Skills

The MCP tools say what rowlab can do. The skills in row-and-reason/agent-skills say how to do it well: the methodology of the guides, written for an agent. They work in Claude Code, Codex, Cursor and any other client that reads SKILL.md files.

npx skills add row-and-reason/agent-skills --skill rowlab-labeling
npx skills add row-and-reason/agent-skills --skill rowlab-model-selection
npx skills add row-and-reason/agent-skills --skill rowlab-labeling-at-scale

Skill	When the agent uses it
rowlab-labeling	any “label each row as…” request: define the question, try it on a few rows, run on all, check, deliver
rowlab-model-selection	which model to use, whether a cheaper one will do, how accurate a column is
rowlab-labeling-at-scale	large datasets, several questions over the same rows, or a run whose estimate is more than expected

The skills tell the agent to report the estimate before a run on all rows, to read answers against their rows rather than counts, and to name the datasets it made so you can look at them yourself.