# HAC-29 Browse And Search Baseline Stabilization

Date: 2026-03-31 UTC
Owner: Senior Frontend Engineer
Issue: HAC-29
Scope: `desktop-v1` browse/search hardening only

## Context

- `apps/desktop` already supports live browse pagination and DB-scoped search, but the visible search surface still leaves a few operator-facing gaps:
  - no explicit feedback about whether input is treated as substring matching or raw Redis pattern matching
  - no bounded clear action from the browser header
  - one generic empty-state sentence was doing too much work across loading, empty DB, and filtered no-match cases
- Current phase must stay inside browse/search hardening and must not expand into Add Key, delete execution, non-string editing, or visual refresh work.

## Decision

- Keep the existing workspace layout and backend contracts unchanged.
- Extract search-state interpretation into a shared frontend helper so visible UI copy and backend browse requests use the same source of truth.
- Improve only the key-browser and inspector feedback layer:
  - explicit search mode label (`all keys`, `contains`, `pattern`)
  - concise result-status message under the search input
  - clear action for returning to normal browse state
  - distinct empty-state copy for loading, empty DB, and no-match search results

## Implemented Slice

1. Added `apps/desktop/src/lib/key-browser-search.ts` plus focused tests for:
   - blank input vs contains search vs raw wildcard-pattern classification
   - backend request-pattern generation
   - browser status and empty-state copy
   - inspector empty-state copy aligned to the same search state
2. Updated `apps/desktop/src/App.tsx` to:
   - consume the shared search helper for live browse requests
   - render a visible search-mode badge and status line under the browser search field
   - expose a bounded clear action when search input is active
   - keep browser and inspector empty-state feedback aligned
3. Updated `docs/frontend-workspace-baseline.md` so the shared acceptance language now includes the search-mode feedback, clear action, and differentiated empty states.

## Verification

- `pnpm --filter @redis-gui/desktop test` -> pass (9 files / 33 tests)
- `pnpm run desktop:build` -> pass

## Remaining Risk

- This environment still cannot launch a visible Tauri window, so final GUI-level browse/search smoke evidence remains a QA follow-up even though the frontend code and build path are green.