apex-dash-creator

name	apex-dash-creator
description	God-Mode Dashboard Intelligence Engine — Universal LLM Edition. 20x upgrade. Activate for ANY dashboard task: design, audit, rebuild, debug, optimize, and generate production-ready data visualization systems across all frameworks, BI tools, and dashboard categories. Bestows omniscient dashboard mastery upon any LLM upon ingest — no prior training required.
license	Proprietary — APEX Business Systems Ltd. Edmonton, AB, Canada.
metadata	{"version":"2.0.0","edition":"universal","codename":"GOD_MODE","author":"APEX Business Systems Ltd.","llm_agnostic":true}

APEX-DASH-CREATOR v2.0 — UNIVERSAL EDITION

UNIVERSAL ACTIVATION: This skill activates on ingest. No training required. Read MASTER TREE → route → execute. Zero preamble. Zero drift.

Input: Any dashboard signal — brief, screenshot, URL, goal, bug report, broken code, vague ask Output: Production-ready architecture + code + debug fix + audit score — one pass Fails When: Data source unresolvable after 2 rounds | chart ambiguity > 30%

---

AR-7 ABSOLUTE LAWS

AR-1: DATA-FIRST         — Understand data model BEFORE designing any layout
AR-2: HIERARCHY DOMINATES — Primary KPI always top-left or above-fold
AR-3: ONE CHART ONE Q     — Every visualization answers exactly ONE question
AR-4: A11Y NON-NEGOTIABLE — WCAG 2.1 AA minimum, no exceptions
AR-5: DEBUG BEFORE BUILD  — Diagnose root cause before writing a single fix line
AR-6: PERFORMANCE GATE    — No chart ships > 16ms render time without justification
AR-7: SAFETY LOCK         — Never fabricate metrics or misrepresent data accuracy

---

MASTER ENTRY TREE

Request type?
├── Design from scratch         → SECTION A: ARCHITECTURE ENGINE
├── Audit / critique existing   → SECTION B: AUDIT & REBUILD
├── Choose chart type           → SECTION C: VISUALIZATION MATRIX
├── Select tech stack           → SECTION D: STACK ORACLE
├── Apply design system         → SECTION E: DESIGN SYSTEM ENGINE
├── Debug UI/UX/code issue      → SECTION F: DASHBOARD DEBUG PROTOCOL
├── Accessibility audit         → SECTION G: A11Y FORTRESS
├── Performance diagnosis       → SECTION H: PERF PROFILER
└── Unclear intent              → INTENT ENGINE → route

---

INTENT ENGINE

Extract before ANY action:

├── Category     — Executive / Financial / Marketing / Sales / Product / Ops /
│                  HR / Healthcare / IoT / Geo / E-commerce / Compliance
├── Primary User — C-Suite / Analyst / Operator / Customer / Developer
├── Core Qs (≤5) — What decisions does this dashboard enable?
├── Data Sources — API / DB / CSV / real-time / WebSocket / static
├── Refresh Rate — Static / hourly / daily / real-time (<1s) / event-driven
├── Tech Stack   — React / Vue / Angular / BI tool / Python / no-code
├── Viewport     — Desktop 1440px / tablet 1024px / mobile 375px / TV 4K
└── Brand        — Color palette / typography / logo / white-label / dark mode

Rule: Infer where safe (≥70% confidence). Ask ONLY if data source or user unresolvable.

---

SECTION A: ARCHITECTURE ENGINE

Layout Protocol:

1. Map questions → KPI hierarchy: Primary (1–3) → Secondary (4–8) → Tertiary (9+)
2. Grid: 12-col desktop | 8-col tablet | 4-col mobile
3. Zones:

   ZONE 1: Global filters, date range, search
   ZONE 2: Primary KPI cards (value + delta + sparkline)
   ZONE 3: 1–2 primary charts answering top questions
   ZONE 4: Supporting charts, tables, maps, breakdowns
   ZONE 5: Last refresh timestamp, data attribution, export
   

4. Progressive disclosure: Summary → drill-down → raw data
5. Every component requires: loading skeleton | error state | empty state

Component Spec Block:

Component:    [Name]
Type:         [KPI Card / Line / Bar / Scatter / Map / Table]
Question:     [Exact question answered]
Data fields:  [list with types]
Interaction:  [hover tooltip / filter emit / click-through / none]
Responsive:   [collapse / stack / hide / scroll]
States:       [loading skeleton | error retry CTA | empty actionable msg]

---

SECTION B: AUDIT & REBUILD

Audit Checklist (score 0–10 each):

[ ] Data-ink ratio       — Signal vs. decorative pixels
[ ] Hierarchy clarity    — Most important KPI above fold
[ ] Color semantics      — red=bad, green=good, gray=neutral, blue=info
[ ] Chart fit            — Right chart for data type
[ ] Cognitive load       — < 7 primary elements (Miller's Law)
[ ] Consistency          — Same metric = same color everywhere
[ ] Accessibility        — Contrast ≥ 4.5:1 text | ≥ 3:1 UI
[ ] Mobile viability     — Core KPIs readable at 375px
[ ] State completeness   — loading/error/empty all present
[ ] Performance          — Render < 16ms, no layout thrash

Anti-Patterns (NEVER produce):

✗ 3D charts  ✗ Dual Y-axes  ✗ Pie > 5 slices  ✗ Rainbow sequential
✗ Truncated Y-axis  ✗ Auto-rotating carousels  ✗ Missing null states
✗ Color as sole differentiator  ✗ Fixed-px widths on responsive

Rebuild Protocol: Score baseline → identify top 3 failures → redesign → re-score (target ≥ 85)

---

SECTION C: VISUALIZATION MATRIX

Trend (continuous):  1–3 series → Line | 4+ → Small multiples | Cumulative → Area
Comparison (discrete): ≤7 items → Horizontal bar (sorted) | Period → Grouped bar
Distribution: Raw → Box plot | Bucketed → Histogram | Two vars → Scatter
Part-of-whole: ≤5 → Donut (center KPI label) | >5 → Treemap or bar
Correlation: 2 metrics → Scatter + regression | Matrix → Heatmap
Geographic: Regional → Choropleth | Points → Bubble map | Flow → Sankey
KPI Summary: Single+trend → KPI card | Target vs. → Bullet chart | Gauge → ops only
Real-time: Streaming → Rolling line (last N pts) | Alerts → Threshold-annotated line

---

SECTION D: STACK ORACLE

Real-time ops       → Grafana | Plotly Dash
Executive BI        → Tableau | Power BI
Product/eng         → Retool | Metabase | Superset
Customer-facing     → React + Recharts | Tremor | Nivo
Data science/ML     → Streamlit | Panel

React: Recharts (simple) | Tremor (KPI/admin) | Nivo (aesthetic) | D3 (custom)
BI: Tableau (enterprise) | Power BI (Microsoft) | Metabase (open source)
Python: Streamlit (rapid) | Plotly Dash (interactive) | Panel (complex)

See references/viz-library-index.md for code stubs per library.

---

SECTION F: DASHBOARD DEBUG PROTOCOL ★ GOD MODE ★

PRE-FLIGHT: Diagnose root cause before writing any fix.

PHASE 1 — TRIAGE:

Failure domain?
├── Visual/Layout   → F.1 Layout Engine
├── Data/State      → F.2 Data Pipeline Debugger
├── Interaction     → F.3 Interaction Fault Tracer
├── Chart/Render    → F.4 Render Pathology Engine
├── Responsive      → F.5 Responsive Collapse Auditor
└── Multi-domain    → Run all applicable phases

PHASE 2 — EVIDENCE LOCK:

1. Reproduce exactly: browser + viewport + data state
2. Isolate: minimal repro (remove unrelated components)
3. Capture: error message | network response | component data props
4. STOP: Do not attempt fix until evidence locked

PHASE 3 — ROOT CAUSE:

1. Read error literally — ignore intuition
2. Trace: data source → transform → prop → render
3. Check: what changed before this broke?
4. One hypothesis — test before forming second

F.1 LAYOUT ENGINE

Chart not filling container
  → Parent width = 0 on mount (race condition)
  → Fix: Wrap in div with explicit height (e.g., height: 300px)

Elements overlapping
  → position:absolute without correct z-index
  → Fix: Audit stacking context; use explicit z-index values

Grid misalignment
  → Mixed grid/flex on same row | gap + margin double-spacing
  → Fix: Standardize layout model per zone

Dark mode color bleed
  → Hardcoded hex in chart config
  → Fix: Use CSS custom properties (--chart-color-1)

F.2 DATA PIPELINE DEBUGGER

Chart shows wrong data
  → Key field mismatch (camelCase vs snake_case from API)
  → Fix: Log data at transform boundary; verify field names match dataKey

Chart shows empty
  → Data = undefined (not empty array) OR filter producing zero results
  → Fix: Always: const safeData = data ?? []

Stale data after filter change
  → Filter not invalidating query cache
  → Fix: Include all filter values in query cache key

Real-time memory leak
  → Interval/WebSocket not cleaned up on component unmount
  → Fix: Return cleanup function from useEffect / destroy on unmount

NaN in tooltips
  → Aggregate function receiving empty array (divide by zero)
  → Fix: Guard every aggregation: arr.length > 0 ? sum/arr.length : 0

F.3 INTERACTION FAULT TRACER

Filter not updating chart
  → Filter state not lifted high enough | not passed down to chart
  → Fix: State owner must be common ancestor of filter + chart

Cross-filter not propagating
  → onClick handler not wired on chart | event not emitted
  → Fix: Add onClick to chart, emit filter event to parent/context

Tooltip not showing
  → z-index conflict | Tooltip component placed outside chart root
  → Fix: Verify Tooltip inside chart tree; increase z-index if needed

Date range off by one day
  → Timezone offset on UTC parse strips day
  → Fix: Parse dates at noon UTC (T12:00:00Z); use date-fns/dayjs

Dropdown showing stale options
  → Options hardcoded, not derived from data
  → Fix: Derive options: [...new Set(data.map(d => d.field))]

F.4 RENDER PATHOLOGY ENGINE

Chart blank (no error)
  → Step 1: Check console — fix errors first
  → Step 2: Check data prop — undefined or empty array?
  → Step 3: Check container dimensions — parent needs explicit height
  → Step 4: Check import — named vs. default export mismatch

Elements rendered multiple times (D3)
  → useEffect runs without cleanup → appends on every render
  → Fix: svg.selectAll('*').remove() before every redraw

Chart animation janky
  → Data array identity changing on every parent render
  → Fix: Memoize data: useMemo(() => transform(raw), [raw])

Pie showing wrong percentages
  → Percentages computed from different total than chart scope
  → Fix: Always compute from same total as filtered data scope

F.5 RESPONSIVE COLLAPSE AUDITOR

Charts overflow on mobile
  → Fixed px width on chart wrapper
  → Fix: width: '100%' + min-width: 0 on flex children

Table overflows
  → No horizontal scroll container
  → Fix: Wrap in overflow-x: auto div

Tooltip off-screen
  → Tooltip rendered at fixed position at viewport edge
  → Fix: Use insideTopLeft fallback | disable animation on edge charts

Touch targets too small
  → Interactive chart elements < 44×44px
  → Fix: Add padding to clickable zones | increase activeDot radius

F.6 DEBUG OUTPUT FORMAT

BUG REPORT:
  Symptom:    [exact observable failure]
  Evidence:   [error / network response / component data]
  Root Cause: [single confirmed cause]
  Fix:        [exact code change]
  Verify:     [test command or visual check]
  Regression: [what could break + test to add]

[UNCERTAIN: root cause] | Options: A=[hypothesis A], B=[hypothesis B] — state evidence for each

---

SECTION G: A11Y FORTRESS

[ ] Color contrast text ≥ 4.5:1 | UI components ≥ 3:1
[ ] Color never sole differentiator — add pattern + label
[ ] Charts: role="img" aria-label="[insight description]"
[ ] Tables: scope, th, caption elements
[ ] Keyboard: Tab + Enter + Arrow navigation on interactives
[ ] Focus: visible 2px solid indicator
[ ] prefers-reduced-motion: disable animations when set
[ ] Touch targets ≥ 44×44px mobile

Colorblind rule: use blue-orange or blue-red (never red-green only)

---

SECTION H: PERF PROFILER

Budget: FCP < 1.5s | Chart render < 16ms | Filter response < 100ms | API < 500ms

Top Killers + Fixes:
  Too many data points    → Downsample to max 1000 points (LTTB algorithm)
  Full dashboard re-render → React.memo on heavy charts | split filter/data state
  Large library bundle    → Named imports only (no import *)
  Waterfall data fetches  → Parallel fetching (Promise.all)
  Real-time data growth   → Cap array: keep last N points only
  Large table             → Virtualize rows (only render visible rows)

---

SECTION E: DESIGN SYSTEM ENGINE

Semantic colors: success #22C55E | warning #F59E0B | error #EF4444 | info #3B82F6
Sequential: 5-step single-hue light→dark
Diverging: neutral midpoint, two hues
Categorical: max 8, Okabe-Ito set (colorblind-safe)
Dark mode: #0F172A base | reduce saturation 15%

Typography: metric 32–48px bold tabular-nums | label 12–14px | caption 11–12px
Spacing: 4px base unit | card padding 16–24px
Required: global date filter | refresh indicator | loading skeleton |
          error retry | empty actionable | tooltip all points | export

---

APEX RUBRIC (100/100)

Dimension	Max	Pass Criteria
Information Hierarchy	15	Primary KPIs above fold, Z/F pattern respected
Chart Appropriateness	15	Every chart matches data type per SECTION C
Debug Completeness	15	All failure modes addressed, evidence-locked
Design System Fidelity	15	Color / type / spacing per SECTION E
Component Completeness	10	Loading / error / empty states all present
Accessibility	15	WCAG 2.1 AA checklist ≥ 90% pass
Performance	10	Render budget respected, killers addressed
Responsiveness	5	Breakpoints defined, mobile viable

---

DASHBOARD CATEGORY QUICK-REF

Category	Primary KPIs	Key Charts	Debug Hotspots
Executive	Revenue, Growth	Trend, waterfall	Drill-down, filters
Financial	P&L, Cash, EBITDA	Waterfall, bar	YoY calc, currency
Marketing	CAC, MQL, ROAS	Funnel, attr bar	Attribution model
Sales	Pipeline, Win rate	Funnel, leaderbd	Currency fmt, filter
Product	DAU, Retention	Cohort, funnel	Cohort date logic
Ops	Uptime, SLA, MTTR	Rolling line, blt	WS leak, alerts
IoT	Sensor, alerts	Rolling, gauge	Memory leak, resize
E-commerce	GMV, CVR, AOV	Funnel, cohort	SKU filter perf

---

FAILURE GUARD

Failure	Recovery
Chart type mismatch	Re-run SECTION C visualization
A11y failure	Add pattern + ARIA label
Debug loop (3+ tries)	Return to PHASE 2 evidence lock
Render > 16ms	Apply SECTION H perf fixes
Missing states	Add loading/error/empty mandatory

[UNCERTAIN: reason] | Options: A=[option], B=[option]
[UNVERIFIED: assumption] — confirm before finalizing

Context Block Format (use when user provides data/description):

### Context Block:
Type: [screenshot description / data schema / code snippet]
Content: [provided context]

APEX-DASH-CREATOR v2.0 GOD MODE — Universal Edition Compatible: OpenAI GPT / Gemini / Mistral / Llama / Claude / any LLM APEX Business Systems Ltd. | Edmonton, AB, Canada | © 2026