How to use the Statistical Reporting agent skill

What does the statistical-reporting skill make the agent do?

The skill is a reference guide written into the agent's instructions. It is not software that runs calculations on your data. Its description says it covers statistical test selection, assumption checking, and APA-formatted reporting.

When the skill is active, the agent is guided to suggest a test based on your data type and group structure, to remind you to check assumptions such as normality and equal variance, and to format results in APA style with effect sizes and confidence intervals.

• Test selection: maps situations such as two independent groups with non-normal data to a suitable test, for example Mann-Whitney U.
• Assumption checks: points to Shapiro-Wilk tests or Q-Q plots for normality, Levene's test for equal variance, and VIF for multicollinearity in multiple regression.
• APA formatting: shows report patterns for t-tests, ANOVA, correlations, chi-square tests and regression.
• Reporting habits: reminds the agent to report effect sizes alongside p-values, use confidence intervals, and state whether tests are one- or two-tailed.

How do you install the statistical-reporting skill?

You can install the skill in two ways. Download the ZIP from this page and either unzip it into your user skills folder or into a project's skills folder, or upload it through the Claude apps.

The skill's files are listed as SKILL.md. The folder should contain that file so the agent can read it.

• Personal install: unzip the download into ~/.claude/skills/.
• Project install: unzip the download into the .claude/skills/ folder of your project.
• Claude apps: go to Settings → Capabilities → Skills and upload the skill.

How do you trigger the skill with natural requests?

You do not need a special command. The skill's description tells the agent to use it when analyzing experimental results or writing results sections, and its metadata lists trigger keywords such as statistic, hypothesis test, p-value, regression, ANOVA, t-test, effect size and confidence interval.

Describe your data and goal in plain language. The more detail about group sizes, paired or independent measures, and whether data look normal, the more useful the suggested test will be.

• "I have two independent groups of patients and their scores look skewed. Which test should I use?"
• "Help me write the results section for an ANOVA comparing three teaching methods."
• "Format this t-test result in APA style and include an effect size."
• "Check whether my data meet the assumptions before I run a t-test."

What output should you expect?

Expect advice and wording, not computed numbers. The agent will tell you which test fits and what to check, then show the APA pattern to use. You still need to run the analysis in your own software and enter the real values.

For example, a t-test is written as t(df) = X.XX, p = .XXX, d = X.XX, and a correlation includes a 95% confidence interval. The skill asks that exact p-values be reported unless p is below .001.

What does the skill need, and is it safe?

The listed requirements are nothing. The skill does not run scripts and does not use the network. It contains no code; it consists of a single instruction file, SKILL.md.

Because it only adds guidance to the agent, it cannot access your files or the internet through its own actions. Any data you share in a conversation is handled by the agent app you are using, so review that app's privacy terms before sharing sensitive data.

• Requires: nothing listed.
• Runs scripts: no.
• Uses network: no.
• Licence: MIT.

What are the limits of the guidance?

The guidance is a short, generic checklist. It may oversimplify complex designs. Its test list does not cover mixed models or survival analysis, so do not rely on it for those.

For high-stakes studies or complex designs, have a statistician review the analysis and the write-up.

• Do not use it as a replacement for a statistician on complex or high-stakes designs.
• Expect it to help with common tests, not specialized models.
• Use it alongside your own software output, not instead of running the analysis.
• Check the agent's suggested test against your study design before committing.

What tips make the results better?

Give the agent your design before your numbers. Tell it whether groups are independent or paired, how many groups there are, and what the outcome type is. Ask it to flag any assumption you have not checked.

The skill also warns against common errors, so ask the agent to review your draft against them.

• Never write that non-significant results mean there is no effect.
• Avoid causal wording for observational data.
• Apply multiple comparison corrections such as Bonferroni or FDR when running many tests.
• State whether tests are one-tailed or two-tailed and justify the choice.