How to use the cohort-analysis agent skill for retention

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when a request matches the skill's description. This skill's description is: perform cohort analysis on user engagement data, covering retention curves, feature adoption trends and segment-level insights.

Once triggered, the skill tells the agent to work in five steps: read and validate your data, generate quantitative analysis, create visualizations, identify insights and patterns, and suggest follow-up research.

• Checks cohort identifiers, time periods and engagement metrics, and looks for missing values and data quality issues
• Calculates cohort retention rates, feature adoption rates and period-over-period changes
• Generates retention heatmaps (cohorts against time periods) and line charts of cohort progression
• Flags early churn, late-stage engagement changes, feature adoption clusters and seasonal trends
• Suggests user interviews, feature surveys, session replays, win/loss analysis and A/B tests

How do you install the cohort-analysis skill?

Download the ZIP for this skill from this page and unzip it. The skill is a single file, SKILL.md, so the unzipped folder should contain that file. You can install it in one of three places:

• Personal skills folder: unzip into ~/.claude/skills/ so the skill is available across your projects
• Project skills folder: unzip into the .claude/skills/ folder of a specific project so it is shared with that project
• Claude apps: upload the skill under Settings → Capabilities → Skills

What do you need before you start?

The skill expects a user cohort dataset. Your file can be CSV, Excel, JSON, or the output of a SQL query. If you want the agent to produce analysis scripts, you also need Python with pandas and numpy installed on the machine where the agent runs code.

The data should include a cohort identifier, a time dimension and at least one engagement or retention metric. The skill's own tips recommend at least three to four cohorts for meaningful patterns, and recommend stating how cohorts are defined (for example signup month or feature launch date).

• A cohort dataset in CSV, Excel, JSON or SQL query output
• Python with pandas and numpy, only if you request analysis scripts
• Context about product changes, launches or events during the period

How do you trigger the skill with a natural request?

You do not need a special command. Describe your data and what you want to know, and mention retention, cohorts, feature adoption or churn so the agent matches the skill. Name the file you are uploading and the columns it contains.

The skill's examples show three request styles: a CSV with a question about why one quarter's cohorts underperform another, a plain-language description of monthly cohorts with a question about long-term retention, and an Excel file of feature adoption data asking which cohorts adopted fastest.

• Name the file and its columns, for example cohort_month, weeks_active, user_id, feature_x_usage, engagement_score
• Ask a specific question, such as which cohorts retain best long term
• Ask for the output you want: charts, a Python script or research recommendations

What output should you expect?

The skill's output format lists the parts you should receive. Not every request produces every part; the analysis scripts appear only if you ask for them.

• Data Summary: cohort overview and data quality assessment
• Quantitative Findings: key metrics, retention rates and trend analysis
• Visualizations: charts showing retention curves and adoption patterns
• Pattern Identification: 2-3 significant insights
• Research Recommendations: qualitative and quantitative follow-ups
• Analysis Scripts (if requested): Python code for reproducible analysis
• Next Steps: prioritized actions based on findings

What are the safety notes?

This skill runs scripts, so review any code the agent writes before running it on your machine. It does not use the network, according to its listed requirements.

Your data is the bigger concern. Uploading raw user records to an AI tool may breach privacy rules. Anonymize or pseudonymize user identifiers and remove personal details before you upload.

• Review generated scripts before running them
• Anonymize user data before uploading it to an AI tool
• Treat outputs as directional, not as statistically tested conclusions

What tips give the best results?

Clear inputs produce clearer analysis. The skill's own guidance is to make the cohort definition explicit, provide data across several time periods, and include context about product changes during the period. Include more than one metric, such as retention, engagement, feature usage or revenue, because the skill can compare them.

• Include the time dimension: data across multiple periods
• Define cohorts explicitly: signup month, feature launch date and so on
• Explain launches and events that happened during the period
• Include several metrics, not just one
• Use at least 3-4 cohorts
• Ask for the specific outputs you need