AI Agents

aievals skill manages the skill file — a generated reference that teaches an AI coding agent how to drive this CLI. Install it once per project and the agent stops guessing at commands.

Install the skill file

Run this inside your project directory:

aievals skill install

This auto-detects Claude Code or Cursor and writes the skill file so the agent knows how to use the CLI. After this, you can give Claude Code prompts like:

List all datasets in the project and create a new one called "my-test-dataset"

Or a full experiment workflow:

Using the aievals CLI, create a dataset with 3 items, create a code evaluator, create a metrics group, create a prompt with model params, create an experiment linking them all, and run it with --wait

Without the skill file, Claude Code still works — it'll discover the CLI via --help. Installing the skill just makes it faster (skips the --help discovery step).

Target a specific agent

aievals skill install --agent claude
aievals skill install --agent cursor
aievals skill install --agent generic

claude-code is accepted as an alias for claude.

Auto-detection. With --agent omitted, the CLI looks for each agent's binary on your PATH and then for its directory in the current folder. Every agent it finds gets configured — so a project with both .claude/ and .cursor/ gets both files in one run. If it finds nothing, it falls back to generic.

Print the skill definition

aievals skill print

Writes the full skill definition to stdout instead of installing it. Useful for piping into a different agent's config format, or for reviewing what the agent will be told.

The definition is generated from the CLI's live command tree, so it always reflects the binary you have installed — including any commands added since the docs were last updated.