| name | csv-data-analyst |
| description | This skill analyzes CSV files for data quality, descriptive statistics, trends, business performance, marketing funnels, binary classification, large-scale full scans, and trend forecasting. Use it when a user uploads or names a CSV and asks for analysis, reports, insights, modeling, or large-data processing. |
CSV Data Analyst
Release
- Public version: v1.0.0
- Release ID: OSS-1.0.0
Purpose
Analyze CSV data through a deterministic Python workflow and produce structured, auditable outputs. Support general analysis, business analysis, marketing analysis, binary classification, large-data analysis, and trend forecasting.
Users may describe their goals in ordinary language. Select the appropriate workflow from the task meaning and the confirmed field information; do not require users to name a technical mode.
Core workflow
- Inspect the CSV structure and the user’s objective.
- Identify the required mode and any essential fields.
- Ask only for missing, decision-critical information. For modeling, confirm the target field, label meaning, and feature availability before creating final outputs. For forecasting, confirm the date field, numeric field, horizon, and holdout when they are not clear.
- Run
scripts/analyze.py once with the user-approved output directory and the mode parameters justified by the request.
- Run
scripts/verify_output_manifest.py against the same output directory.
- Summarize facts, limitations, and registered output files in chat. Keep follow-up explanations in chat instead of writing extra files to the output directory.
Mode selection
| User goal | Execution path | Expected outcome |
|---|
| Data quality, statistics, distributions, correlations, equipment inspection, or ordinary trends | General analysis | Data profile, quality report, summary, charts, and runtime metadata |
| Orders, revenue, cost, profit, customers, products, regions, channels, or management review | --business --intent | KPI summary, trend, dimension breakdown, contribution analysis, and business report |
| Campaigns, traffic, funnel stages, ads, experiment groups, or conversion review | --business --intent | Business analysis plus the marketing submode products |
| Binary prediction, classification, scoring, or churn analysis with a confirmed target field | --modeling --target-field | Leakage review, train/test evaluation, feature importance, and OOF evidence when applicable |
| Large CSV, full scan, chunked processing, constrained memory, or no sampling | --bigdata --chunk-size | Resource preflight, chunked full scan, controlled aggregates, and run metadata |
| Date-based trend prediction with a confirmed numeric measure | --bigdata --forecast | Full scan plus fixed OLS extrapolation, time-order validation, rolling validation, and risk signals |
Do not activate modeling or large-data processing from column names or file size alone. Use the user’s stated objective and required confirmations.
Commands
Run commands from the Skill directory. Always use an explicit output directory.
python scripts/analyze.py /path/to/input.csv --output-dir /path/to/output
python scripts/analyze.py /path/to/input.csv --output-dir /path/to/output --business --intent "月度经营复盘"
python scripts/analyze.py /path/to/input.csv --output-dir /path/to/output --modeling --target-field "流失标签"
python scripts/analyze.py /path/to/input.csv --output-dir /path/to/output --bigdata --chunk-size 50000
python scripts/analyze.py /path/to/input.csv --output-dir /path/to/output --bigdata --forecast --date-field "日期" --forecast-field "销售额" --forecast-horizon 3 --forecast-holdout-periods 6 --chunk-size 50000
After each completed run, validate the output package:
python scripts/verify_output_manifest.py --output-dir /path/to/output --expected-mode general
Set --expected-mode to one of general, business, marketing, modeling, bigdata, or forecast.
Analysis boundaries
- Treat correlation and descriptive patterns as observations, not causal proof.
- Treat forecasting as statistical extrapolation, not a business commitment or automated decision.
- Exclude IDs and confirmed post-outcome fields from modeling features.
- Do not generate production action lists, automated retention actions, ad-spend decisions, or dispatch instructions.
- Do not add reports, CSV files, JSON files, logs, or temporary artifacts to an analysis output directory after validation.
Package contents
scripts/analyze.py: deterministic analysis engine.scripts/verify_output_manifest.py: output-directory and metadata validator.resources/sample.csv: small business-analysis sample.examples/equipment_inspection_sample.csv: general-analysis sample.examples/showcase_financial_pl_data.csv: business-analysis sample.README.md: installation, usage, mode overview, and output guidance.GITHUB_PUBLISHING_GUIDE.md: GitHub publishing instructions.
