gmtool/docs/superpowers/plans/2026-07-20-director-slice5.md

The Director — Slice 5 Implementation Plan

Goal: Add the convention hub — a landing screen listing every slot of a convention with live "live now / up next / done / upcoming" status, that deep-links into each ported scenario and lets the GM return to the hub from a running session.

Architecture: Same pattern as the other slices — a pure engine (analyzeSchedule) that classifies slots from a wall-clock now, a <con-hub> light-DOM component that renders the slot list + summary and emits open-scenario, a scenario registry (src/scenarios/index.js) resolving a slot's scenarioId to its module, and shell routing that switches between the hub and a session (hiding the Rail/stage in hub mode) plus a 🏠 rail chip to return. No fabricated content: only AFTERIMAGE is ported, so continuum-2026.js marks the other four slots scenarioId: null ("Not yet ported") with startsAt: null (times TBC, not invented); a clearly-labelled example-con.js provides fully-timed slots as the demo/test fixture.

Tech Stack: JavaScript (ES modules, no TypeScript), Web Components (light DOM), Vite, Vitest + jsdom. No new dependencies. Fully offline.

Global Constraints

• Offline-first, no runtime framework, light DOM, no TypeScript.
• Con model: { meta: { id, title }, slots: [ { id, scenarioId, title, slot, system?, players?, playMinutes?, startsAt? } ] }. slot is the display label; startsAt (ISO) drives timing; scenarioId deep-links via the registry (null ⇒ not yet ported). A null/absent startsAt ⇒ unscheduled: listed, never live/next.
• Determinism: analyzeSchedule(slots, nowMs) is pure; <con-hub> takes an injected now() like the shell.
• Escaping: all author text (titles, slot labels, systems) via escapeHtml.
• Rail invariant preserved: the hub is a pre-session screen; the Rail is hidden in hub mode and restored (never removed) when a session is open.

Tasks (all complete)

1. Schedule engine — src/con/schedule.js (analyzeSchedule, slotStartMs) + tests/con/schedule.test.js (6).
2. Scenario registry — src/scenarios/index.js (SCENARIOS, getScenario) + tests/scenarios/registry.test.js (2).
3. <con-hub> component — src/components/con-hub.js + tests/components/con-hub.test.js (6).
4. Con data — src/con/continuum-2026.js (real slate, honest placeholders) + src/con/example-con.js (template/fixture).
5. Shell + rail wiring — gm-shell.js (loadCon/showHub/showSession/onOpenScenario, open-hub), director-rail.js (🏠 chip), styles.css, boot in main.js → loadCon(continuum2026) + tests/components/gm-shell-hub.test.js (4).

Validation

• npm test → 122 passing (34 files); build succeeds (35 modules). Manual: hub renders the Continuum slate, "Open" deep-links AFTERIMAGE into a live Rail session, 🏠 returns to the hub. Console clean.

Follow-up (completed after the slice)

• All six Continuum 2026 slots ported + scheduled. continuum-2026.js rewritten from the authoritative con-desk.html SLOTS array (real days/times/systems), every startsAt set so live/up-next is live. Five new scenario modules (day-one, vain-crown, silvery-moon, chopper, princes-bride) transcribe their timelines from the existing gm-utility consoles and are registered; tests/con/continuum-2026.test.js guards resolvability + schedule.

  • Correction found while porting: the real slate is six slots (Slot 5 Silvery Moon was missing from the first hub draft), and "Another Fine Mess" is Get to the Chopper — Another Fine Mess.

• All six scenarios' clue trails ported into the data model (transcribed from the console CLUES/gbclue/module sources via extract-clues.mjs), so the Slice 4 safety-net now works against real content. tests/scenarios/clues.test.js guards shape + solvability. The four GM-brief-sourced trails (afterimage, vain-crown, silvery-moon, chopper) mark every tracked clue essential: true (their source is a single curated spine); day-one and princes-bride carry the source's own essential/optional split.

• All six scenarios' cast rosters ported (NPCs, kind: 'npc') via extract-cast.mjs, normalising the two source schemas ({id,name,ess,sec} consoles and {name,tag,secret} modules) to { id, name, kind, note, secret? }. tests/scenarios/cast.test.js guards shape + unique ids. Note: the cast field has no UI consumer yet (the design reserves it for a future cast dashboard) — this is a data-model port. Pregens/PCs are intentionally excluded (separate character-sheet/handout system).

• Cast tray built. <cast-tray> (src/components/cast-tray.js) renders scenario.cast — name + one-line note, each secret behind a tap-to-reveal — wired into the shell (cast tool key) with a 👥 rail chip. Reveal state is ephemeral component state. tests/components/cast-tray.test.js + gm-shell-cast.test.js cover it.

• BRP dice pack built. brp-d100 (src/dice/rulepacks/brp-d100.js) — fumble on 100, critical ≤ skill/20, special ≤ skill/5, success ≤ skill (bands transcribed from the Day One console). Day One's meta.system switched from coc-d100 to brp-d100. tests/dice/brp-d100.test.js covers it.

• Break timer built. <break-timer> pauses the session clock for a break (reusing the clock's existing pause/resume), so break time never poisons drift. Shell owns the pause/resume + persistence (break store key); the rail shows a ☕ on-break badge; fmtCountdown (M:SS) added to format.js. Tests: break-timer.test.js, gm-shell-break.test.js (incl. the drift-honesty case).

• Parking-lot note + wake-lock built. <parking-lot> captures timestamped threads (session-time stamp, persisted under parkingLot, 📝 chip). src/core/wake-lock.js is an injectable Screen Wake Lock controller (re-acquires on foreground, graceful no-op where unsupported), toggled by a 💡 rail chip that lights when active. Tests: parking-lot.test.js, wake-lock.test.js, gm-shell-parking.test.js, gm-shell-wake.test.js. This completes every design §5 tray tool.

• Markdown → scenario-data generator built. tools/scenario-md.js (pure parse + emit) + tools/md-to-scenario.mjs (CLI, npm run gen:scenario) turn a structured scenario markdown file into a validated scenario module. Deterministic parser over a defined format (frontmatter + Timeline / Clues / Cast lists) — not freeform-prose scraping. tests/tools/scenario-md.test.js (round-trip

  • validation); tools/ is dev-only, never bundled.

• Online art "Generate" built. src/art/generate.js (configurable endpoint, injectable fetch, graceful offline/unconfigured degradation) + src/art/library.js (pure add/dedupe/cap) + art-tray integration: prompt → configured endpoint → image cached into a persisted, searchable library (gmd.art.generated, capped 24). No endpoint/key hardcoded — the GM pastes their gen URL. Tests: generate.test.js, library.test.js, art-tray-generate.test.js. Network never load-bearing.

• Native iPad wrapper scaffolded (Capacitor). capacitor.config.json (webDir dist, appId uk.timevans.thedirector, iOS safe-area/theme), @capacitor/{core,cli,ios} as devDependencies (web app keeps zero runtime deps), ios:add/ios:sync/ios:open scripts, main.js skips the SW under Capacitor, iOS <meta> hints, .gitignore for native build output, and docs/native-ipad-wrapper.md (the Xcode + CocoaPods build guide). tests/tools/capacitor-config.test.js guards the config. The actual iOS project generation/build needs a Mac with full Xcode + CocoaPods (not available in this env) and is documented for the user to run.

Done — the design is fully built.