The PKCE verifier cookie, state encoding, redirect-URI construction,
and code exchange were coordinated across three route handlers that
each knew part of the protocol. Two latent bugs lived in the gaps: a
push-initiated re-authorization skipped PKCE entirely (harmless today
only because the sole pusher, Wahoo, is not a PKCE provider), and the
push-resume callback path never cleared the spent verifier cookie.
oauth-flow.server.ts now owns the lifecycle: initiateOAuthFlow builds
the provider redirect (state + verifier cookie, on every entry path),
and completeOAuthFlow consumes the callback — state decode, verifier
recovery, code exchange, connection linking — returning a
discriminated result the route maps to redirects. The three routes are
thin adapters; the next OAuth provider reuses the flow, not the
pattern.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Garmin Connect as the third connected-services provider (spec:
garmin-import). The interesting parts:
- Push-first ingestion: Garmin has no list endpoint. The webhook
normalizes ping (callbackURL) and push (inline) notification batches
into events; the slow work (authorized FIT download, FIT→GPX via the
shared converter, activity creation) runs in a garmin-import-activity
pg-boss job so the webhook answers fast. Callback URLs are validated
against Garmin's API host before any fetch (SSRF guard).
- History via backfill requests: /sync/import/garmin is a date-range
requester with honest async progress (no pick list — the concept
doesn't exist in a push model). Ranges chunk to Garmin's 90-day cap;
overlaps are free via sync_imports dedupe. Requests persist in
import_batches via two new nullable columns (range_start/range_end).
- OAuth2 + PKCE on the existing oauth credential kind. Design
correction from apply: the verifier rides a short-lived httpOnly
cookie scoped to the callback path — the state param is visible in
redirect URLs and must never carry it. Manifests opt in via pkce:true.
- Deregistration notifications flip the connection to 'revoked'
(row kept for audit, imports retained, re-connect prompt shown).
- Framework evolutions, all additive: parseWebhook returns
WebhookEvent[] (Garmin batches; Wahoo adapted), manifest gains
configured()/importUrl/pkce, importActivity accepts summary stats
for FIT-less imports, manager gains markRevoked.
- Env-gated: no GARMIN_CLIENT_ID → provider hidden on
/settings/connections. Privacy manifest entry (DE+EN). i18n en+de.
Rollout (§6) stays gated on the Garmin Developer Program application
(submitted 2026-06-07). Fixtures are doc-shaped; the staging soak
swaps in recorded payloads if shapes differ.
Gate: typecheck ✓ lint ✓ unit+integration ✓ e2e 70/72 + both known
flakes green isolated ✓ openspec validate ✓
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Follow-up to PR #406 — addresses the two items deferred from the audit:
#7 — Centralize auth helpers
- New `requireSessionUser(request)` in lib/auth/session.server.ts that
returns the user or throws a redirect to /auth/login.
- New `requireSessionUserJson(request)` companion that throws a 401 JSON
response (for fetcher/JSON endpoints).
- Replace the repeated
const user = await getSessionUser(request);
if (!user) return redirect("/auth/login");
pattern across 18 route loaders/actions. Removes the duplicated guard
preamble and gives a single chokepoint to evolve later (e.g., for
terms-version gating).
#8 — Extract heavy loaders into .server.ts siblings
- routes/home.tsx → home.server.ts (DB count query + listActivities +
listRecentPublicActivities)
- routes/users.$username.tsx → users.$username.server.ts (user lookup +
follow state + counts + listPublicRoutes/Activities + persona check)
- routes/settings.connections.tsx → settings.connections.server.ts
(connected_services join + manifest merge)
Each route file shrinks to a thin delegator: `loader` calls
`loadXxx(request)`. The component module no longer transitively pulls
`getDb` and Drizzle schema into its import graph — Vite's tree-shake
already strips server-only code from the client bundle, but the
explicit `.server.ts` suffix makes that contract local and auditable.
Other 17 routes that mix loader/action with components are left as-is
for now: they're each small enough that the split adds churn without
buying much clarity. The pattern is documented by the three examples;
the rest can convert opportunistically when they grow.
Tests:
- lib/auth/session.server.test.ts (4 cases — redirect for missing
cookie, redirect for ghost userId, success path, JSON 401 variant)
Full repo: pnpm typecheck, pnpm lint, pnpm test all green
(181 passed | 31 integration-gated skipped).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Addresses 8 issues from the Journal architecture audit:
1. DB indexes on routes.ownerId + activities.ownerId. Listing queries on
these tables were full table scans; adds composite indexes matching
the order-by columns (updatedAt/startedAt/createdAt).
2. Zod validation on /api/auth/register body. Previously the action
destructured request.json() with zero schema validation.
3. N+1 GeoJSON batch fetch collapsed to a single ANY($1::text[]) query
in both routes.server and activities.server.
4. Webhook envelope validation in /api/sync/webhook/:provider.
5. AbortSignal.timeout(30s) on all external fetches (Komoot, Wahoo) via
a new fetchWithTimeout helper in lib/http.server.ts.
6. .limit(100) on listPublicRoutesForOwner / listPublicActivitiesForOwner.
9. Welcome email moved off fire-and-forget onto a pg-boss job with
retryLimit: 3 (send-welcome-email).
10. process.env.ORIGIN ?? "http://localhost:3000" centralized into
lib/config.server.ts::getOrigin() across 14 call sites.
Issues 7 (centralized apiError/auth guards across 60+ route files) and
8 (split .server.ts boundaries across 20+ route files) intentionally
deferred — both are pure refactors that would balloon this PR past
reviewability and warrant their own focused PRs.
Tests added:
- lib/config.server.test.ts (2 cases)
- lib/http.server.test.ts (3 cases — timeout abort, success passthrough,
caller-signal composition)
- routes/api.sync.webhook.$provider.test.ts (6 cases)
- routes/api.auth.register.test.ts (7 cases — schema rejection paths +
the new welcome-email enqueue assertion)
Full repo: pnpm typecheck, pnpm lint, pnpm test all green.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two cleanups in one pass:
1. Update import paths app-wide from `~/lib/auth.server` to
`~/lib/auth/session.server` for the four session helpers
(sessionStorage, createSession, getSessionUser, destroySession).
~40 files: 33 simple path swaps where the file imported only session
symbols, 5 splits where it also imported per-method auth functions
(auth.verify.tsx, api.settings.email.ts, activities.\$id.tsx,
routes.\$id.tsx, auth.accept-terms.tsx) — those keep one import
from auth.server (for verifyMagicToken, canView, recordTermsAcceptance,
etc.) and gain a second import from auth/session.server.
Two more files used relative paths and were missed by the first
grep pass (lib/oauth.server.ts and routes/oauth.authorize.tsx) —
migrated too.
The @deprecated re-exports block in auth.server.ts is gone.
2. Rename the new auth files to follow the project's `.server.ts`
convention so Vite/React Router treat them as server-only (they
read process.env.SESSION_SECRET, hit the DB, etc. — must NOT enter
the client bundle):
- auth/session.ts → auth/session.server.ts
- auth/completion.ts → auth/completion.server.ts
- auth/completion.test.ts → auth/completion.server.test.ts
Done with `git mv` so blame is preserved.
Verified: typecheck + lint green; 126 unit tests pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The connect loader was setting state to a raw user.id, but the callback
parses state as base64-JSON via decodeOAuthState and falls back to {}
when the parse fails. That sent users back to /settings, which now
redirects to /settings/profile, hiding the freshly saved connection on
a different tab.
Encode the state as a proper PushOAuthState with returnTo set to
/settings/connections so the callback lands on the page that initiated
the flow.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
