Skip to content

evrentan/skill-generator

v1.0.0Apache-2.0

Create, update and benchmark agent skills: interview the user, draft SKILL.md with bundled files, validate the structure, test with and without the skill, grade the results and tune the description so it triggers reliably.

Changelog

All notable changes to this project are documented here. The format follows Keep a Changelog, and the project uses Semantic Versioning.

Unreleased

1.0.0 - 2026-10-31

Added

  • Create mode: interview the user or mine the current conversation, then draft SKILL.md with bundled scripts, references and assets.
  • Update mode: snapshot the original, audit it against best practices, apply minimal targeted edits and show a diff.
  • Test loop that runs realistic prompts with and without the skill and grades the results (agents/grader.md).
  • Blind A/B comparison and win analysis (agents/comparator.md, agents/analyzer.md).
  • Description tuning with train/held-out trigger evals.
  • scripts/quick_validate.py: stdlib-only structural validator for frontmatter, naming, description quality, body size, file references and credential-like files.
  • references/skill-format.md and references/schemas.md.
  • Plugin manifests for Devin, Claude Code, Codex and Cursor, with the skill under skills/skill-generator/.
  • Project logo, used in the README and the Codex and Cursor plugin cards.
  • Contributing guide, security policy, issue and pull request templates, Dependabot and a CI workflow that validates the skill, checks the license copy stays in sync and verifies the manifest versions match on Linux, macOS and Windows.