Initial commit

skills/xlsx/scenes/analyze.md

@@ -0,0 +1,95 @@
+# Scene: Data Analysis → Excel Output
+
+## When This Applies
+User wants to analyze data (statistics, trends, comparisons, pivots, aggregation) and receive results as an Excel file — possibly with charts, summary tables, or dashboards.
+
+This scene bridges **pandas analysis** with **openpyxl output**. The deliverable is always an .xlsx file.
+
+## Workflow
+
+```
+1. LOAD       → Read input data (CSV/XLSX/JSON/DB)
+2. EXPLORE    → Understand structure, quality, distributions
+3. ANALYZE    → Compute metrics, aggregations, statistical tests
+4. DESIGN     → Plan Excel output (sheets, charts, KPIs)
+5. BUILD      → Write analysis results to .xlsx with formatting
+6. CHART      → Add charts (Excel-native or embedded matplotlib)
+7. QA         → recalc → audit → scan → chart-verify
+8. PIVOT      → If needed, run xlsx.py pivot as final step
+9. VALIDATE   → validate → deliver
+```
+
+## Analysis Framework
+
+### Phase A: Problem Framing
+- What question is the user trying to answer?
+- Who will consume this output? (executive summary vs. detailed analysis)
+- What decisions will be made based on this data?
+
+### Phase B: Data Quality Assessment
+- Missing values: count, pattern (random vs. systematic)
+- Outliers: statistical detection (IQR, z-score)
+- Data types: numeric vs. categorical, date parsing
+- Duplicates: exact and fuzzy
+
+### Phase C: Exploratory Analysis
+- Distributions: histograms, box plots for key variables
+- Correlations: pairwise for numeric columns
+- Segmentation: group-by analysis on categorical dimensions
+- Time patterns: trends, seasonality if time-series data
+
+### Phase D: Insight Extraction
+- Rank findings by business impact, not statistical significance
+- Each insight must be actionable — "so what?" test
+- Cross-validate: check the same insight from a different angle
+
+### Phase E: Cross-Validation
+- Sanity check totals against known benchmarks
+- Verify computed metrics with alternative formulas
+- Document any assumptions or limitations in the output
+
+**Industry-specific frameworks:**
+- **Finance**: Variance analysis → trend decomposition → ratio analysis → peer comparison
+- **Marketing**: Funnel analysis → cohort analysis → attribution → ROI calculation
+- **Operations**: Throughput analysis → bottleneck identification → utilization rates → SLA compliance
+
+---
+
+## Multi-Sheet Report Layout
+
+```
+Sheet 1: "Dashboard"     — KPI cards + summary chart
+Sheet 2: "Detail"        — Full analysis table with formatting
+Sheet 3: "Charts"        — Additional visualizations
+Sheet 4: "Raw Data"      — Original data for reference (tab color: gray)
+```
+
+### KPI Summary Card Pattern
+
+Place 4-6 KPI metrics at the top of Dashboard sheet (row 3-4), each spaced 3 columns apart. Include label (small, gray) and value (large, bold, themed) with appropriate number format.
+
+---
+
+## PivotTable Decision
+
+| Situation | Use |
+|-----------|-----|
+| Need interactive PivotTable in Excel | `"$XLSX_SKILL_DIR/xlsx.py" pivot` |
+| Just need a summary table (static) | pandas `pivot_table` → openpyxl |
+| Simple aggregation (1 dimension) | pandas `groupby` → openpyxl |
+
+**Trigger phrases**: summarize, aggregate, group by, categorize, breakdown, distribution, tally, totals per, cross-tab, 汇总, 透视, 分类统计, 交叉分析
+
+---
+
+## Data Provenance
+
+When analysis uses external data, create a **"Sources" sheet** (tab color: `PRIMARY`) with columns: Data Description | Source Name | Source URL | Access Date.
+
+Skip when user provides all data directly.
+
+---
+
+## Code Recipes
+
+For specific code patterns (aggregation, time series, comparison, cleaning, bridge pattern), load `scenes/analyze-recipes.md` on demand.