Overview
fastskill eval runs skill quality checks using prompts and rubrics under [tool.fastskill.eval] in skill-project.toml. Subcommands are validate, run, report, and score.
eval run requires an agent on PATH for the chosen --agent key (for example codex, claude, gemini, opencode, or agent). Use eval validate --agent <key> to verify availability.
Usage
eval validate
Checks eval configuration and files (optional agent presence check).
eval run
Runs eval cases against an agent; writes artifacts under --output-dir.
eval report
Summarizes a completed run directory.
eval score
Re-scores saved artifacts without invoking the agent again.
See also
