| name | data-analytics-skills-claude |
| description | Use Claude's portable data analytics skills library for quality checks, analysis, documentation, visualization, and stakeholder communication |
| triggers | ["help me run a data quality audit","I need to analyze this cohort data","create a semantic model for our metrics","investigate why this metric dropped","build a dashboard specification","document this SQL query in business terms","plan my data analysis approach","generate an executive summary of these findings"] |
Data Analytics Skills for Claude
Skill by ara.so — Data Skills collection.
A structured library of 31 reusable AI-powered skills that turn Claude into a hands-on analytics partner. Each skill is a self-contained instruction set that works on-demand without setup, asking targeted questions to gather exactly what it needs before executing a complete analytical workflow.
What This Project Does
This is a skill collection, not a library you import. When you describe an analytics task, Claude automatically selects and activates the appropriate skill from six categories:
- Data Quality & Validation (5 skills) - EDA, quality audits, query validation, schema mapping
- Documentation & Knowledge (5 skills) - Semantic models, documentation, SQL translation
- Data Analysis & Investigation (7 skills) - Cohorts, segmentation, funnels, time series, A/B tests
- Data Storytelling & Visualization (5 skills) - Insights, charts, dashboards, narratives
- Stakeholder Communication (5 skills) - Translation, requirements, impact quantification
- Workflow Optimization (4 skills) - Planning, context packaging, peer review
Installation
For Individual Use
- Clone the repository:
git clone https://github.com/nimrodfisher/data-analytics-skills.git
cd data-analytics-skills
- Reference skills in your Claude conversations:
"Use the programmatic-eda skill to analyze this dataset"
For AI Coding Agents
Add to your agent's system prompt or project context:
SKILL_PATH = "./data-analytics-skills"
with open(f"{SKILL_PATH}/01-data-quality-validation/programmatic-eda/SKILL.md") as f:
skill_context = f.read()
For Teams
Create a central repository with customized skills:
git clone https://github.com/your-org/data-analytics-skills.git
cd data-analytics-skills
mkdir -p 01-data-quality-validation/programmatic-eda/references
echo "# Company Schema" > references/company-schema.md
Key Concepts
Skill Structure
Every skill follows this pattern:
skill-name/
├── SKILL.md # Complete skill instructions
├── README.md # Human-readable documentation
└── references/ # Optional: company-specific context
├── company-schema.md
├── metric-definitions.md
└── business-rules.md
How Skills Work
- Minimal Context Request - Skill asks only essential information
- Structured Execution - Step-by-step analytical workflow
- Assumption Surfacing - Flags uncertainties explicitly
- Templated Output - Consistent, shareable results
Common Usage Patterns
Pattern 1: Exploratory Data Analysis
import pandas as pd
df = pd.read_csv('user_data.csv')
print(df.head())
print(df.info())
Pattern 2: Root Cause Investigation
"""
Our daily active users dropped from 10,000 to 8,800 last week.
Activate root-cause-investigation.
"""
Pattern 3: A/B Test Analysis
import pandas as pd
import numpy as np
test_data = pd.DataFrame({
'user_id': range(1000),
'variant': np.random.choice(['control', 'treatment'], 1000),
'converted': np.random.binomial(1, 0.15, 1000)
})
Pattern 4: SQL Documentation
SELECT
DATE_TRUNC('week', u.created_at) AS cohort_week,
COUNT(DISTINCT u.user_id) AS cohort_size,
COUNT(DISTINCT CASE WHEN o.created_at <= u.created_at + INTERVAL '7 days'
THEN o.user_id END) AS week_1_retained
FROM users u
LEFT JOIN orders o ON u.user_id = o.user_id
WHERE u.created_at >= '2024-01-01'
GROUP BY 1
ORDER BY 1;
Say: "Use sql-to-business-logic to document this query"
Claude produces:
- Plain English explanation
- Business logic breakdown
- Key assumptions
- Edge cases handled
- When to use/not use
Pattern 5: Dashboard Specification
"I need a dashboard for marketing performance.
Use dashboard-specification skill."
Claude asks about:
- Primary audience
- Key questions to answer
- Available data sources
- Refresh frequency needs
Then delivers:
- Metric definitions
- Visual layout mockup
- Filter specifications
- Data requirements
- Success criteria
Pattern 6: Chaining Multiple Skills
"""
Use analysis-planning for investigating signup drop.
Context: B2C SaaS, 15% drop in weekly signups over 2 weeks.
"""
"""
Run data-quality-audit on signup table.
[Provide schema and sample data]
"""
"""
Execute root-cause-investigation with validated data.
"""
"""
Use analysis-documentation to document findings.
"""
"""
Generate executive-summary for VP of Growth.
"""
Customization
Adding Company Context
Create a references/ folder in any skill:
cd 02-documentation-knowledge/semantic-model-builder
mkdir references
cat > references/metric-definitions.md << 'EOF'
- **Formula**: (Users completing 3+ key actions in 7 days) / Total signups
- **Threshold**: >40% is healthy
- **Owner**: Growth team
- **Formula**: Sum of all active subscription values
- **Excludes**: One-time payments, paused subscriptions
- **Currency**: USD normalized
EOF
Claude automatically uses these references when the skill runs.
Creating Team Conventions
cat > 06-workflow-optimization/analysis-planning/references/team-conventions.md << 'EOF'
- [ ] Create JIRA ticket
- [ ] Check if similar analysis exists
- [ ] Schedule 15min scoping with PM
- [ ] Log all assumptions in analysis-assumptions-log
- [ ] Validate data with data-quality-audit
- [ ] Document queries in SQL comments
- [ ] Run analysis-qa-checklist
- [ ] Peer review with another analyst
- [ ] Add to company wiki
EOF
Skill Quick Reference
When to Use Each Category
| You Need To... | Category | Start With |
|---|
| Check new data quality | Data Quality | programmatic-eda |
| Review or write SQL | Data Quality | query-validation, schema-mapper |
| Investigate metric change | Analysis | root-cause-investigation |
| Analyze customer behavior | Analysis | cohort-analysis, segmentation-analysis |
| Measure experiment | Analysis | ab-test-analysis |
| Document findings | Documentation | analysis-documentation |
| Define metrics once | Documentation | semantic-model-builder |
| Build dashboard | Storytelling | dashboard-specification, visualization-builder |
| Present to executives | Stakeholder | executive-summary-generator |
| Gather requirements | Stakeholder | stakeholder-requirements-gathering |
| Plan complex analysis | Workflow | analysis-planning |
Most Frequently Used Skills
CORE_SKILLS = [
"programmatic-eda",
"root-cause-investigation",
"analysis-planning",
"query-validation",
"semantic-model-builder",
"ab-test-analysis",
"cohort-analysis",
"executive-summary-generator",
"visualization-builder",
"analysis-documentation"
]
Troubleshooting
Issue: Skill asks too many questions
Solution: Provide more upfront context
"Analyze this data"
"""
Run programmatic-eda on user signup data.
Context:
- Table: signups
- Rows: 50,000
- Timeframe: Last 90 days
- Business: B2C SaaS
- Main concerns: Conversion rate, signup source quality
"""
Issue: Output doesn't match company standards
Solution: Add company references
cd skill-name/references
Issue: Skill produces generic insights
Solution: Provide business context explicitly
"""
Use root-cause-investigation.
Metric: Daily Active Users (DAU)
- Normal range: 8,000-10,000
- Current: 6,500 (35% below normal)
- Started: March 15 (5 days ago)
- Recent changes:
- New onboarding flow deployed March 14
- Email provider switched March 10
- Major competitor launched promo March 12
- Segments to check: Platform (iOS/Android/Web), User cohort, Geography
"""
Issue: Want to use skills programmatically
Solution: Reference skills in system prompts
import os
def load_skill(skill_path):
"""Load skill content for programmatic use."""
with open(f"{skill_path}/SKILL.md", 'r') as f:
return f.read()
skills = [
load_skill("01-data-quality-validation/data-quality-audit"),
load_skill("03-data-analysis-investigation/cohort-analysis"),
load_skill("04-data-storytelling-visualization/insight-synthesis")
]
system_prompt = f"""
You are a data analyst. Use these skills in sequence:
{chr(10).join(skills)}
Follow each skill's workflow completely before moving to the next.
"""
Issue: Skills not activating automatically
Solution: Skills activate based on natural language. Be explicit:
"Use the cohort-analysis skill to..."
"Activate root-cause-investigation for..."
"Run programmatic-eda on..."
"I need to understand retention by signup cohort"
"Why did our conversion rate drop?"
"Check if this dataset is clean"
Real-World Examples
Example 1: Complete Investigation Workflow
"""
PROJECT: Investigate 20% drop in mobile app engagement
STEP 1: Plan the investigation
→ Use analysis-planning skill
Output: Structured approach with hypotheses
STEP 2: Validate data quality
→ Use data-quality-audit on app_events table
Output: Quality report, issues flagged
STEP 3: Run investigation
→ Use root-cause-investigation
Output: Ranked hypotheses with evidence
STEP 4: Deep dive on top cause
→ Use segmentation-analysis on iOS vs Android
Output: iOS users affected, Android stable
STEP 5: Document findings
→ Use analysis-documentation
Output: Reproducible methodology doc
STEP 6: Present to stakeholders
→ Use executive-summary-generator
Output: 1-page summary for leadership
"""
Example 2: Building a Metrics Framework
"""
PROJECT: Create company-wide metric definitions
STEP 1: Build semantic model
→ Use semantic-model-builder skill
Provide:
- List of key metrics (ARR, MRR, CAC, LTV, Churn, etc.)
- Business definitions
- Owners
Output: Standardized metric layer
STEP 2: Add to each relevant skill
→ Save output to references/metric-definitions.md in:
- business-metrics-calculator/references/
- ab-test-analysis/references/
- executive-summary-generator/references/
STEP 3: Create data catalog
→ Use data-catalog-entry for each data source
Result: Everyone uses same definitions automatically
"""
Configuration
Environment Variables
No environment variables required. Skills work with data you provide in conversation.
Optional: Team Configuration
Create a .data-skills-config.json in your project root:
{
"default_references_path": "./company-context",
"metric_definitions": "./company-context/metrics.md",
"schema_docs": "./company-context/schema.md",
"business_rules": "./company-context/rules.md",
"team_conventions": "./company-context/conventions.md"
}
Reference in skills with:
See ${project_root}/.data-skills-config.json for company standards
Advanced Patterns
Pattern: Automated Quality Gates
"""
Before committing analysis:
1. Run analysis-qa-checklist
2. Verify all assumptions logged with analysis-assumptions-log
3. Check documentation exists via analysis-documentation
4. Validate SQL with query-validation
"""
Pattern: Analysis Templates
ANALYSIS_TEMPLATE = """
1. [analysis-planning] → approach.md
2. [data-quality-audit] → quality-report.md
3. [relevant-analysis-skill] → findings.md
4. [analysis-assumptions-log] → assumptions.md
5. [analysis-documentation] → documentation.md
6. [executive-summary-generator] → summary.md
7. [analysis-retrospective] → learnings.md
"""
Pattern: Skill Chaining Syntax
"""
PIPELINE: User retention deep-dive
data-quality-audit → cohort-analysis → segmentation-analysis →
insight-synthesis → visualization-builder → executive-summary-generator
Pass context forward at each step.
Flag any blockers immediately.
"""
Best Practices
- Always start with
analysis-planning for complex work
- Run
data-quality-audit before analyzing unfamiliar data
- Use
semantic-model-builder once, reference forever
- Document with
analysis-assumptions-log as you go
- Chain skills for complete workflows, don't jump to conclusions
- Add company context to
references/ folders for frequently-used skills
- Run
analysis-qa-checklist before sharing any findings
Resources