better-skills 0.7.7

CLI for authoring, evaluating, and iteratively improving Claude Agent Skills.

better-skills

A measurement-first CLI for building and iterating on Claude Agent Skills. Each iteration runs every test case under two configurations (current + baseline), grades the trajectories, and tracks pass-rate deltas across iterations.

Install

pip install better-skills

better-skills --help
better-skills init <skill-path>                                          # scaffolds triggers.json inside the skill
better-skills init-evals <skill-path>-evals --skill-path <skill-path>    # scaffolds evals.json in a sibling dir
better-skills iterate \
  --skill-path <skill-path> \
  --evals-json <skill-path>-evals/evals.json \
  --workspace <name>-eval
better-skills view

Repo layout

The CLI source lives in this package directory; the matching agent skill (SKILL.md, authoring methodology, schemas, prompt templates) lives at <repo-root>/skills/better-skills/ in the skill-creator monorepo.

When the wheel is built, setup.py copies the runtime resources the CLI reads (grader system prompt, eval viewer template) from the skill folder into scripts/data/ so the published wheel is self-contained.

For the authoring methodology and full skill documentation, read the SKILL.md in the skill folder, or visit the repo above.