# BanGUI — Task List

This document breaks the entire BanGUI project into development stages, ordered so that each stage builds on the previous one. Every task is described in prose with enough detail for a developer to begin work. References point to the relevant documentation.

Reference: `Docs/Refactoring.md` for full analysis of each issue.

---

## Open Issues

Issues are grouped by category and ordered roughly by severity. Each entry describes the current state, the desired end state, pitfalls to watch for, and which documentation needs updating when the task is done.

---

## Bug Fixes / Correctness

---

### TASK-001 — Race condition in `useJailBannedIps`: missing AbortController (done)

**Where fixed:** `frontend/src/hooks/useJailBannedIps.ts`, `frontend/src/api/jails.ts`

**Summary:** Added an `AbortController` ref to cancel stale fetches, passed the signal into `fetchJailBannedIps`, and abort on unmount.

**Where found:** `frontend/src/hooks/useJailBannedIps.ts` — the `load` callback is `async` and calls `fetchJailBannedIps` with no AbortSignal.

**Goal:** Add a `useRef<AbortController | null>` to `useJailBannedIps`. Before each fetch, abort the previous controller and create a new one. Pass `signal` to `fetchJailBannedIps`. In the `useEffect` cleanup return, abort the controller. After every `await`, check `signal.aborted` before calling any state setter.

**Possible traps:**
- `fetchJailBannedIps` in `api/jails.ts` does not yet accept a signal — that parameter must be added first (see TASK-005).
- The hook uses a debounced search value; the debounce timer itself does not need to be cancelled, only the in-flight request.
- Setting state after abort causes a React warning in some versions; always guard with `if (!ctrl.signal.aborted)`.

**Docs changes needed:** None.

**Why:** Without cancellation, rapidly changing the search or page can cause an older, slower response to overwrite the result of a newer request. The user sees stale data with no indication it is stale.

---

### TASK-002 — `HistoryPage` filter effect has a stale `appliedQuery` dependency (done)

**Where found:** `frontend/src/pages/HistoryPage.tsx` lines 214–230. The `useEffect` lists `appliedQuery` as a dependency, reads it inside the effect, and then calls `setAppliedQuery` — which triggers the same effect again immediately.

**Goal:** Remove `appliedQuery` from the effect dependency array and use a `useRef` to hold the last-applied query for comparison. The ref must be updated synchronously when the query changes so the comparison is always accurate without creating a circular dependency.

**Possible traps:**
- The `areHistoryQueriesEqual` utility must still be called to avoid redundant fetches; moving it to a ref pattern requires care not to skip the equality check.
- Removing a dependency while keeping `// eslint-disable-next-line` comments is tempting but wrong; restructure so the lint rule is satisfied naturally.

**Docs changes needed:** None.

**Why:** The current code runs the effect one extra time on every filter change — once when filters change, and once because `appliedQuery` just changed. This causes two API calls per user interaction instead of one.

---

### TASK-003 — `BanUnbanForm` floating promises and no double-submit guard (done)

**Where fixed:** `frontend/src/pages/jails/BanUnbanForm.tsx`, `frontend/src/pages/jails/__tests__/BanUnbanForm.test.tsx`

**Summary:** Converted ban and unban handlers to async functions with separate submit states and disabled submit buttons while requests are in flight.

**Where found:** `frontend/src/pages/jails/BanUnbanForm.tsx` — `handleBan` and `handleUnban` are synchronous functions that call `onBan(…).then(…).catch(…)`. The returned promise is not awaited and is not assigned to anything.

**Goal:** Convert `handleBan` and `handleUnban` to `async` functions. Add an `isSubmitting` state variable. Set it to `true` before the API call and `false` in a `finally` block. Disable both submit buttons while `isSubmitting` is true. The promise chain can then become a simple `try/catch/finally`.

**Possible traps:**
- If the outer handler is passed to an `onClick` that is typed as `() => void`, TypeScript will not warn about the lost promise; ESLint rule `@typescript-eslint/no-misused-promises` (see TASK-030) would catch this.
- Two separate actions (ban and unban) share the same `isSubmitting` flag; use two flags (`isBanning`, `isUnbanning`) to allow one to proceed while the other is loading.

**Docs changes needed:** None.

**Why:** Without a double-submit guard, clicking "Ban" quickly fires multiple identical API requests. If the `.catch()` branch itself throws, the error is an unhandled rejection that surfaces as a console warning and is invisible to the user.

---

### TASK-004 — `KVEditor` key rename silently overwrites duplicate keys (done)

**Where fixed:** `frontend/src/components/config/KVEditor.tsx`, `frontend/src/components/config/configStyles.ts`, `frontend/src/components/config/__tests__/KVEditor.test.tsx`

**Goal:** Before applying a key rename, check whether `newKey` already exists in `entries`. If it does, show a validation error inline (a `MessageBar` beneath the affected input row or a red border via `validationState="error"` on the Fluent UI `Input`). Block the `onChange` call until the conflict is resolved.

**Possible traps:**
- The component is fully controlled (parent owns the `entries` object), so the error state must be local to `KVEditor` — adding it to the parent's state would be over-engineering.
- While the user is mid-edit (typing the new key name), partial names that match existing keys should not be flagged until the input loses focus, to avoid premature errors.

**Docs changes needed:** None.

**Why:** Silently dropping the value of an existing key when the user renames another key to the same name is destructive data loss with no warning. Config settings that share a key name are semantically invalid in fail2ban anyway.

---

### TASK-005 — `SetupGuard` redirects to `/setup` when the backend is temporarily unreachable (done)

**Where fixed:** `frontend/src/components/SetupGuard.tsx`

**Where found:** `frontend/src/components/SetupGuard.tsx`. When `useSetup` returns `{ loading: false, status: null }` due to a network error, the guard treats this the same as "setup not completed" and redirects to `/setup`.

**Goal:** Destructure `error` from `useSetup` in addition to `status` and `loading`. Add an error branch: if `error` is non-null and `status` is null, render an error card (Fluent UI `MessageBar` with a Retry button) instead of redirecting. Only redirect to `/setup` when `status` is definitively `{ completed: false }`.

**Possible traps:**
- `useSetup` may need to be updated to expose an `error` field if it does not already.
- There is a brief window on initial load where both `loading` and `error` are false and `status` is null; this state must not trigger a redirect.

**Docs changes needed:** None.

**Why:** A transient network hiccup during app startup causes a setup-complete user to be dropped into the setup wizard and potentially overwrite their configuration. This is a silent data-integrity risk.

---

## Security

---

### TASK-006 — No 401 interceptor: expired sessions show broken pages instead of redirecting

**Where found:** `frontend/src/api/client.ts`, `request` function. All non-2xx responses including 401 are thrown as a generic `ApiError`. Consumers render "Failed to load…" messages instead of redirecting.

**Goal:** After the `if (!response.ok)` check, add a dedicated branch: if `response.status === 401`, dispatch a custom DOM event (`bangui:session-expired`) before throwing. In `AuthProvider`, listen for this event with `window.addEventListener` and call the existing logout cleanup logic, then navigate to `/login`. The event approach avoids coupling the API client to React context.

**Possible traps:**
- The event listener must be added and removed inside a `useEffect` in `AuthProvider` to avoid leaks.
- Pages that show a brief error flash before the redirect fires should have their error messages suppressed for `ApiError` with status 401 — wrap this in a shared utility (`isAuthError(err)`).
- If the backend returns 403 (forbidden) rather than 401 for an expired token, the interceptor must also handle that status.

**Docs changes needed:** Update `Docs/Backend-Development.md` to document that 401 responses trigger client-side logout.

**Why:** Currently a user with an expired session sees multiple "Failed to load" error boxes and must manually navigate to `/login`. This is confusing and the broken state can be mistaken for a backend outage.

---

### TASK-007 — Setup page password validation too weak

**Where found:** `frontend/src/pages/SetupPage.tsx`, `validate()` function. Only `masterPassword.length < 8` is checked.

**Goal:** Add minimum complexity rules: at least one uppercase letter, one number, and one special character (e.g. `!@#$%^&*()`). Show a password-strength indicator (a simple four-step progress bar is sufficient) that updates as the user types. Add a validation message per unmet rule rather than a single generic string.

**Possible traps:**
- The complexity rules must match whatever the backend enforces; check `backend/app/routers/setup.py` for server-side validation and keep the two in sync.
- The "confirm password" field comparison must run after the strength check, not instead of it, so both errors can be shown simultaneously.

**Docs changes needed:** Document the password policy in `Docs/Instructions.md`.

**Why:** An 8-character minimum allows trivially weak passwords such as `12345678`. Because this is the single master credential for the entire application, it warrants stronger client-side guidance.

---

## Performance

---

### TASK-008 — `buildBanColumns` and `HISTORY_COLUMNS` recreated on every render

**Where found:**
- `frontend/src/components/BanTable.tsx` — `buildBanColumns(styles)` called unconditionally in the render body.
- `frontend/src/pages/HistoryPage.tsx` — `HISTORY_COLUMNS(onClickIp, styles)` called unconditionally in the render body.

**Goal:** Wrap both column-definition arrays in `useMemo`. In `BanTable`, the dependency is `[styles]`. In `HistoryPage`, the dependency is `[styles]` — the `onClickIp` callback should be wrapped in `useCallback` first so it has a stable reference.

**Possible traps:**
- Fluent UI `createTableColumn` creates objects that are compared by reference inside `DataGrid`; if the array is recreated on every render, `DataGrid` will re-render all cells even when data has not changed.
- The `styles` object from `makeStyles` is stable across renders (Fluent UI guarantees this), so `[styles]` as a dependency is effectively `[]` in practice.

**Docs changes needed:** None.

**Why:** Recreating column definitions every render triggers unnecessary cell re-renders in `DataGrid`, which is one of the heaviest Fluent UI components. On a page with 100 rows and 7 columns, this means 700 unnecessary cell renders per parent re-render.

---

### TASK-009 — `resolveFluentToken` calls `getComputedStyle` on every render

**Where found:** `frontend/src/utils/chartTheme.ts`, `resolveFluentToken` function. Called 2–3 times per render in `BanTrendChart`, `TopCountriesPieChart`, `TopCountriesBarChart`, and `JailDistributionChart`.

**Goal:** In each chart component that calls `resolveFluentToken`, wrap the calls in `useMemo` with an empty dependency array `[]` (the theme never changes at runtime in the current implementation). When dark mode support is added (TASK-015), the dependency should change to the active theme object.

**Possible traps:**
- `resolveFluentToken` reads a CSS custom property from a DOM element; calling it at module level (outside a component) would read before the `FluentProvider` has injected its tokens, returning an empty string.
- The function itself should not be memoized globally — only the results per component, since each component may be mounted in a different theming context in tests.

**Docs changes needed:** None.

**Why:** `getComputedStyle` on every render is a forced style recalculation. On low-end hardware or when many charts are visible simultaneously, this measurably degrades frame rate.

---

### TASK-010 — No code splitting: all pages bundled into the main chunk

**Where found:** `frontend/src/App.tsx` — all page imports are static (`import { DashboardPage } from "./pages/DashboardPage"`). `frontend/vite.config.ts` has no `build.rollupOptions.manualChunks`.

**Goal:** Convert all seven page imports in `App.tsx` to `React.lazy(() => import(…))`. Wrap the `<Routes>` block in `<React.Suspense fallback={<Spinner size="large" />}>`. In `vite.config.ts`, add `manualChunks` to split `react`/`react-dom`, Fluent UI, Recharts, and the D3/TopoJSON geo stack into separate vendor chunks.

**Possible traps:**
- `React.lazy` requires the module to have a default export; all current pages use named exports. Either add a re-export default in each page file, or use `import(…).then(m => ({ default: m.PageName }))` in the lazy call.
- The `<Suspense>` fallback renders during navigation; place it inside `<BrowserRouter>` but outside `<AuthProvider>` so the spinner is not blocked by auth context loading.
- The D3/TopoJSON data (`world-atlas/countries-110m.json`) is a large JSON file (~100 KB gzipped) and should be placed in the `geo` chunk alongside `d3-geo` and `topojson-client`.

**Docs changes needed:** None.

**Why:** Every user who opens the dashboard must download, parse, and execute JavaScript for the map, config editor, and blocklist pages — even if they never visit those routes. Code splitting allows the initial load to be significantly smaller.

---

### TASK-011 — No `React.memo` on any heavy component

**Where found:** Every component in `frontend/src/components/` — zero uses of `React.memo` exist in the codebase.

**Goal:** Apply `React.memo` to the components whose props rarely change but whose render is expensive: `WorldMap`, `BanTrendChart`, `JailDistributionChart`, `TopCountriesPieChart`, `TopCountriesBarChart`, `BanTable`, and `ServerStatusBar`. Use the default shallow-equality comparator for props that are primitives. For `WorldMap`, provide a custom comparator that compares the `countries` record by key count and values (shallow object comparison).

**Possible traps:**
- Inline function props (e.g. `onSelectCountry={() => …}`) will always be a new reference and defeat `React.memo`. Ensure all callback props passed to memoized components are wrapped in `useCallback` at the call site.
- Inline `style={{…}}` objects passed as props also defeat memoization; these must be moved to `makeStyles` first (see TASK-018).
- `React.memo` does not help if the component itself calls an expensive hook that always returns a new object.

**Docs changes needed:** None.

**Why:** `DashboardPage` re-renders whenever the 30-second server status poll fires, causing all five chart components and the ban table to re-render even though the filter state has not changed.

---

### TASK-012 — `useMapData` sets `loading=true` before the debounce fires

**Where found:** `frontend/src/hooks/useMapData.ts`, `load` callback — `setLoading(true)` is called at the top of `load`, but the actual fetch is deferred inside a `setTimeout` of 300 ms.

**Goal:** Move `setLoading(true)` inside the `setTimeout` callback, immediately before the `abortRef.current?.abort()` call. This ensures the spinner only appears when a real network request is about to start.

**Possible traps:**
- The caller (`MapPage`) uses `loading` to fade the map table with `opacity: 0.5`. There will now be a 300 ms window where filters have changed but the table is still showing old data at full opacity; this is acceptable and more correct than showing a spinner immediately.
- If the component unmounts during the debounce window, the `clearTimeout` in the cleanup must run; verify the existing `useEffect` cleanup correctly returns `clearTimeout`.

**Docs changes needed:** None.

**Why:** Users see a loading spinner for 300 ms before any network activity has started, which makes the UI feel slower than it is.

---

## Architecture / Code Quality

---

### TASK-013 — Consolidate `api/config.ts` and `api/file_config.ts` duplicate functions

**Where found:** Both `frontend/src/api/config.ts` and `frontend/src/api/file_config.ts` export identical functions: `fetchJailConfigFiles`, `createJailConfigFile`, `fetchJailConfigFileContent`, `updateJailConfigFile`, `setJailConfigFileEnabled`, `fetchFilterFiles`, `fetchFilterFile`, `updateFilterFile`, `createFilterFile`, `fetchActionFiles`, `fetchActionFile`, `updateActionFile`, `createActionFile`.

**Goal:** Delete `api/file_config.ts`. Update the four files that import from it (`hooks/useFilterRawFile.ts`, `hooks/useActionRawFile.ts`, `components/config/JailFilesTab.tsx`, `components/config/ExportTab.tsx`) to import the same functions from `api/config.ts` instead. Verify that both modules currently export functions with identical signatures before the merge; resolve any differences first.

**Possible traps:**
- `file_config.ts` lacks `AbortSignal` parameters on all its functions, while the corresponding functions in `config.ts` may have them. After migration, all callers must be updated to pass the signal.
- Running a global search for `from.*file_config` after the migration should return zero results; add this as a CI check via ESLint `no-restricted-imports` if desired.

**Docs changes needed:** None.

**Why:** Two modules exporting the same API functions will drift apart over time. Any bug fix or new feature applied to one will silently not apply to the other.

---

### TASK-014 — Add `AbortSignal` to all API functions missing it

**Where found:**
- `frontend/src/api/dashboard.ts` — `fetchServerStatus`, `fetchBansByJail`, `fetchBanTrend` have no signal parameter.
- `frontend/src/api/jails.ts` — `fetchJailBannedIps` (and others) have no signal parameter.
- `frontend/src/api/blocklist.ts` — `fetchSchedule` (used by the polling hook) has no signal parameter.

**Goal:** Add `signal?: AbortSignal` as the last parameter to every `get`/`post`/`put`/`del` wrapper call in these modules. The pattern is already established in `api/config.ts` and `api/map.ts` and should be replicated uniformly. After adding the parameters, update the consuming hooks to pass their `abortRef.current.signal`.

**Possible traps:**
- Functions used by polling hooks (`fetchServerStatus` in `useServerStatus`, `fetchSchedule` in `useBlocklistStatus`) must pass the signal from a `useRef<AbortController>` rather than from a `useEffect`-local controller, because the poll fires outside the effect's lifecycle.
- Adding a signal parameter is non-breaking (it is optional), so no consumers need to change immediately; they can be updated incrementally.

**Docs changes needed:** None.

**Why:** Without a signal, requests fired by polling hooks cannot be cancelled when the component unmounts, leaking network activity and potentially updating state on unmounted components (React warning).

---

### TASK-015 — Standardise AbortController pattern across all hooks

**Where found:** Three different patterns exist in `frontend/src/hooks/`:
1. `useRef<AbortController | null>` with manual abort before each fetch (correct — used in `useActiveBans`, `useActionList`).
2. Inline `AbortController` created inside `useEffect` and aborted in cleanup (acceptable for single-fetch effects — used in `useSchedule`).
3. No AbortController at all (incorrect — used in `useJailBannedIps`, `useBlocklistStatus`).

**Goal:** Hooks that expose a manual `refresh()` function must use pattern 1 (the `useRef` approach), because the abort must survive across effect re-runs. Hooks that only fetch once on mount and never need manual refresh may use pattern 2. Remove all instances of pattern 3. Document the chosen conventions in a comment block at the top of a new file `frontend/src/hooks/README.md` (or inline in `fetchError.ts`).

**Possible traps:**
- When migrating `useBlocklistStatus` from a boolean `cancelled` flag to an `AbortController`, the underlying `fetchSchedule` call must first accept a signal (TASK-014).
- Some hooks use both a debounce timer and an abort controller; the abort must cancel the in-flight request, while the timer cancellation prevents a new request from starting. These are independent concerns and should not be conflated.

**Docs changes needed:** Add an ADR (Architecture Decision Record) or hook convention note to `Docs/Backend-Development.md` or a new `Docs/Frontend-Development.md`.

**Why:** Inconsistent patterns make it hard for a new developer to know which approach to follow. The presence of pattern 3 causes memory leaks and stale-state React warnings in production.

---

### TASK-016 — Extract generic `useListData` hook to eliminate duplicated fetch-list pattern

**Where found:** `frontend/src/hooks/useActionList.ts`, `useFilterList.ts`, `useJailConfigs.ts`, and `useBlocklists.ts` each contain ~40 lines of nearly identical code: `useState` for data/loading/error, a `useRef<AbortController>`, a `refresh` callback with abort-create-set-loading-fetch-set logic, and a `useEffect` that calls `refresh` on mount.

**Goal:** Create `frontend/src/hooks/useListData.ts` that exports a generic hook:
```ts
function useListData<TResponse, TItem>(options: {
  fetcher: (signal: AbortSignal) => Promise<TResponse>;
  selector: (response: TResponse) => TItem[];
  errorMessage: string;
}): { items: TItem[]; loading: boolean; error: string | null; refresh: () => void }
```
Rewrite the four hooks listed above as thin wrappers around `useListData`.

**Possible traps:**
- Some hooks (e.g. `useJailConfigs`) expose additional operations like `reload` and `update` beyond just listing. These should remain in their own hook and call `useListData` only for the list-loading portion.
- The `fetcher` function must accept a signal; ensure all underlying API functions have been updated (TASK-014) before replacing the hook internals.
- Generic hooks with complex type parameters can confuse TypeScript's inference; provide explicit type arguments at each call site to avoid `unknown` leaking out.

**Docs changes needed:** None.

**Why:** Four copies of identical logic means bug fixes must be applied four times. The pattern is stable enough to abstract — every list hook has the same contract.

---

### TASK-017 — Move `source` derivation out of page components

**Where found:** `frontend/src/pages/DashboardPage.tsx` and `frontend/src/pages/MapPage.tsx` both contain the identical line:
```ts
const source = timeRange === "24h" ? "fail2ban" : "archive";
```

**Goal:** Create a utility function `getDataSource(timeRange: TimeRange): "fail2ban" | "archive"` in `frontend/src/utils/constants.ts` (or a new `queryUtils.ts` entry). Import and call it in both pages. Remove the inline ternary.

**Possible traps:**
- This rule ("24h uses live fail2ban data, everything else uses the archive") is a backend contract, not a UI preference. If the backend ever adds a new time range or changes the source mapping, both places must be updated in sync; centralising it makes that change a one-liner.
- The function name should make the business rule obvious — `getLiveDataSource` is more descriptive than `getDataSource`.

**Docs changes needed:** Document the time range → source mapping in `Docs/Backend-Development.md` if not already described.

**Why:** Duplicated business logic in UI components is fragile. If the rule changes or a new time range is added, it is easy to update one page and miss the other.

---

### TASK-018 — Move inline `style={{…}}` objects to `makeStyles`

**Where found:** 30+ instances across `frontend/src/pages/MapPage.tsx`, `frontend/src/pages/map/MapBansTable.tsx`, `frontend/src/pages/DashboardPage.tsx`, `frontend/src/pages/jail/PatternsSection.tsx`, `frontend/src/pages/jails/BanUnbanForm.tsx`, `frontend/src/layouts/MainLayout.tsx`, and others.

**Goal:** For each instance of `style={{…}}` that is not dynamic (i.e. the style values are constants or token references, not computed from props), move the declaration into the nearest `makeStyles` call. Dynamic styles that depend on runtime values (e.g. `style={{ opacity: loading ? 0.5 : 1 }}`) should remain inline or be converted to conditional `mergeClasses` with two separate class definitions.

**Possible traps:**
- The `MainLayout` logout button uses `style={{ width: "100%", justifyContent: collapsed ? "center" : "flex-start" }}` — this is dynamic and must become two `makeStyles` entries toggled with `mergeClasses`.
- The `WorldMap` country `<path>` uses CSS custom properties (`--country-fill`) as inline styles for CSS variable injection; this is intentional and should not be changed.
- Running ESLint with `@typescript-eslint/no-unsafe-assignment` can catch some object literals, but a manual pass is more reliable.

**Docs changes needed:** None.

**Why:** Inline style objects create a new object reference on every render, defeating `React.memo` (TASK-011) and preventing the browser from reusing cached styles. Fluent UI's `makeStyles` uses atomic CSS which is far more cache-efficient.

---

### TASK-019 — Replace index keys with stable keys in editable lists

**Where found:** 19 instances identified, including:
- `frontend/src/components/config/StringListEditor.tsx` line 34 — `key={index}` on editable `Input` rows.
- `frontend/src/components/config/RegexList.tsx` line 66 — `key={i}` on editable regex rows.
- `frontend/src/components/config/JailsTab.tsx` lines 384, 714, 726 — `key={i}` and `key={idx}`.
- `frontend/src/components/blocklist/BlocklistScheduleSection.tsx` line 147 — `key={i}` on `<option>` elements.

**Goal:** For editable list components (`StringListEditor`, `RegexList`), each item needs a stable identity. Since the items are plain strings with no natural ID, generate a stable ID when an item is added (e.g. a running counter stored in a `useRef`, or a `crypto.randomUUID()` call). Store the list as `{ id: string; value: string }[]` internally and emit `string[]` via `onChange`. For static read-only lists (`<option>` elements, skeleton items), index keys are acceptable and do not need to be changed.

**Possible traps:**
- Changing the internal representation of `StringListEditor` from `string[]` to `{ id, value }[]` changes the component's internal state shape but not its public interface (`items: string[]`, `onChange: (next: string[]) => void`), so callers are unaffected.
- If two items have the same string value (which is valid in some config fields), using the value itself as the key will cause the same de-duplication problem as using indices. Only a generated ID is reliably stable.

**Docs changes needed:** None.

**Why:** When React sees `key={index}` on input elements in an editable list and an item is deleted from the middle, it reuses the DOM node at that index rather than the one associated with the deleted item. The result is that the wrong input field appears to clear or change value.

---

### TASK-020 — Standardise `loading` vs `isLoading` naming across hooks

**Where found:** Dashboard and country-data hooks (`useBans`, `useBanTrend`, `useDashboardCountryData`) return `isLoading`. All other hooks (`useActiveBans`, `useJails`, `useBlocklists`, `useFilterList`, `useActionList`, etc.) return `loading`.

**Goal:** Standardise on `loading: boolean` throughout. Update the three hooks that use `isLoading` to rename the field. Update all call sites that destructure `isLoading` from those hooks to use `loading`.

**Possible traps:**
- A global search-and-replace on `isLoading` will also match any variable named `isLoading` that is locally derived (e.g. `const isLoading = loading && …`); review each replacement manually.
- The rename affects destructuring at the call site, not just the hook's return object; both must be updated atomically.

**Docs changes needed:** None.

**Why:** Inconsistent naming forces developers to check the hook signature every time they use one, and increases the chance of misreading state (treating `loading` as `isLoading` when they have different semantics in some codebase convention).

---

### TASK-021 — Persist sidebar collapse preference to `localStorage`

**Where found:** `frontend/src/layouts/MainLayout.tsx` — `useState(() => window.innerWidth < 640)` initialises collapsed state but never saves the user's toggle preference.

**Goal:** Replace the `useState` initialiser with one that reads from `localStorage` first (`localStorage.getItem("bangui_sidebar_collapsed")`), falling back to the `window.innerWidth < 640` check. Add a `useEffect` that writes to `localStorage` whenever `collapsed` changes.

**Possible traps:**
- `localStorage` access can throw in private-browsing mode in some browsers; wrap it in a try/catch and fall back silently to the width-based default.
- The existing `useEffect` that watches `matchMedia` for viewport changes should continue to work but should not override a persisted preference; only auto-collapse/expand when there is no saved preference.

**Docs changes needed:** None.

**Why:** Users who prefer the expanded sidebar on a wide monitor should not have to re-expand it on every page refresh.

---

### TASK-022 — Add dark mode support and OS preference detection

**Where found:** `frontend/src/App.tsx` line 25 — `<FluentProvider theme={lightTheme}>` is hardcoded. `frontend/src/theme/customTheme.ts` already exports `darkTheme` but it is never used.

**Goal:** In `App.tsx` (or a new `ThemeProvider` wrapper), initialise theme from `localStorage.getItem("bangui_theme")`. If no preference is stored, fall back to `window.matchMedia("(prefers-color-scheme: dark)").matches`. Add a toggle button in `MainLayout`'s sidebar footer that switches between light and dark and persists the choice. Pass the active theme to `<FluentProvider>`.

**Possible traps:**
- `resolveFluentToken` (used by chart components) reads CSS custom properties injected by `FluentProvider`. When the theme switches, the memoized color values from TASK-009 must be invalidated; add `theme` as a dependency to the `useMemo` calls in chart components.
- The `matchMedia` listener for OS-level theme changes should update app state only when no explicit user preference has been saved.
- The toggle button icon should be accessible — use `SunRegular` / `WeatherMoonRegular` from `@fluentui/react-icons` with an `aria-label`.

**Docs changes needed:** Update `Docs/Web-Design.md` to mention dark theme support.

**Why:** Dark mode is a standard accessibility and user-comfort feature. The infrastructure (`darkTheme`) already exists in the codebase; wiring it up is the only remaining work.

---

### TASK-023 — Replace loose `string` types with union types in config models

**Where found:** `frontend/src/types/config.ts` — `JailConfig.use_dns`, `JailConfig.log_encoding`, `JailConfig.backend`, and several action/filter fields are typed as plain `string` despite having a fixed set of valid values defined by fail2ban.

**Goal:** Define union types for these fields:
```ts
type DNSMode = "yes" | "warn" | "no" | "raw";
type LogEncoding = "auto" | "ascii" | "utf-8" | "latin-1";
type BackendType = "auto" | "polling" | "pyinotify" | "systemd" | "gamin";
```
Update `JailConfig` (and the corresponding `JailConfigUpdate` patch type) to use these. Update any `<Select>` or `<Input>` components that render these fields to use the narrowed types.

**Possible traps:**
- The backend Pydantic models should be the source of truth. Check `backend/app/models/` to ensure the union type lists match the actual validation constraints before committing the frontend types.
- If the backend returns a value not in the union (e.g. a custom backend plugin), TypeScript will flag it as a type error. A `string & {}` escape hatch or a broader union including `string` for unknown values may be needed.

**Docs changes needed:** None.

**Why:** Plain `string` types allow invalid values to be passed to API endpoints and rendered in select dropdowns without compile-time errors. Union types shift error detection from runtime to build time.

---

### TASK-024 — Add `useReducer` to `ServerTab` and `ConfFilesTab`

**Where found:**
- `frontend/src/components/config/ServerTab.tsx` — 11 `useState` calls managing logLevel, logTarget, dbPurgeAge, dbMaxMatches, flushing, msg, isReloading, isRestarting, and three map threshold fields.
- `frontend/src/components/config/ConfFilesTab.tsx` — 10 `useState` calls with a subtle synchronisation dependency between `contents` and `editedContents`.

**Goal:** For `ServerTab`, define a `ServerTabState` type and a `serverTabReducer`. All action buttons (flush, reload, restart) dispatch `{ type: "FLUSH_START" }` / `{ type: "FLUSH_SUCCESS", result }` / `{ type: "FLUSH_ERROR", message }` actions that update related state atomically. For `ConfFilesTab`, use `useReducer` to ensure `contents` and `editedContents` are always updated in the same dispatch to prevent one being stale relative to the other.

**Possible traps:**
- The `useAutoSave` hook in `ServerTab` subscribes to the `updatePayload` memo, which derives values from state. If the reducer replaces the individual `useState` calls, the memoisation dependencies must be updated to read from `state.logLevel` etc.
- `useReducer` does not replace `useRef` for the `abortRef` pattern; keep abort controllers in refs, not in reducer state.

**Docs changes needed:** None.

**Why:** When multiple state values must change in response to a single action (e.g. `isFlushing = true` and `msg = null` both change when flush starts), `useState` requires two separate `set` calls that fire two renders. `useReducer` batches them into one.

---

### TASK-025 — Consolidate barrel re-export files or replace with a single `index.ts`

**Where found:** `frontend/src/hooks/useJails.ts`, `hooks/useConfig.ts`, and `hooks/useBlocklist.ts` are pure re-export files that do nothing except forward exports from other hook files.

**Goal:** Either (a) delete the three barrel files and update all 20+ import sites to use direct imports (e.g. `from "../hooks/useJailList"` instead of `from "../hooks/useJails"`), or (b) consolidate them into a single `frontend/src/hooks/index.ts` that re-exports everything from the hooks folder, and update all imports to `from "../hooks"`. Option (a) is preferred because it makes the import graph explicit and avoids accidentally importing unused hook modules.

**Possible traps:**
- Some consumers import both `useJails` (the list hook) and the `useJails` name from `useJails.ts` (the barrel); after refactoring, the hook function formerly exported as `useJails` from the barrel now lives in `useJailList.ts` as `useJails` — the function name is the same, only the file path changes.
- A global find-replace from `from "../hooks/useJails"` to `from "../hooks/useJailList"` should cover most cases, but verify by running the TypeScript compiler with `--noEmit` afterwards.

**Docs changes needed:** None.

**Why:** Barrel files that just re-export create an extra indirection layer that obscures the real source of a function. IDEs sometimes fail to navigate through barrel files correctly, breaking "Go to Definition".

---

### TASK-026 — Props drilling: introduce a Dashboard filter context

**Where found:** `frontend/src/pages/DashboardPage.tsx` passes `timeRange`, `originFilter`, and `source` individually as props to `BanTrendChart`, `BanTable`, `TopCountriesPieChart`, `TopCountriesBarChart`, and `JailDistributionChart`.

**Goal:** Create `frontend/src/providers/DashboardFilterProvider.tsx` that wraps the dashboard content and exposes `{ timeRange, originFilter, source, setTimeRange, setOriginFilter }` via context. Individual chart components read the filter values from the context instead of receiving them as props. `DashboardFilterBar` writes to the context rather than receiving callback props.

**Possible traps:**
- Introducing a context for the dashboard is only justified if at least one deeply-nested component (more than one level) needs access; currently all consumers are direct children of `DashboardPage`, making this a marginal improvement. Evaluate whether the added complexity is worth it before starting.
- If any chart is used outside the dashboard (e.g. in a modal or embedded view), removing the props would break that usage; keep props as optional overrides.

**Docs changes needed:** None.

**Why:** Passing the same three values to five sibling components is repetitive and makes each component's prop signature larger than necessary. A context also makes it easier to add new filter dimensions in the future.

---

## Developer Experience / Tooling

---

### TASK-027 — Add missing ESLint rules

**Where found:** `frontend/eslint.config.ts` — only `react-hooks/recommended`, `no-explicit-any`, `explicit-function-return-type`, and `no-unused-vars` are enabled. Several important rules are absent.

**Goal:** Add the following rules to `eslint.config.ts`:
- `@typescript-eslint/no-floating-promises: "error"` — catches unawaited promises like those in `BanUnbanForm` (TASK-003).
- `@typescript-eslint/no-misused-promises: "error"` — catches async functions passed as event handlers without `void`.
- `react/no-array-index-key: "warn"` — requires adding `eslint-plugin-react` to `devDependencies`.
- `react/no-unstable-nested-components: "error"` — prevents component definitions inside render functions.
- `@typescript-eslint/switch-exhaustiveness-check: "error"` — ensures `switch` on a union type covers all cases.

Install `eslint-plugin-react` if not already present. Run `npm run lint` after each rule addition and fix all violations before enabling the rule at `"error"` level.

**Possible traps:**
- `no-floating-promises` will flag every `void someAsync()` call; add `void` as an explicit acknowledgement only where the promise is intentionally fire-and-forget (e.g. the `handleLogout` `onClick`).
- `no-misused-promises` has a known false-positive with Fluent UI's `onCheckedChange` prop, which passes a React synthetic event; configure the rule to ignore `checkAsync` patterns if needed.
- Adding `eslint-plugin-react` requires configuring the plugin with `{ version: "detect" }` in settings.

**Docs changes needed:** None.

**Why:** Several of the bugs identified in this task list (floating promises, index keys, unstable nested components) would be caught automatically at lint time with these rules. Prevention is cheaper than code review.

---

### TASK-028 — Add security and SEO meta tags to `index.html`

**Where found:** `frontend/index.html` — the file contains only `charset`, `viewport`, and `title` meta tags.

**Goal:** Add the following:
- `<link rel="icon" href="/favicon.ico" />` (create a simple icon asset in `frontend/public/`).
- `<meta name="description" content="BanGUI — fail2ban management interface." />`
- `<meta name="theme-color" content="#0F6CBD" />` (matches the brand primary colour).
- `<meta name="robots" content="noindex, nofollow" />` — this is a private admin interface; it should not be indexed by search engines.
- `<noscript>BanGUI requires JavaScript to run.</noscript>`

Do **not** add a Content-Security-Policy meta tag; CSP should be set as an HTTP response header from the nginx configuration (`Docker/nginx.conf`) to prevent it being stripped by proxies.

**Possible traps:**
- The `theme-color` meta tag is a hint to mobile browsers for the browser chrome colour; it should match `lightTheme` brand primary. When dark mode (TASK-022) is implemented, it can be updated dynamically via JavaScript.
- Adding `noindex` is important because BanGUI runs on a local network and may be accidentally accessible via a public IP; preventing search engine indexing adds a minor layer of security through obscurity.

**Docs changes needed:** None.

**Why:** Missing `noindex` on an admin interface risks it being discovered via search engine crawlers if accidentally exposed. Missing `<noscript>` leaves users with JS disabled seeing a completely blank page with no explanation.

---

### TASK-029 — Configure Vite proxy target via environment variable

**Where found:** `frontend/vite.config.ts` line 31 — `target: "http://backend:8000"` is a hardcoded Docker service hostname.

**Goal:** Change the proxy target to `process.env["VITE_BACKEND_URL"] ?? "http://backend:8000"`. Document the variable in a `frontend/.env.example` file with the value `VITE_BACKEND_URL=http://localhost:8000` for local development without Docker.

**Possible traps:**
- Vite's `loadEnv` helper can be used inside `defineConfig` to read `.env` files, but `process.env` also works for variables set in the shell or Docker Compose environment. Either approach is acceptable.
- The fallback `"http://backend:8000"` ensures the Docker compose stack continues to work without any changes.

**Docs changes needed:** Add `VITE_BACKEND_URL` to `Docs/Backend-Development.md` or a new `Docs/Frontend-Development.md` setup guide.

**Why:** Developers running the backend locally on a different port or hostname cannot run `npm run dev` without modifying the committed `vite.config.ts`.

---

### TASK-030 — Add manual chunk splitting to `vite.config.ts`

**Where found:** `frontend/vite.config.ts` — `build` section is absent entirely; Rollup uses its default single-chunk strategy.

**Goal:** Add a `build.rollupOptions.output.manualChunks` function that groups modules into the following chunks: `react-vendor` (react, react-dom, react-router-dom), `ui-vendor` (@fluentui/react-components, @fluentui/react-icons), `chart-vendor` (recharts), `geo-vendor` (d3-geo, topojson-client, world-atlas). This is complementary to TASK-010 (lazy loading) but independent — chunk splitting improves cache efficiency even without lazy loading.

**Possible traps:**
- `manualChunks` can be a function `(id: string) => string | undefined`; use `id.includes("node_modules/react")` style checks rather than exact package names to handle monorepo sub-packages.
- Fluent UI has many sub-packages (`@fluentui/react-components`, `@fluentui/merge-styles`, `@griffel/react`, etc.); group all `@fluentui` and `@griffel` packages into the `ui-vendor` chunk.
- The `geo-vendor` chunk containing the world atlas JSON will be large (~100 KB gzipped) and only needed on the `/map` route; this makes TASK-010 (lazy loading) particularly important for the map page.

**Docs changes needed:** None.

**Why:** Without chunk splitting, all vendor code is in a single bundle. When any vendor dependency is updated, the user must re-download the entire bundle. Splitting by vendor allows browsers to cache stable chunks across deployments.