Skip to main content
Use kumo-rfm-agent when you want Codex to work directly with KumoRFM. It is designed for workflows where the agent should inspect the schema, repair graph links if needed, translate a natural-language request into a valid query, and run predictions right away.

Download Skill

RFM Quickstart

This is the better fit when the work sounds like “figure out the right query and score it now,” especially when you do not want a training workflow. For example, this skill works well if you want Codex to inspect a Snowflake schema and predict which customers are likely to churn in the next 90 days, or if you want it to turn “rank the top 10 products each active user is most likely to buy next month” into a valid KumoRFM query.

​
Install

Download https://raw.githubusercontent.com/kumo-ai/kumo-skills-catalog/master/kumo/kumo-rfm-agent/SKILL.md, then install it into Codex: 
mkdir -p ~/.codex/skills/kumo-rfm-agent
curl -L https://raw.githubusercontent.com/kumo-ai/kumo-skills-catalog/master/kumo/kumo-rfm-agent/SKILL.md -o ~/.codex/skills/kumo-rfm-agent/SKILL.md

​
Before You Start

Set this environment variable before starting Codex: 
export KUMO_API_KEY="your-kumo-api-key"

​
Example Prompts

Use the kumo-rfm-agent skill on the ecommerce dataset at s3://acme-rfm-datasets/online-shopping/ to predict which users are likely to churn in the next 90 days.

Use the kumo-rfm-agent skill on the shopping dataset at s3://acme-rfm-datasets/online-shopping/ to turn this request into a valid query: rank the top 10 products each active user is most likely to buy in the next 30 days.

Use the kumo-rfm-agent skill on the lead scoring dataset at s3://acme-rfm-datasets/lead-scoring/lead_scoring.csv to confirm the columns are usable and predict which leads are most likely to convert.
The value of this skill is that it gives Codex a safer path through KumoRFM tasks. It tells the agent to inspect the graph before querying, avoid inventing schema details, and return a working query with sample outputs rather than stopping at a rough draft. If you want more detail on RFM itself, start here: