bytelyst/learning_ai_common_plat
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
1,450 Commits 2 Branches 3 Tags 61 MiB
57a09c31dd
Commit Graph

3 Commits

Author SHA1 Message Date
saravanakumardb1
57a09c31dd feat(ai-ui): @bytelyst/ai-ui@0.5.0 — Wave 13.C trust surfaces (CostMeter / ConfidenceTag / RefusalCard) 
Three new primitives — every product chat / agent surface should
adopt these to make the model honest about what it is doing.

──────────────────────────────────────────────────────────────────
<CostMeter>  ·  Wave 13.C.1
──────────────────────────────────────────────────────────────────
  packages/ai-ui/src/CostMeter.tsx (new)

  - Live token + (optional) USD readout
  - 4 tiers: neutral (no budget) · ok (<70 %) · warn (>=70 %) ·
    danger (>=95 %) — token-tinted via color-mix() so all four
    palettes degrade gracefully on browsers without var() support
  - NaN-safe — non-finite / negative inputs floor to 0
  - role=status + aria-live=polite + aria-label assembles a
    screen-reader-friendly sentence
  - Mini-bar visual indicator at the end of the pill when budget
    is provided
  - Pure passive surface — never warns / prompts / blocks

──────────────────────────────────────────────────────────────────
<ConfidenceTag>  ·  Wave 13.C.2
──────────────────────────────────────────────────────────────────
  packages/ai-ui/src/ConfidenceTag.tsx (new)

  - Accepts `number | 'high' | 'medium' | 'low' | 'unknown'`
  - Default thresholds 0.8 / 0.5 — both overridable
  - Out-of-range numerics map to `unknown` (no false confidence)
  - Optional `showScore` renders a tabular-nums percent suffix
  - 4 token-tinted palettes (success / warning / danger /
    neutral) — pair naturally with <CitationChip> for the full
    'show your work' story

──────────────────────────────────────────────────────────────────
<RefusalCard>  ·  Wave 13.C.3
──────────────────────────────────────────────────────────────────
  packages/ai-ui/src/RefusalCard.tsx (new)

  - 6 reason archetypes — safety / policy / capability /
    authorization / rate-limit / unknown — each with a typed
    heading + glyph
  - Calm warning palette (never red) — refusals are not errors
  - Up to 3 actionable next steps (further entries silently
    clipped) — one is markable `primary` to render as filled CTA
  - Optional `footer` slot for policy doc links
  - role=note + composite aria-label covering heading +
    explanation

──────────────────────────────────────────────────────────────────
Quality gates
──────────────────────────────────────────────────────────────────
  ✓ pnpm -F @bytelyst/ai-ui test  →  67/67 passing (was 53/53)
    +14 new trust-surface tests in src/__tests__/trust.test.tsx
  ✓ Exports wired in src/index.ts under '0.5 surfaces' section
  ✓ package.json 0.4.0 → 0.5.0

──────────────────────────────────────────────────────────────────
Roadmap tracker — 5 boxes flipped (§11)
──────────────────────────────────────────────────────────────────
  13.C.1  CostMeter shipped
  13.C.2  ConfidenceTag shipped
  13.C.3  RefusalCard shipped
  13.C.7  trust-surfaces showcase (lands in paired showcase commit)
  MAG.3   the trust-surfaces customer-magnet ✨

Wave 13 Futurism: 5/39 → 9/39 (23%)
Magnet demos:     1/8  → 2/8  (25%)
TOTAL:            19/202 → 24/202 (12%)

Vendored snapshot + showcase /ai-ui/* + /futurism/trust-surfaces
routes land in the paired showcase commit.

Pending in 13.C: ProvenanceDrawer (.4) · DebugOverlay (.5) ·
PrivacyBadge (.6).
 2026-05-27 16:45:07 -07:00
saravanakumardb1
e2eea086dc feat(packages): Wave 2 v0.4 + Wave 3 v0.1 — ai-ui expanded, command-palette new 
═══════════════════════════════════════════════════════════════════════
@bytelyst/ai-ui  bump 0.1.0 → 0.4.0
═══════════════════════════════════════════════════════════════════════
Folds three more roadmap milestones into the flagship package.

  0.2: <ToolCallCard>   — disclosure card; status pill, JSON preview
       <CitationChip>   — inline citation marker + hover preview
       useToolCalls()   — per-turn tool-invocation state machine
                          (begin/update/settle/clear); preserves insertion
                          order across updates; auto-computes durationMs
  0.3: <AgentTimeline>  — vertical think→act→observe→respond trace;
                          embeds ToolCallCard for kind='tool_call' steps
       <ModelPicker>    — model dropdown with capability chips, cost,
                          latency, context window, disabled gating
  0.4: <ToolPalette>    — searchable tool list with MCP-style discovery
                          (source can be ToolDescriptor[] OR an
                          'mcp://...' URL resolved via a discover
                          adapter; default adapter is fetch+JSON)

Types extended:
  - ToolInvocation, ToolCallStatus, Citation added
  - Message gains optional toolInvocations + citations

Tests: 53/53 (27 old + 26 new) · typecheck clean · 7.65 KB / 35 KB

═══════════════════════════════════════════════════════════════════════
NEW PACKAGE: @bytelyst/command-palette@0.1.0
═══════════════════════════════════════════════════════════════════════
Wave 3 deliverable — Cmd-K dialog with three modes and pluggable command
registration. Roadmap §Wave 3 of ROADMAP_2026.md.

What's exported:
  <CommandRegistryProvider>  — wrap your app once
  <CommandPalette>           — the dialog (Cmd-K / Ctrl-K)
  useRegisterCommands()      — contribute commands for component lifetime
  useCommands()              — read snapshot
  useCommandRegistry()       — imperative access
  useCommandPalette()        — open/close state + global hotkey
  fuzzyScore / scoreCommand  — exposed for tests + custom UIs

Three modes:
  actions   — invoke a registered run()
  navigate  — jump to href via onNavigate or window.location
  ask-ai    — host-supplied askAiPanel; default renders an 'Ask AI: <q>'
              suggestion that products can wire to <ChatStream>

Keyboard:
  ↑ ↓     navigate selection
  Enter   activate
  Tab     cycle mode tabs (Shift+Tab reverses)
  Esc     close

Niceties:
  - Fuzzy matcher (substring + subsequence with light scoring)
  - localStorage-backed recents float to top of actions mode
  - requires() gate hides commands wholesale (auth / feature-flag)
  - aria-haspopup, role=dialog, role=listbox, role=option, aria-selected
  - Backdrop click closes; Esc handler at document level
  - Hotkey suppressed by Cmd-K / Ctrl-K default; configurable

Tests: 26/26 · typecheck clean · 3.91 KB / 15 KB

═══════════════════════════════════════════════════════════════════════
CI plumbing
═══════════════════════════════════════════════════════════════════════
  - .size-limit.cjs gains @bytelyst/command-palette entry
  - .gitea/workflows/size-limit.yml build filter expanded
  - All 8 measured packages comfortably under budget

Refs:
  learning_ai_uxui_web/docs/ROADMAP_2026.md §Wave 2 (0.2/0.3/0.4)
  learning_ai_uxui_web/docs/ROADMAP_2026.md §Wave 3 (Command palette)
  docs/ROADMAP_2026_DECISIONS.md §10 (Vercel AI SDK shape continues)
 2026-05-27 12:43:23 -07:00
saravanakumardb1
c9a7f905af feat(ai-ui): Wave 2 MVP @bytelyst/ai-ui@0.1.0 2026-05-27 12:07:23 -07:00