Improve: Skills docs, chart guidelines, and homepage dynamics

- Update discover skill with required coverage check before scanning
- Add advanced narrative patterns to voice reference (streaks, offsets, cumulative)
- Expand chart style guide with accessibility and color neutrality rules
- Make homepage featured articles load dynamically from articles.json
- Add utility scripts for chart color and label position fixes
- Gitignore automation logs

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: dshkol

.claude/skills/the-daily-discover/SKILL.md

@@ -19,7 +19,18 @@ Rscript r-tools/discover_stories.R
 
 ## Discovery Workflow
 
-### 1. Scan Available Data
+### 1. Check Existing Coverage (REQUIRED FIRST STEP)
+
+Before scanning CANSIM, identify tables already covered:
+
+```bash
+# List existing articles to extract table numbers
+ls docs/en/*/index.md | xargs grep -h "statcan.gc.ca/t1/tbl1" | grep -oE "[0-9]{2}-[0-9]{2}-[0-9]{4}" | sort -u
+```
+
+**Exclude these table numbers from recommendations.** The goal is diversity—never suggest a table that's already been covered unless explicitly asked for an update.
+
+### 2. Scan Available Data
 
 ```r
 library(cansim)
@@ -34,34 +45,24 @@ candidates <- cubes %>%
   arrange(desc(cubeEndDate))
 ```
 
-### 2. Score Candidates
+### 3. Score Candidates
 
 | Dimension | Weight | Criteria |
 |-----------|--------|----------|
 | Recency | 25% | Days since release (fresher = higher) |
-| Diversity | 25% | Sector gap from existing articles |
+| Diversity | 25% | Sector gap from existing articles—deprioritize sectors covered in last 7 days |
 | Narrative | 25% | Regional variation, trend reversals |
 | Public Interest | 15% | Topic relevance to general audiences |
 | Data Quality | 10% | Complete coverage, national totals |
 
-### 3. Validate Top Candidates
+### 4. Validate Top Candidates
 
 Before committing:
-1. Fetch sample data to verify structure
-2. Check for StatCan Daily release
-3. Verify national totals exist
-4. Confirm table is current (not deprecated)
-
-## Check Existing Coverage
-
-```bash
-ls docs/en/*/index.md | head -20
-```
-
-Avoid redundancy:
-- If CPI just published, deprioritize other price indices
-- Retail Trade + Wholesale Trade complement each other
-- Housing Starts + Building Permits complement each other
+1. **Verify table numbers**: Run `search_cansim_cubes("keyword")` to confirm table numbers are current—they go stale as StatCan discontinues/replaces tables
+2. Fetch sample data to verify structure
+3. Check for StatCan Daily release
+4. Verify national totals exist
+5. Confirm table is current (not deprecated)
 
 ## Output Format
 

.claude/skills/the-daily-generator/references/french.md

@@ -86,3 +86,7 @@ d.toFixed(1).replace(".", ",") + " %"  // 2,2 %
 // French currency
 d.toFixed(1).replace(".", ",") + " G$"  // 165,4 G$
 ```
+
+## Chart Layout
+
+French labels are typically longer than English. Increase left margins by ~20-40px for horizontal bar charts to prevent truncation (e.g., `marginLeft: 200` vs `marginLeft: 160` in English).

.claude/skills/the-daily-generator/references/voice.md

@@ -218,3 +218,100 @@ On landing page cards:
 
 ### Bad Component Analysis
 "Food prices continued their alarming rise, squeezing household budgets and forcing families to make difficult choices." (emotional, speculative)
+
+## Advanced Narrative Patterns
+
+These patterns are observed in real Statistics Canada Daily articles and add depth to statistical storytelling.
+
+### Streak Mentions
+
+When a metric moves in the same direction for multiple consecutive periods, highlight the streak:
+
+- "increased for the **fourth consecutive month** in October"
+- "fell for a **second consecutive month** in October"
+- "marked the **first decline since July**"
+
+### Cumulative Context
+
+Provide multi-period cumulative context for sustained trends:
+
+- "bringing the **cumulative gain to 16,700 (+2.0%) since July**"
+- "The three-month gain in the sector **partially offset the cumulative decline of 10,300 (-0.8%)** recorded from April to July 2025"
+
+### Offset Framing
+
+When a current-period change reverses or partially offsets a prior period:
+
+- "**largely offsetting** a decline of 24,300 (-0.1%) in September"
+- "**more than offsetting** the uptick in September (+11,000; +2.3%)"
+- "**partially offset by** declines in administration and support"
+
+### Lone Exception Pattern
+
+Highlight the outlier when most categories move in one direction:
+
+- "**The only sector** to record a year-over-year increase was manufacturing"
+- "truck transportation (-4,500; -2.1%) was the **lone subsector** to record a year-over-year decline"
+
+### Historical Milestone
+
+When a value reaches a notable historical level:
+
+- "The number of vacancies in October was the **lowest since October 2017**"
+- "to a level **comparable with January 2016**"
+- "the **highest level since April 2024**"
+
+### Cross-Survey References
+
+When relevant, link to related releases for complementary data:
+
+- "unemployment rate rose from 6.6% to 6.9% (according to the **Labour Force Survey**)"
+- "For more information, see the **Consumer Price Index** release"
+
+### Provincial Extremes with Context
+
+When presenting provincial data, highlight both extremes:
+
+- "**7.2 unemployed persons for every job vacancy** in Newfoundland and Labrador, the **highest ratio** among the provinces. In comparison, the... ratio was **lowest in Quebec (2.3)**, followed by Saskatchewan (2.5)"
+
+### Causal Attribution (when data supports)
+
+Provide brief explanations for derived metrics:
+
+- "This increase in the job vacancy rate in manufacturing was **primarily due to the decline in payroll employment** over the period"
+- "Growth in average weekly earnings can reflect a range of factors, including changes in wages, **composition of employment**, hours worked and **base-year effects**"
+
+## Article Types
+
+### Single-Indicator Releases
+Most common type. Covers one metric (CPI, Retail Trade, GDP).
+
+**Structure:**
+1. Headline with key number
+2. Highlights (3-5 bullets)
+3. Lede paragraph
+4. Component analysis
+5. Regional analysis
+6. Note to readers
+
+### Multi-Indicator Releases
+Combines related metrics from the same survey (e.g., SEPH covers payroll employment + weekly earnings + job vacancies).
+
+**Structure:**
+1. Headline (usually the primary metric)
+2. Highlights covering multiple metrics
+3. Lede paragraph for primary metric
+4. Section for each indicator with its own heading
+5. Cross-metric analysis
+6. Note to readers (may cover multiple surveys)
+
+**Section heading pattern:** "[Sector/Topic] [verb] [direction] in [period]"
+- "Health care and social assistance records payroll employment increase in October"
+- "Job vacancies decrease in four sectors and increase in one"
+- "Average weekly earnings continue to increase on a year-over-year basis"
+
+### Data Release Announcements
+Short articles announcing new tables are available. Not suitable for D-AI-LY (no narrative).
+
+### Product Announcements
+Short articles pointing to dashboards or tools. Not suitable for D-AI-LY.

.gitignore

@@ -18,3 +18,4 @@ output/
 .Rhistory
 .RData
 .Rproj.user/
+automation/logs/