connections, session
lifecycle, and reactive message state.
- Add built-in message types for `narrative`, `intent`, `roll`,
and `article`.
- Provide utilities for message validation, optimistic local updates,
and message reversion.
Implement a journal server that uses an Aedes MQTT broker to
persist stream messages to JSONL files and manage session metadata
via a manifest.
- Add `aedes`, `mqtt`, and `zod` dependencies
- Implement `createJournalServer` in `src/cli/journal.ts`
- Integrate journal server into the `serve` command
- Add `--mqtt-port` option to the CLI
- Add session lifecycle management (create, update, delete) via
retained MQTT topics
- Add Tailwind source scanning for color utilities
- Implement icon masking support
- Reformat CSS for better readability and consistency
- Expand utility definitions into multi-line blocks
Add a unique `_key` to each layer config to maintain stable sortable IDs
across reorder operations. Update LayerRow to use the sortableId prop
and the `use:sortable` directive. Refactor layer-crud utilities to
generate and propagate `_key` values on creation and initialization.
- Use createMemo for stable IDs to prevent re-renders
- Add opacity and transition styles during drag
- Remove "所有字段已添加" message when no available fields
Introduce a dialog for switching between built-in content and a
user-selected local folder. Persist the folder handle in IndexedDB
so the selection survives page refreshes. Scan the folder for .md,
.yarn, and .csv files to build the file index.
Update App and Sidebar to pass file tree and heading data, and
add TypeScript declarations for the File System Access API.
hypercross
hyper