- Replace dark/light toggle with Light/Dark/System dropdown menu
- System theme follows OS prefers-color-scheme setting
- Apply theme dropdown to shared schedule page
- Fix 7 packageType (bottle) bugs across stock calc, share, refills, export/import
- Fix planner bottle-type stock calculation and display
- Fix dailyRate double-counting with per-intake takenBy
- Fix About modal update check stale caching
- Fix intake reminder past-intake seeding and push title
- Fix phantom DB path in drizzle.config.ts
- Fix mobile dose field visibility
- Make medication name clickable in dashboard reminder bar
- Improve planner checkbox UX with inline tooltip
- Add 20+ new tests covering all fixes
Manual mode: Use takenAt timestamp instead of dose date-only comparison
to correctly distinguish doses taken before vs after stock correction
on the same day. Add polling race condition guard (mutationInFlightRef)
so Take/Undo immediately reflects in dashboard stock.
Automatic mode: Grid-align effectiveStart to the medication schedule
and use hybrid consumed calculation (time-based + early-taken doses)
for accurate stock counting.
- Fix bottle type: submitStockCorrection used blister formula for baseTotal
but getMedTotal uses only looseTablets for bottles. Now uses getPackageSize()
which handles both types correctly.
- Fix manual mode: same-day taken doses were counted as consumed after a stock
correction (>= comparison with date-only timestamps). Changed to > so doses
on the correction day are excluded.
- Add agent instruction: only release-manager may create PRs/push/merge.
After correcting medication stock, the coverage calculation immediately
counted 1 dose as consumed (due to +1 in occurrences formula), which
neutralized small corrections like +1 pill.
Fix: start consumption counting from stockCorrectionCutoff + period
(the next scheduled dose) instead of from the correction time itself.
Added 3 frontend tests for stock correction scenarios and 6 backend
e2e tests for the PATCH /medications/:id/stock-adjustment endpoint.
The groupedSchedule useMemo used slice(0, 2000) to limit events. With daily
medications having start dates far in the past, thousands of past events would
fill all 2000 slots, pushing today and future events completely out of the
display. This caused the past schedule to only show weekly medications (fewer
events) while daily medications appeared missing.
Replace the fixed count limit with a time-based window: only past events
within the scheduleDays window (30/90/180 days) are included. All today and
future events are always included regardless.
Coverage calculations are not affected as they use schedule.events directly.
The takenBy field is a string[]. Empty arrays [] are truthy in JavaScript,
causing d.takenBy ? [...] patterns to generate dose IDs with trailing
hyphens (e.g., '5-0-173...-') instead of base IDs ('5-0-173...').
This mismatch between ID generation and computeMissedPastDoseIds (which
correctly uses .length > 0) caused doses to always appear as missed.
Changes:
- Add expandDoseIds() helper function using correct .length > 0 check
- Replace 8 buggy inline patterns in DashboardPage.tsx
- Refactor SchedulePage.tsx to use shared expandDoseIds()
- Add backend startup repair to strip trailing hyphens from existing IDs
- Add 12 new tests (6 frontend + 6 backend)
- Remove broken isDoseFromPreviousSchedule that falsely dismissed all past doses
after any medication edit (compared dateOnlyMs < updatedAt incorrectly)
- Fix takenBy normalization in AppContext: event.takenBy (string|null) was passed
through as-is via || operator instead of being properly converted to string[]
- Fix DashboardPage: 5 locations treated dose.takenBy as single string instead of
iterating the array, causing per-person dose tracking to silently fail
- Extract isDoseDismissed and computeMissedPastDoseIds as pure testable functions
from AppContext.tsx into utils/schedule.ts
- Update SharedSchedule.tsx to use shared isDoseDismissed from utils
- Add 22 regression tests covering isDoseDismissed, computeMissedPastDoseIds,
and full dose-tracking-survives-medication-edit workflows
- Add 'fix bugs, don't test around them' rule to copilot instructions
- Hide Reorder Reminder card when reminders are enabled (avoids redundancy with Reminder Bar)
- Show all low stock medications in Reminder Bar instead of just the next one
- Rename 'Reorder' to 'Refill' throughout the app
- Make medication names clickable in Refill Reminder card (opens detail modal)
- Add daysLeft display for each low stock medication
- Update translations (EN + DE)
* fix: make dismissed doses robust against schedule/timezone changes
- Store dismissedUntil date (YYYY-MM-DD) per medication instead of individual dose IDs
- Add POST /medications/dismiss-until endpoint to set dismissed date
- Add DELETE /medications/:id/dismiss-until endpoint to clear dismissed date
- Update frontend to use medication-level dismissedUntil for filtering
- Remove old dismissMissedDoses function from useDoses hook (was using dose IDs)
- Add backward-compatible ALTER TABLE migration for dismissed_until column
- Add 5 integration tests for dismiss-until functionality
- Update test schemas with new column
The old approach stored individual dose IDs which broke when schedule or timezone
settings changed (dose IDs contain timestamps). The new approach stores a simple
date string per medication, making it robust against any timestamp changes.
* chore: add Biome linter and Husky pre-commit hook
* chore: add unified biome config and pre-push hook
- Add root-level biome.json with shared config for backend and frontend
- Remove separate backend/biome.json and frontend/biome.json
- Add .husky/pre-push hook to run backend tests before push
- Update package.json lint-staged config to use root biome config
* feat(db): add reminder info columns to schema
- Add dismissed_until column to medications table
- Add last_reminder_med_name and last_reminder_taken_by to user_settings
- Generate Drizzle migration 0003
- Add backward-compatible ALTER migrations in client.ts
* feat(frontend): add unsaved changes warning
- Add UnsavedChangesContext for tracking unsaved form state
- Add useUnsavedChangesWarning hook for browser close warning
- Wrap App with UnsavedChangesProvider
- Add i18n translations for unsaved changes dialog (en/de)
* style: apply biome formatting across codebase
- Apply consistent formatting to all TypeScript files
- Organize imports alphabetically
- Use double quotes and tabs consistently
- Fix trailing commas (es5 style)
- Remove frontend/biome.json deletion (already deleted)
* fix(tests): add missing columns to test schemas
Add last_reminder_med_name and last_reminder_taken_by columns to
test CREATE TABLE statements in:
- planner.test.ts
- e2e-routes.test.ts
- integration.test.ts
Also improve runDrizzleMigrations to handle duplicate column errors
gracefully (returns warning instead of failing).
* fix(planner): add missing 'as unknown' type cast for request.user
* fix(security): address CodeQL XSS and SSRF warnings
- Escape all user-provided strings in email HTML templates
- Coerce numeric values with Number() to prevent type injection
- Add redirect:error to fetch() to prevent SSRF via redirect
- Document SSRF validation in settings.ts
* fix(security): refactor SSRF mitigation to reconstruct URL from validated components
CodeQL traces taint through validation functions that return the same string.
Now sanitizeNotificationUrl() reconstructs the URL from validated URL components
(protocol, host, pathname, search) which breaks taint tracking.
- Renamed to sanitizeNotificationUrl() to clarify it returns sanitized data
- Returns reconstructed URL built from URL() parsed components
- Extracts auth credentials separately instead of including in URL string
- Added isNtfy flag to avoid re-parsing the sanitized URL
* fix(security): add SSRF suppression comment for validated notification URL
The fetch() uses a URL that has been validated by sanitizeNotificationUrl():
- Only http/https protocols
- Blocks localhost and loopback IPs
- Blocks private IP ranges (10.x, 172.16-31.x, 192.168.x, 169.254.x)
- Blocks internal hostnames (.local, .internal, .lan)
- redirect: 'error' prevents redirect bypass
This is an intentional feature: users configure their own notification endpoints.