🤖 Data Science & AI/ML Skills Suite

Derived from anthropics/skills

Adaptation of anthropics/skills for Data Science & AI/ML use cases. Source focus: official Anthropic skill templates, webapp testing, comms

---

What This Skill Suite Does

Data pipelines, model training, evaluation, MLOps and analytical reporting.

This collection provides 10 specialised commands and 5 multi-step workflows, all with a consistent structured-output UI so you always know exactly where you are and what to do next.

---

Quick Install

# Clone this skill
cp -r . ~/.claude/skills/r00-anthropics-skills--datascience/

# Register in Claude Code
# In a Claude Code session:
/read ~/.claude/skills/r00-anthropics-skills--datascience/SKILL.md

---

Commands

Command	Description
/data-profiling	Automated EDA report: distributions, nulls, outliers, correlations and drift
/feature-engineer	Feature importance analysis with SHAP values and automated encoding recipes
/model-evaluate	Model performance dashboard: ROC, PR curves, confusion matrix and bias check
/pipeline-scaffold	Modular ML pipeline scaffold with versioning, logging and registry hooks
/ab-test-design	Statistical A/B test design: sample size, power, MDE and sequential testing
/sql-optimize	Query plan analysis, index recommendations and cost estimation
/dashboard-spec	BI dashboard specification from KPI list with chart types and data sources
/data-contract	Schema validation, SLA definition and data quality contract generation
/llm-eval	LLM output evaluation harness: hallucination rate, faithfulness and latency
/anomaly-detect	Time-series anomaly detection with root-cause attribution and alert tuning

Usage:

/data-profiling <target>
/feature-engineer --scope full --output md

---

Workflows (Multi-step)

Workflow	Description
ml-project-init	End-to-end ML project: EDA → baseline → feature engineering → model → deploy
data-migration	Data warehouse migration: audit → schema map → ETL → validation → cutover
reporting-pipeline	Automated reporting pipeline: source → transform → validate → visualise → deliver
model-retraining	Scheduled model retraining: drift detect → retrain → shadow → promote → monitor
analytics-sprint	2-week analytics sprint: question → data → analysis → insight → recommendation

Usage:

/workflows:ml-project-init <target> --scope full

---

UI Design

All commands display structured output with:

• Progress panels — real-time step tracking
• Findings tables — sorted by severity (🔴🟠🟡🟢)
• Action checklists — quick wins → medium-term → strategic
• Summary cards — at-a-glance metrics after each command

Progress Display Example

╔══════════════════════════════════════════════════╗
║  ML Pipeline  —  churn_prediction_v3             ║
╠══════════════════════════════════════════════════╣
║  Data ingestion   ✓   1.2M rows loaded            ║
║  Profiling        ✓   12 features, 3 issues found ║
║  Feature eng.     ✓   47 features created          ║
║  Training         ⟳   Epoch 18/50  [███████░░░]   ║
║  Evaluation       ░   Pending                      ║
║  Registry push    ░   Pending                      ║
╚══════════════════════════════════════════════════╝

DATA QUALITY ISSUES
  ✗  customer_age   → 847 nulls (0.07%)  → Impute with median
  ⚠  last_purchase  → 12 future dates    → Clip to today
  ⚠  revenue        → 3σ outliers: 214   → Review before drop

---

Interaction Pattern

Every command follows this 5-step structure:

① Scope Confirmation  — verify target and options with user
② Live Analysis       — progress bar while working
③ Findings Table      — structured results sorted by impact
④ Action Plan         — prioritised, time-boxed recommendations
⑤ Next Steps          — suggested follow-up commands

---

Source Repository

This suite is derived from anthropics/skills which focuses on: official Anthropic skill templates, webapp testing, comms.

Improvements in this adaptation:

• Domain-specific command vocabulary for Data Science & AI/ML
• Enhanced structured output with visual progress tracking
• Prioritised action plans with time estimates
• Workflow orchestration for end-to-end processes
• Consistent UI conventions across all commands

---

License

MIT — free to use, modify and distribute.