Spec Kitty

Tasks: Offline Queue Enqueue Performance

Work Packages

Mission 6 is intentionally a single WP because the cache invariant must hold across every mutation path simultaneously. Splitting the cache introduction from the mutation-path updates would force an intermediate commit where the cache and disk disagree, and the existing test suite would either fail or need to be temporarily relaxed.

WP01 — Cached row count for offline queue enqueue + benchmark

See tasks/WP01-cached-row-count.md for the full prompt and acceptance criteria. Subtasks:

• T001 — Add _row_count, _load_row_count(), and _size_from_disk() to OfflineQueue.
• T002 — Rewrite queue_event() to use the cached counter; preserve eviction + warning.
• T003 — Rewrite append() to use the cached counter; preserve OfflineQueueFull.
• T004 — Update _try_coalesce, mark_synced, clear, remove_project_events, process_batch_results, drain_to_file, and size() to keep the counter coherent.
• T005 — Add tests/sync/test_offline_queue_counter.py covering cache initialization, increment/decrement on every mutation path, eviction preserves cap, coalesce does not change count, strict-cap behavior, and disk/cache equality invariant.
• T006 — Add scripts/benchmarks/bench_queue_enqueue.py and record before/after numbers in kitty-specs/offline-queue-enqueue-performance-01KRC4HN/benchmark.md.

Acceptance

regressions beyond pre-existing infrastructure-related failures.

before/after delta documented in benchmark.md.

• SPEC_KITTY_ENABLE_SAAS_SYNC=1 uv run pytest tests/sync/test_offline_queue.py tests/sync/test_offline_queue_counter.py -q is fully green.
• Full SPEC_KITTY_ENABLE_SAAS_SYNC=1 uv run pytest tests/sync/ -q shows no
• scripts/benchmarks/bench_queue_enqueue.py produces a measurable, repeatable