BF-011-BF-014: Improve tool discovery search, filters, favorites, and recents - #209

Summary:

• Added a shared weighted tool-search metadata layer with aliases, task synonyms, category terms, popularity boosts, and localized aliases for zh-CN, zh-TW, ja, ko, de, and fr.
• Wired Command Palette and All Tools to the same scoring path, including privacy-safe query-length/count analytics only.
• Reworked All Tools filters into a mobile drawer with active counts, removable chips, clear/reset controls, tags, empty-state suggestions, and no persisted free-text query state.
• Added local-only Favorites and Recent Tools surfaces on All Tools and tool pages, with timestamped storage records limited to tool IDs.
• Updated Trust Center / Privacy Policy copy across supported locales to document favorites, recents, timestamps, and forbidden payload persistence.

Issues:

• Closes [P1] BF-011: Improve search ranking with popularity, category relevance, and task synonyms #116
• Closes [P1] BF-012: Add localized search support for Chinese, Japanese, Korean, German, and French queries #118
• Closes [P1] BF-013: Redesign All Tools filters for mobile and complex multi-filter workflows #120
• Closes [P1] BF-014: Add local-only Favorites and Recent Tools without storing tool payloads #122

Implementation details:

• Search: src/core/search/tool-search-metadata.ts centralizes weighted metadata and localized aliases; scoreToolSearch applies popularity, favorite, and recent bonuses without making those metadata markers searchable.
• Routing/metadata: No new routes or indexable filtered pages were added. All Tools continues using safe URL params for category/filter state, and free-text search is not written to the URL.
• Storage: src/core/storage/tool-discovery-state.ts migrates legacy arrays to { toolKey, updatedAt } records and keeps read helpers returning key lists for existing UI code.
• UI: All Tools gets a 390px-friendly filter drawer and hydrated personalized favorites/recent panels; tool pages get a favorite button near the trust header.
• Tests: Added real generated-index search ranking regressions, localized query regressions, storage-shape assertions, All Tools drawer/favorite tests, and a guard against payload-shaped favorite/recent persistence.
• Content: Updated Privacy Policy and Trust Center storage wording in en, zh-CN, zh-TW, ja, ko, de, and fr.

Privacy review:

• No tool payloads are stored by default.
• Favorites and recents store only tool IDs plus timestamps in localStorage.
• No tool input, output, files, logs, tokens, secrets, full URLs, or search query text are persisted.
• No new external request runs automatically.
• Analytics remains disabled by default and search telemetry passes only language, query-length bucket, result count, and source page.
• localStorage/sessionStorage/IndexedDB/cache usage was reviewed; this PR only changes the approved localStorage discovery state shape.

SEO review:

• Clean /[lang]/all-tools remains indexable.
• Filter states remain parameterized and noindex/canonical behavior from BF-002 is unchanged.
• Free-text search is intentionally not serialized into URLs.
• Sitemap behavior is unchanged; no parameter pages or new routes were added.
• Hreflang/canonical behavior was covered by the build post-checks.

Accessibility review:

• Mobile filter drawer uses role="dialog", aria-modal, explicit close controls, and reachable clear/done actions.
• Favorite controls use aria-label and aria-pressed.
• Existing visible focus ring classes were preserved on interactive controls.
• Dynamic personalized sections hydrate after client storage sync and remain keyboard reachable.

Tests:

• Targeted checks run:
  • npm run test -- tests/unit/command-search.test.ts tests/unit/tool-discovery-state.test.ts tests/component/all-tools-discovery.test.tsx tests/component/route-page-chrome.test.tsx
  • npm run test -- tests/unit/command-search.test.ts tests/unit/tool-discovery-state.test.ts tests/component/all-tools-discovery.test.tsx tests/component/route-page-chrome.test.tsx tests/guards/sensitive-storage-audit.test.ts tests/guards/command-palette-search-guard.test.ts tests/guards/all-tools-route-guard.test.ts
  • npm run test -- tests/unit/command-search.test.ts tests/component/all-tools-discovery.test.tsx tests/guards/command-palette-search-guard.test.ts tests/guards/sensitive-storage-audit.test.ts
  • npm run test -- tests/component/all-tools-discovery.test.tsx tests/unit/tool-discovery-state.test.ts
  • npm run test -- tests/unit/command-search.test.ts tests/component/all-tools-discovery.test.tsx tests/guards/sensitive-storage-audit.test.ts
  • npm run check:i18n
  • npm run check:types -- --pretty false
  • npm run lint -- src/core/search/command-search.ts src/core/search/tool-search-metadata.ts src/core/storage/tool-discovery-state.ts src/features/tool-discovery/all-tools-discovery.tsx src/components/layout/command-palette.tsx src/components/layout/route-page-chrome.tsx src/components/layout/tool-favorite-control.tsx tests/unit/command-search.test.ts tests/unit/tool-discovery-state.test.ts tests/component/all-tools-discovery.test.tsx tests/component/route-page-chrome.test.tsx tests/guards/sensitive-storage-audit.test.ts
  • npm run lint -- src/components/layout/command-palette.tsx src/core/search/command-search.ts src/core/search/tool-search-metadata.ts tests/unit/command-search.test.ts tests/guards/command-palette-search-guard.test.ts
  • npm run lint -- 'src/app/[lang]/all-tools/page.tsx' src/features/tool-discovery/all-tools-discovery.tsx src/core/search/command-search.ts src/core/search/tool-search-metadata.ts tests/unit/command-search.test.ts tests/component/all-tools-discovery.test.tsx
  • npm run lint -- src/features/tool-discovery/all-tools-discovery.tsx tests/component/all-tools-discovery.test.tsx
  • git diff --check
• Batch-level checks run:
  • npm run build passed after moving personalized All Tools controls out of static HTML and keeping the route within the performance budget.
  • Earlier npm run build attempts failed on /en/all-tools HTML budget; those failures were used to reduce static page weight before the final passing build.
• Full checks run:
  • npm run test passed: 228 files, 1154 tests.
  • npm run lint passed.
  • npm run build passed.
• Full CI:
  • Remote PR checks will run once on this PR; no extra CI-triggering pushes planned unless a failure requires it.