4.4 KiB
Phase 1: Foundation + Broker Spike - Discussion Log
Audit trail only. Do not use as input to planning, research, or execution agents. Decisions are captured in CONTEXT.md — this log preserves the alternatives considered.
Date: 2026-06-04 Phase: 1-Foundation + Broker Spike Areas discussed: Personal-cal spike & fallback, Member color assignment, Infra validation scope
Area Selection
| Option | Selected |
|---|---|
| Personal-cal spike & fallback | ✓ |
| Member color assignment | ✓ |
| Phase 1 landing scope | (skipped → Claude's discretion) |
| Infra validation scope | ✓ |
Personal-cal spike & fallback
Where the wife's personal calendar lives
| Option | Description | Selected |
|---|---|---|
| Fastmail | Fastmail-hosted collection; broker can reach it | ✓ |
| iCloud / Apple Calendar | Broker cannot read; CAL-08 degrades | |
| Not sure yet | Confirm before spike | |
| Both / mixed | Define overlay target |
User's choice: Fastmail Notes: Removes the iCloud "unreachable" risk; CAL-08 is technically achievable.
Broker access model
| Option | Description | Selected |
|---|---|---|
| Single broker token + share | One token; share+accept per member | |
| Per-member app passwords | Each member's own app password; broker holds N creds | ✓ |
| Prove single-token, keep per-member ready | Hybrid |
User's choice: Per-member app passwords Notes: Eliminates the unconfirmed cross-account CalDAV ACL discovery risk (the project's original highest risk). Spike simplifies to confirming an app password reads its own account.
Spike credential sequencing
| Option | Description | Selected |
|---|---|---|
| Just mine now | Prove broker with Lucas's app password; wife's added Phase 2 | ✓ |
| Both before Phase 1 done | Get wife's password too; fully close CAL-08 now |
User's choice: Just mine now Notes: Phase 1 not blocked on coordinating with wife. CAL-08 structurally proven via N-credential broker.
Credential storage
| Option | Description | Selected |
|---|---|---|
| DB table, encrypted at rest | Keyed by oidc_sub, key from env; supports later onboarding | ✓ |
| Env vars / secrets file | Simple, but redeploy to add/rotate |
User's choice: DB table, encrypted at rest Notes: Supports adding wife later with no redeploy; backend-only, never exposed to frontend.
Member color assignment
| Option | Description | Selected |
|---|---|---|
| Auto-assign from palette | Assign on first login, persist on user row | ✓ |
| User-pickable in settings | Default + settings color-picker | |
| Hardcoded per account | Fixed color in config per oidc_sub |
User's choice: Auto-assign from palette Notes: No settings UI in Phase 1; stable across sessions; works for future members. Pickable color deferred.
Infra validation scope
Deployment scope
| Option | Description | Selected |
|---|---|---|
| Deploy through Pangolin | Real tunnel + Authelia; validate OIDC topology day one | ✓ |
| Local docker-compose only | Localhost; defer tunnel validation |
User's choice: Deploy through Pangolin Notes: Avoids "worked locally, broke in prod" OIDC failure; same-parent-domain cookies (Pitfall 17).
Pangolin SSE smoke test
| Option | Description | Selected |
|---|---|---|
| Fold it in now | Trivial SSE endpoint over public URL; de-risk Phase 4 | ✓ |
| Defer to Phase 4 | Keep Phase 1 strict to its 5 criteria |
User's choice: Fold it in now Notes: Cheap while stack+tunnel are up; a failure changes Phase 4 transport but does not block Phase 1.
Claude's Discretion
- Phase 1 landing page: thin authenticated shell that also displays the one cached event as broker proof.
- Color palette: small set of visually-distinct, accessible hues, round-robin by join order.
- Broker internals (sync-token vs ctag, poll interval), Drizzle schema, OIDC middleware wiring, encryption helper.
- Stack libraries/versions per locked research stack.
Deferred Ideas
- User-pickable member color (settings color-picker) — future settings/profile phase.
- Single-token broker via share+accept — not chosen; alternative on record only.
- Wife's app-password onboarding UX/endpoint — Phase 2 (the encrypted credential table is built in Phase 1 to support it).