A fresh local `yarn db:migrate`/`yarn dev` failed with SQLITE_CANTOPEN (14)
because libsql does not create the parent directory for a file: URL and the
gitignored apps/api/data/ does not exist on checkout. Added ensureDbDir() and
call it before createClient in both the client and the migrator. No-op for
non-file URLs and for the Docker /data volume. Verified: fresh db:migrate now
creates data/ and all tables; tests + typecheck still green; live round-trip
(/health, sign-up, sign-in, /api/me, 401) works.
The suppressed peer-dep warnings (YN0060/YN0002/YN0068/YN0086) and their comment
referenced @lshay/ui, expo-three and @expo/cli — all removed in the cleanup.
Dropping the filters surfaces the remaining warnings, which are now legitimate
signals about the apps/api stack (drizzle vs better-auth peer ranges), not noise.
Dropped @expo/ngrok (and its 40 platform binary packages) from apps/mobile.
It was only used by `expo start --tunnel` (public dev-server tunnel) — never
imported in app code and unrelated to token storage (that is expo-secure-store).
Removed per IT request; no functional impact on the app or auth.
The guard compared import.meta.url against `file://${process.argv[1]}`, which
never matches on Windows (argv[1] is a drive-letter/backslash path), so
`yarn db:migrate` silently no-opped for local non-Docker dev. Use
pathToFileURL(process.argv[1]).href instead, which is correct on Windows and
Linux alike. Verified `yarn db:migrate` now creates all tables on Windows; the
Docker start path is unaffected.
The Task 3 schema drifted from what better-auth@1.6.18's CLI generates: it used
mode:'timestamp' (epoch seconds) instead of 'timestamp_ms', left verification
timestamps nullable, and omitted the default expressions and helper indexes.
Regenerated src/db/schema.ts from `@better-auth/cli generate` (authoritative per
the plan's version-drift rule) and rebuilt migration 0000 from scratch (no data
exists yet). Converted the index callbacks from the CLI's array form to the
object form required by drizzle-orm@0.36.4's types. Adds session_userId_idx,
account_userId_idx, verification_identifier_idx and the unixepoch defaults.
Tests (health, db, auth sign-up/sign-in, /api/me round-trip) all pass; typecheck
clean; db:generate reports no pending changes.
Tracked planning docs under docs/ (the project's documentation source of
truth per CLAUDE.local.md):
- docs/roadmap.md — vision, decisions, architecture, 6-phase roadmap
- docs/plans/phase-0-foundation.md — TDD plan to stand up the dockerized
Hono + better-auth + Drizzle + libsql backend with an auth round-trip
No migration shipped in the export. Recreates production_tasks and
time_logs (columns derived from the API queries) so the backend can run
against a fresh self-hosted Postgres.
- .gitignore: the bare 'logs' rule was matching the api/logs/ source
dir, so the route was never committed. Anchor it to /logs/ and repair
two log globs where '*' had been mangled to '_'.
- route.ts: add the missing 'import sql' — the file used sql but never
imported it, so every GET/POST /api/logs threw ReferenceError and 500'd.
The export shipped a lockfile out of sync with the manifest (lock had
expo 54.0.35 etc., package.json pins 54.0.34). yarn install reconciled
the lock down to the manifest versions.
The backend is not remote: apps/web is a Next.js 16 app with the API
routes, better-auth, and a Neon Postgres data layer. Document the data
model, the mobile<->web API contract, and the missing /api/logs route.
Insole-production time tracker exported from the Create/Anything AI
platform. Baseline snapshot before any reverse-engineering or cleanup.
- apps/mobile: Expo Router app (iOS/Android/web), the only workspace
- publisher/: standalone OpenNext/AWS deploy tooling for the web side
- Backend (/api/tasks, /api/logs + DB) lives remotely, not in this repo
