Remove undici 7.18.2 override, bump wrangler to 4.83.0

docs/kanban.md

@@ -1,22 +1,26 @@
 ---
+
 kanban-plugin: board
+
 ---
 
 ## Todo
 
 - [ ] [[todos/telegram-login-link|Simplify Telegram bot auth with login link]]
 - [ ] [[todos/e2e-do-sync-testing|E2E testing for DO-to-DO sync]]
-- [ ] [[todos/remove-undici-override|Remove undici 7.18.2 override]]
 - [ ] [[todos/livestore-testing-ui|Livestore UI feature tests (RTL)]]
 - [ ] [[todos/progress-tracker-sqlite-review|Review stateful SQLite ProgressTracker]]
 - [ ] [[todos/managed-effect-runtime-do|Explore ManagedRuntime for LinkProcessorDO]]
-- [ ] [[todos/links-list-performance|Fix links list rendering performance at 150+ links]]
-- [ ] Rethink app design (inspiration: shiori.sh)
 - [ ] Develop CLI for ingestion and management
 - [ ] Review and consolidate rate limiting / usage limits
 
+
 ## In Progress
 
+- [ ] Rethink app design (inspiration: shiori.sh)
+- [ ] [[todos/links-list-performance|Fix links list rendering performance at 150+ links]]
+
+
 ## Done
 
 - [ ] [[todos/publish-raycast-extension|Publish Raycast extension to Store]]
@@ -42,10 +46,11 @@ kanban-plugin: board
 - [x] [[todos/done/monorepo-conversion|Convert project to monorepo]]
 - [x] [[todos/done/raycast-ingestion|Add Raycast ingestion path]]
 
-%% kanban:settings
 
+
+
+%% kanban:settings
 ```
 {"kanban-plugin":"board"}
 ```
-
-%%
+%%

docs/todos/e2e-do-sync-testing.md

@@ -29,3 +29,27 @@ Unit tests can't cover the DO's subscription-driven concurrency because it requi
 6. **Queue batch concurrency** — send batch of 5 messages → verify `ingestAndProcess` called concurrently (not sequentially)
 
 See [[incidents/2026-04-link-processor-sync-bug]] for full context.
+
+## Current status — blocked on upstream
+
+PR [#46](https://github.com/bohdanbirdie/cloudstash/pull/46) implements 12 miniflare-based e2e tests covering cold/warm boot sync, duplicate detection, concurrent ingests, cross-org isolation, fetch trigger path, and queue batch handling via real DOs. Tests pass locally on macOS but CI on Linux hangs indefinitely.
+
+### Root cause
+
+`vitest-pool-workers` per-test storage isolation snapshots/pops DO SQLite state between tests. LinkProcessorDO's livestore opens SQLite in WAL mode, creating `.sqlite-shm`/`.sqlite-wal` sidecars. The pool's pop logic asserts filenames end in `.sqlite` and throws on the sidecars (`AssertionError: Expected .sqlite, got ...sqlite-shm`). After that, the livestore push fiber is left mid-push against inconsistent state and retries forever — that's the hang.
+
+Passes on macOS because SQLite checkpoints WAL more eagerly than Linux, so sidecars usually don't exist at pop time. Environmental flake, not a test bug.
+
+### Upgrade path blocked
+
+- `@cloudflare/vitest-pool-workers@0.13+` removes isolated storage entirely (would fix this)
+- 0.13+ requires `vitest ^4.1.0`
+- `@effect/vitest` stable is pinned to `vitest ^3.2.0` — no stable release supports vitest 4 yet (only 4.0.0-beta tags)
+
+### Plan
+
+Leave PR #46 open. When `@effect/vitest` stable supports vitest 4, upgrade `vitest` + `@cloudflare/vitest-pool-workers` together and unblock.
+
+Alternatives rejected: `isolatedStorage: false` loosens isolation for all e2e tests; skipping on CI defeats the purpose.
+
+Related: [workers-sdk#11031](https://github.com/cloudflare/workers-sdk/issues/11031).

package.json

@@ -106,7 +106,7 @@
     "zustand": "5.0.12"
   },
   "devDependencies": {
-    "@cloudflare/vite-plugin": "1.32.2",
+    "@cloudflare/vite-plugin": "1.32.3",
     "@cloudflare/vitest-pool-workers": "0.12.21",
     "@cloudflare/workers-types": "4.20260414.1",
     "@effect/language-service": "0.81.0",
@@ -138,10 +138,7 @@
     "vite-tsconfig-paths": "6.1.1",
     "vitest": "3.2.4",
     "web-vitals": "^5.1.0",
-    "wrangler": "4.82.2"
-  },
-  "overrides": {
-    "undici": "7.18.2"
+    "wrangler": "4.83.0"
   },
   "packageManager": "bun@1.3.11",
   "cloudflare": {