Implement Phase-1 system testing improvements

.claude/settings.local.json

@@ -24,7 +24,22 @@
       "Bash(gh label list:*)",
       "Bash(gh release:*)",
       "WebSearch",
-      "Bash(git tag:*)"
+      "Bash(git tag:*)",
+      "Bash(yarn e2e)",
+      "Bash(yarn playwright test:*)",
+      "Bash(npx playwright show-report)",
+      "Bash(pkill:*)",
+      "Bash(git restore:*)",
+      "Bash(yarn e2e:*)",
+      "Bash(cat:*)",
+      "Bash(npx tsc:*)",
+      "Bash(tee:*)",
+      "Bash(while ! grep:*)",
+      "Bash(do sleep 5)",
+      "Bash(done)",
+      "Bash(npx playwright:*)",
+      "Bash(unzip:*)",
+      "Bash(timeout 60 yarn playwright test:*)"
     ],
     "deny": [],
     "ask": []

.gitignore

@@ -28,3 +28,7 @@ playwright-report/
 /e2e/test-results/
 /e2e/playwright-report/
 /e2e/playwright/.cache/
+
+# SQLite temporary files
+*.sqlite-shm
+*.sqlite-wal

CLAUDE.md

@@ -270,3 +270,5 @@ yarn test path/to/test-file.test.ts
   - `_ordering` - Sort (prefix `-` for DESC)
   - `_filters` - Key:value pairs with operators (`__eq`, `__neq`, `__null`, etc.)
   - `_search` - Full-text search
+- memory - we should move away from using `waitForLoadState('networkidle')`. It is proving fragile.  We should examine the relevant component, and look for actual UI text, or better still, class-names or data-ids.  Where we have failing tests because it looks like we are waiting for something, it's more likely that we have an invalid selector - and we should specify a data-id.
+- memory - no UI operation takes more than 2 seconds.  It is NEVER necessary to specify timeouts greater than 2 seconds - this just delays the process of us finding out that a test is broken,