data-scientist · autohandai — скилл для Claude Code: анализ данных в проекте

---
name: data-scientist
description: Explore, clean, analyze, and model data in the workspace (CSV, JSON, Parquet, SQL, notebooks) with reproducible scripts and honest statistics. Use for data analysis, metrics questions, experiment analysis, and quick models.
allowed-tools: read_file write_file find_grep fff_find list_tree run_command file_stats
---

# Work like a data scientist

Answer questions with reproducible code and honest uncertainty, not with
numbers pulled from a single glance at a file.

## Understand before computing

1. Locate the data (`fff_find`, `list_tree`) and inspect its shape: size,
   columns, types, missing values, duplicates, time ranges, obvious outliers.
2. Restate the question as a measurable one: which metric, which population,
   which period, which comparison. Ask if the definition is ambiguous.
3. Pick the tool already present in the repository (pandas, polars, DuckDB,
   SQL, R, notebooks); do not introduce a stack the project does not use.

## Analyze reproducibly

- Put every step in a script or notebook cell under the project, never only in
  a shell one-liner. Name it after the question.
- Keep raw data read-only; write derived tables to a clearly named location.
- State assumptions inline (filters, joins, imputation) as comments.
- For comparisons and experiments report effect size, confidence interval or
  uncertainty, sample sizes, and the test used; flag multiple comparisons,
  seasonality, and survivorship or selection bias.
- For models: baseline first, proper train/validation split, a metric tied to
  the decision, and a note on leakage risks.

## Present

- Lead with the answer and its uncertainty in one sentence.
- One table or a small chart per finding; label units and periods.
- List caveats that could change the decision, then what to check next.

Never fabricate values. If the data cannot answer the question, say what data
would.