# Deferred: enhancements and nice-to-haves

Smaller, mostly independent improvements. None are blocking; ordered roughly by
value.

## Undo / redo

- Add a history stack around the reducer (past, present, future) with
  Cmd/Ctrl+Z and Shift+Cmd/Ctrl+Z.
- The catch: continuous actions (setCrop, setZoom, setRotation from the slider)
  fire many times. Coalesce them - snapshot history only on discrete commits
  (template, aspect, rotateBy, flip, effect preset, format, quality) and treat a
  drag as a single entry (push on interaction start, replace until it settles).
- Medium effort; the coalescing is the only subtle part.

## Persist preferences

- Save format, quality, maxEdge, and last template to localStorage; restore on
  load. Do not persist the image.
- Guard for SSR/StrictMode (read in an effect, not during render).
- Small effort.

## Batch / favicon set export

- The brief mentions a "favicon set". Export several sizes at once
  (16, 32, 48, 180, 192, 512) as a zip.
- Add a tiny zip writer (for example fflate) in the worker; or trigger multiple
  downloads (some browsers throttle multiple downloads, so zip is nicer).
- Small-to-medium effort.

## Richer adjustments (need real pixel math)

- Exposure, temperature/tint (white balance), sharpen, vignette, levels/curves.
- CSS/canvas filter cannot do these well. Two options:
  - Implement them with wasm-vips operations in the worker (vips has sharpen,
    linear, recomb, etc.), or
  - Write small WebGL/Canvas shaders for live preview plus a matching worker pass.
- The live-preview-must-match-export rule makes this non-trivial; vips-only
  (export-time) is simpler but loses live preview for those controls.
- Medium-to-large effort.

## Explicit output dimensions

- Beyond the current "limit long edge" cap, allow typing an exact width/height
  for custom crops (with an aspect lock to avoid stretching). The worker already
  accepts an output size; this is mostly UI.
- Small effort.

## User-defined / managed templates

- Let users create, edit, reorder, and persist their own templates (localStorage
  or export/import JSON). Templates are already plain data, so the model fits.
- Medium effort.

## Live size estimate

- Show an estimated output file size before export. Requires a debounced encode
  of a downscaled proxy (full encode per keystroke is too expensive).
- Medium effort; easy to get wrong on cost.

## Committed E2E test suite

- The app was verified with Playwright, but no suite is committed because the
  browser binary path is environment-specific (launch used an explicit
  executablePath). To commit a suite: add @playwright/test as a dev dependency,
  add a config that resolves the browser normally (npx playwright install), and
  port the scenarios from VERIFICATION.md. Wire a "test:e2e" script.
- Small-to-medium effort.

## Mobile polish

- The responsive drawers work, but could add an explicit close button inside
  each drawer and swipe-to-dismiss. Small effort.

## Misc

- i18n of UI strings.
- A "compare" press-and-hold to temporarily disable adjustments/crop.
- Keyboard nudge of the crop with arrow keys.
- Drag-over highlight on the whole editor (not just the empty dropzone).