01
Profile unfamiliar data
Inspect schemas, types, units, ranges, duplicates, category levels, date coverage, missingness, and suspicious values before choosing a method.
Evidence-first analysis workflow
CategoryData & Research
Codex Skills for Data Analysis
Codex Skills for Data Analysis give Codex a dependable process for inspecting datasets, cleaning values, testing assumptions, writing analysis code, creating useful charts, and explaining results. Download the skills to make exploratory and repeatable analysis easier across real projects.
Data cleaningEDAStatisticsReproducible reports
01
02
03
04
05
06
07
08
09
02
03
04
05
06
07
08
09
task: explain customer churn
inspect:
- schema and grain
- missing values
- target leakage
verify: tests + charts + findingsResults depend on the decisions before them
A chart can look convincing even when the data grain, missing values, or comparison group is wrong.
Useful analysis begins by understanding what one row represents, where each field came from, how missingness is encoded, and which transformations are justified. Only then should summaries, statistical tests, or predictive models be trusted.
These skills guide Codex through data cleaning, exploratory data analysis, statistical analysis, visualization, and reporting with checks that make the work easier to inspect and repeat.
What the downloadable skills can do
What Codex Skills for Data Analysis can handle
02
Build a documented cleaning pipeline
Standardize fields, resolve types, handle nulls, join tables, reshape records, and preserve a clear account of what changed.
03
Analyze patterns and uncertainty
Calculate suitable summaries, compare groups, investigate relationships, test assumptions, and report uncertainty without overstating causation.
04
Produce reviewable outputs
Create readable plots, tables, notebooks, scripts, validation checks, and a plain-language account of methods, limitations, and findings.
From raw file to supported finding
Treat data preparation, analysis, and interpretation as separate jobs
The workflow records the question and data grain first, then keeps transformations visible so another person can rerun and challenge the result.
workflow.statusREADY
Context → Plan → Work → Verify
01
Frame the decision
Clarify the question, population, unit of analysis, time window, comparison, outcome, and what action the result may inform.
02
Inspect and clean
Create a quality report, define missing-value treatment, correct types, check joins, test totals, and save derived fields in code.
03
Explore and model
Use distributions, grouped summaries, visualizations, statistical tests, or models that fit the question and data-generating process.
04
Validate and communicate
Check sensitivity, leakage, residual issues, edge cases, and reproducibility, then explain what the evidence does and does not support.
Works with familiar analysis stacks
Keep the method consistent while the tools change
The skill can adapt its instructions to the files and libraries already used in your repository.
Python and pandas
Clean, merge, reshape, aggregate, visualize, and validate tabular data in scripts or notebooks.
R and tidy workflows
Prepare data, run statistical analysis, build graphics, and render repeatable research or business reports.
SQL analytics
Check table grain, write readable queries, validate joins, calculate cohorts, and reconcile output with source totals.
Notebooks and reports
Combine code, results, charts, interpretation, assumptions, and limitations in an ordered analysis record.
Common analysis requests
Use the skills for questions that begin with messy evidence
The workflow supports quick investigations and longer analytical projects without skipping the checks that make conclusions trustworthy.
01Profile a CSV, workbook, or database extract
02Clean missing and inconsistent values
03Join datasets without multiplying records
04Create grouped summaries and cohort tables
05Explore distributions and outliers
06Select and evaluate statistical tests
07Build clear analytical visualizations
08Write a reproducible findings report
People who need defensible answers
Who Is This For?
These skills are useful when the analysis must be clear enough for someone else to inspect, rerun, and act on.
Data analysts
Move from an unclear request to a tested dataset, analysis plan, useful output, and concise explanation.
Researchers
Document cleaning, assumptions, statistical choices, sensitivity checks, figures, and limitations.
Product and operations teams
Analyze funnels, cohorts, experiments, service levels, forecasts, and recurring business metrics.
Developers working with data
Create reliable transformation scripts, validation checks, reporting jobs, and analysis-ready tables.
Good to know:
The skills assist with analytical work but do not make a weak dataset representative or turn association into causation. Important financial, medical, policy, and scientific decisions require qualified review and independent validation.
Set up an analysis repository
Installation Guide
Place the skill where it can read the data dictionary, code, validation rules, and expected deliverables.
Download and extract the data analysis skills
Keep the SKILL.md file with any analysis checklists, code conventions, and supporting references included in the folder.
Choose where the skill should apply
Use project scope for one dataset and its business rules, or personal scope for a general analysis workflow.
Document the dataset
Add field meanings, units, grain, source systems, refresh dates, known quality issues, sensitive columns, and access restrictions.
State the approved toolchain
Record the language, libraries, environment, output format, testing method, and whether notebooks or production scripts are expected.
Start with the question, not a chart
Tell Codex the decision, population, period, comparison, desired evidence, and constraints before requesting analysis.
Before you analyze
Frequently Asked Questions
Practical answers about files, statistics, dashboards, machine learning, and sensitive information.
Which data formats can the skills work with?+
They can guide work with formats supported by your environment, including CSV, TSV, JSON, spreadsheets, Parquet, database tables, and API exports.
Can they choose the correct statistical test?+
They can narrow the options by checking the question, variable types, design, assumptions, and sample structure. A qualified analyst should review consequential results.
Do the skills support machine learning?+
Yes. They can structure preprocessing, train-test separation, pipelines, cross-validation, metrics, error analysis, and leakage checks when prediction is the actual goal.
Will they create dashboards?+
They can help prepare metrics, queries, chart specifications, and dashboard code. The workflow first verifies the calculations and intended decisions behind each view.
How should private data be handled?+
Follow your organization's access, retention, de-identification, and approved-tool rules. Do not place protected data in an environment that is not authorized to process it.
Make the path to the answer inspectable
Give Codex a data workflow that checks the inputs, records the transformations, and earns the conclusion.
Known grain. Tested calculations. Honest uncertainty. Results people can review.