familysync/.planning/.continue-here.md

context, phase, task, total_tasks, status, last_updated

context	phase	task	total_tasks	status	last_updated
default	none (pre-Phase-1 — project initialized, planning not started)	0	5	paused	2026-06-03T20:34:20.534Z

<current_state> /gsd-new-project ran to completion. FamilySync is fully initialized: PROJECT.md, config.json, research (STACK/FEATURES/ARCHITECTURE/PITFALLS/SUMMARY), REQUIREMENTS.md (20 v1 reqs), and an approved 5-phase ROADMAP.md (MVP mode) are all written and committed. Working tree is clean. No phase/plan/task work has started yet. The immediate next step is /gsd-discuss-phase 1. </current_state>

<completed_work>

• PROJECT.md written + committed (ad2aad8)
• config.json: interactive mode, standard granularity, parallel, balanced (Sonnet) models, research + plan-check + verifier + drift-guard all on (1ce0348)
• Domain research: 4 parallel researchers + synthesis (0f79277)
• REQUIREMENTS.md: 20 v1 requirements — AUTH-01..03, CAL-01..08, LIST-01..04, NOTIF-01..03, PWA-01..02 (a20afd4)
• ROADMAP.md + STATE.md: 5 phases, 20/20 coverage; CLAUDE.md generated (ac82d32) </completed_work>

<remaining_work>

5 phases, none started:

1. Foundation + Broker Spike — AUTH-01/02/03, CAL-01, CAL-08
2. Calendar Display — CAL-02, CAL-03
3. Event Write-Back + PWA Install — CAL-04/05/06/07, PWA-01/02
4. Shared Lists + Live Sync — LIST-01/02/03/04
5. Web Push Notifications — NOTIF-01/02/03 </remaining_work>

<decisions_made>

• Fastmail calendars are CalDAV-only — JMAP-for-calendars is not available (pending RFC 8984). Broker authenticates with a Fastmail app password, NOT an API token (tokens are JMAP-only). This resolves the long-open JMAP-vs-CalDAV question.
• Stack: Hono + Drizzle (mysql2/MariaDB) + tsdav + ical.js + rrule + vite-plugin-pwa + web-push + @hono/oidc-auth; SSE (+ optional Redis) for live list sync.
• CAL-08 personal-calendar overlay is spike-gated in Phase 1 with a shared-family-only fallback.
• v1 scope: all 4 calendar views; recurring create+display only (single-instance edit -> v1.x); all 3 push types (event reminder, list-change, event-change).
• Vertical MVP structure (each phase = end-to-end user capability).
• Authelia already deployed + both members have accounts -> Phase 1 auth is OIDC-client integration only. </decisions_made>

- None blocking. One important open question (see Pre-Execution / human actions): where the wife's personal calendar actually lives (Fastmail vs iCloud) — determines whether CAL-08 is achievable.

Required Reading (in order)

1. .planning/ROADMAP.md — the 5-phase plan and per-phase success criteria
2. .planning/research/SUMMARY.md — cross-cutting findings; read before Phase 1 planning
3. .planning/research/PITFALLS.md — phase-mapped risks (CalDAV RRULE/ETag, iOS Web Push, Authelia, Pangolin)
4. .planning/REQUIREMENTS.md — 20 v1 requirements + traceability
5. .planning/PROJECT.md — project context, constraints, open questions
6. .planning/notes/familysync-architecture.md — reasoning behind dropped options (Baikal/Vikunja)

Critical Anti-Patterns (do NOT repeat these)

• ANTI-PATTERN: Assuming JMAP for calendar read/write → it does not exist on Fastmail. Use CalDAV (tsdav) only.
• ANTI-PATTERN: Building calendar UI before proving the personal-calendar share works → Phase 1 spike gates CAL-08 first.
• ANTI-PATTERN: Treating iOS Web Push as working from a Safari tab → it requires an installed (Add-to-Home-Screen) PWA; the guided install UX is load-bearing, not optional.

Infrastructure State

• Unraid + Docker Compose host; Authelia deployed (both members have accounts); Pangolin/Newt tunnel for public access (no open ports); MariaDB available, Redis available, no PostgreSQL.
• Nothing for FamilySync deployed yet — greenfield repo, only .planning/ + generated CLAUDE.md exist.

Pre-Execution Critique Required

• Before/at Phase 1: confirm where the wife's PERSONAL calendar lives (Fastmail vs iCloud). If iCloud, CAL-08 (personal overlay) is unreachable via the Fastmail broker and degrades to shared-only (v1.x). This is the Phase 1 go/no-go spike.

This is a clean handoff right after project setup — no mid-task state to untangle. The roadmap is approved as-is. The one piece of judgment to carry forward is the CAL-08 risk: front-loaded into Phase 1 deliberately. Knowing the wife's calendar location before planning Phase 1 would sharpen it.

<next_action> Start with: /gsd-discuss-phase 1 — and open by confirming where the wife's personal calendar lives, since it drives the CAL-08 go/no-go spike. </next_action>