ReignsAgent now has a creator-focused workspace for building, reviewing, previewing, and shipping Reigns-like card narratives. Local Web, the Node ZIP, and portable Electron ZIPs share the Creator Server and filesystem Workspace; the Hosted PWA uses OPFS and a browser backend over the same domain contracts.
The current priority is depth rather than another host or shell: make Content, Story, Review, AI Assist, and the production player clearer and more actionable for non-technical authors while keeping backend and package boundaries stable.
Story and Content work should now optimize for non-technical content authors: the creator surface should explain what each card does, when it appears, how it moves the story, what is broken, and what the next repair action should be.
Recent merged work established the baseline that future phases should preserve:
- PR #21 shipped one React Creator across local Web, a cross-platform Node ZIP, portable Electron ZIPs, and an offline-capable Hosted PWA. It also delivered durable TOML configuration, multi-project filesystems, OPFS persistence, Workspace/project ZIP interchange, direct-CORS Hosted AI, and shared release verification.
- PR #22 made explicit workbench routes authoritative over persisted panel state, anchored packaged Creator exports under
ReignsAgentData/Buildsindependently of the launch working directory, and added cached-shell fallback for non-success Hosted navigations. - PR #24 delivered responsive pinned/compact/floating navigation, desktop-only panel shortcuts, project management in the header, client-aware English/Simplified Chinese locale controls, and Player launch/return context preservation. Optional rail preferences now degrade safely when
localStorageis unavailable or throws. - Creator onboarding now provides a localized, responsive spotlight tour across the complete authoring workflow, remembers completion per client, preserves deep-link and workspace panel state, and can be replayed immediately from Settings.
- The first repository release pipeline defines
v0.1.0as the canonical identity, verifies legal and private-data boundaries inside every archive, assembles exactly five Creator ZIPs, generates SHA-256 checksums, and keeps manual native runs non-publishing while tag publication remains explicitly gated. - Windows Project release packaging now publishes a validated active Project as a single x64 EXE from local Node/Electron Creator, with deterministic build IDs, project-scoped release history, portable
Buildsoutput, and a restricted native WebView2 host. Hosted Creator retains Web Player ZIP export. /workbenchis the primary Creator surface, with overview, content, story, review, AI Assist, preview, build, and settings panels over shared backend contracts.- Review diagnostics and seeds remain creator-facing and reproducible; developer preview may expose debug state while the production player remains focused on play.
- Skins remain a presentation layer shared through configuration and URL context rather than a dependency of Creator workflows or player behavior.
- Replace remaining implementation-shaped controls and raw fields with author-facing summaries, explanations, repair actions, and progressive disclosure.
- Tighten project lifecycle UX around blank, sample, import, backup/restore, rename, and destructive actions without weakening portable or Hosted persistence contracts.
- Continue cross-skin, responsive, keyboard, and Simplified Chinese visual QA as panels gain richer content.
- Keep direct routes, Player return context, client-local navigation preferences, and shared locale behavior covered by Hosted browser tests.
- Preserve existing API endpoints and content bundle schemas unless a reviewed contract change justifies migration.
- Keep the player interaction model pure: card text plus binary left/right choices. Add richness through authored narrative state, not through inventory, equipment, shops, builds, or other RPG management loops.
- Treat chapters, themes, arcs, endings, and status as data-driven authoring concepts expressed through tags, variables, and metadata. Core must not hard-code chapter progression rules; it should keep handling generic eligibility and state.
- Reframe Content from a field editor into a card authoring desk: human-readable card summaries, appearance-condition explanations, choice-consequence explanations, validation messages, and repair links should be visible before raw JSON or advanced state editing.
- Reframe Story from a graph visualizer into a story structure workspace: graph navigation, chapter/theme grouping, reachability, reviewer coverage, ending paths, breakpoints, and issue navigation should all point back to concrete editing actions.
- Reframe Review from a balance-only panel into narrative QA. Keep gauge pressure, game-over rate, and coverage, but add pacing, dead paths, early endings, unreachable story groups, unvisited endings, and chapter/theme coverage as review concerns.
- Replace the legacy AI Edit framing with AI Assist: a context-aware creator assistance layer over Overview, Content, Story, and Review. It should not become a chat-first product or a separate editing mode that hides the normal workflow.
- Settings should support user-supplied endpoints with a NewAPI-style channel setup: provider channel type with official/brand logo, API key with show/hide control, editable model presets/model id, editable base URL, real backend validation, optional
/modelsfetch, and capability toggles for vision, structured JSON, tool/function calling, reasoning/thinking, and streaming. Protocol, route mode, compatibility family, and JSON mode preference belong in Advanced. Image or vision endpoints may be configured separately only when the call shape requires it. - Avoid profile-management burden in the default UX. Lightweight frontend-owned endpoint/model presets and a manual
/modelsmetadata probe are acceptable; provider profile management, automatic model discovery, multiple saved profiles, MCP, skills, and raw tool integrations are optional developer-mode enhancements after the basic endpoint and action flow works. - Overview should handle empty/sample initialization with a brief composer and safe actions for blank project, sample, import, or generated draft. Full regenerate and clear belong in project settings or a project menu with destructive styling and confirmation.
- AI entry points should be contextual: selected cards, graph nodes/edges, review issue cards, and overview project state open the same action popover/drawer with context summary, recommended actions, optional prompt, draft preview, and apply controls.
- Review should become the strongest AI-assisted diagnostics surface: coverage matrices, ending reachability, gauge pressure, issue cards, and repair proposals should be visible before AI generates fixes.
- AI work needs explicit interaction states. Simple edits should show inline loading; longer edits should show progress stages such as context, request, model response, parse, validate, and ready. User-facing errors should offer retry; raw call details belong in Dev Mode logs.
- Image Endpoint execution now supports capability-driven Generate, Edit, Inpaint, and Outpaint across OpenAI Images-compatible, Gemini Interactions, and Stability Stable Image adapters. Generated binaries remain drafts until explicit Apply, are committed by content hash, and stay wholly outside Core and deployable player AI dependencies.
- Do not migrate the content schema during the docs-first phase.
- Prefer lightweight metadata before engine contracts:
metadata.story.groupsdescribes chapters, themes, arcs, and ending groups;metadata.presentation.gaugescan rename, describe, or hide the default four neutral gauge slots (gauge0throughgauge3) without creating new built-in stats. - Cards should continue to express progression through
requirements, choiceeffects.tags, and choiceeffects.variables. Current requirements may combine tag gates, exact variable gates, and default-gauge threshold gates throughrequirements.factions; this remains story eligibility data, not a new player stat model. Legacyfaith/people/military/treasurykeys are import aliases only.
packages/contracts: Shared schemas for cards, content bundles, diagnostics reports, connector requests, and build manifests.- Fastify remains an optional future transport upgrade if the shared
apps/creator-serverHTTP implementation outgrows Node's built-in server. - Extend the hosted Reviewer Worker with richer incremental progress and resumable runs beyond the bounded v1 workload.
- Production player surface evolution: the shipped Web runtime and Windows native host now use a dedicated responsive decision stage with reliable directional card transitions, direct click/keyboard/drag choices, immediate restart, shared Creator Skin tokens, build-scoped Reign session records, authored-language switching, project/About attribution, and DPI-aware native sizing. Continue with a full translation authoring matrix, broader interaction preferences, custom icons, signing, and additional platforms only through reviewed release work.
- Preserve Browser/Vite development, the Node ZIP, and Electron as parallel hosts over the same WebUI and local API.
- Keep Electron isolated in
apps/desktop-electron; no package or Creator Web code may import desktop APIs. - Preserve the utility-process Creator Server, ZIP-only Windows x64/macOS x64+arm64/Linux x64 Creator outputs, and beside-app
ReignsAgentDataportability model. ZIP-only constrains Creator Electron distributions, not single-file Project player releases. - Keep Windows Project EXEs player-only: embedded authored content, Core/player runtime, and assets are allowed; Creator, Pipeline, Reviewer, AI connectors, settings, and credentials are forbidden.
- Preserve the shared TOML/file Workspace contract for Electron, Node ZIP, and local Web, and the equivalent Hosted OPFS projections owned by
apps/creator-web. - Keep Hosted AI direct-CORS only; do not introduce a public relay or compile server secrets into the static application.
- Treat signing, notarization, publishing, automatic updates, native file dialogs, menus, notifications, and protocol handlers as later opt-in release work rather than v1 shell requirements.
- No full engine migration to Unity, Godot, or another game editor.
- No Fastify or database introduction.
- No content schema migration.
- No default multi-provider/profile management system before the lightweight AI Assist endpoint UX is proven.
- No MCP/skills/tool-agent surface in the normal creator UX.
- No second Creator frontend or host-specific fork of the shared WebUI.
- No built-in RPG management systems such as inventory, equipment, pets, shops, rarity, crafting, character builds, skill trees, or loot progression.