Skip to content

Releases

Audit installed skills, update from the registry and 11 new skills

Choose a tag to compare

@Lethe044 Lethe044 released this 13 Nov 21:42

Added

  • askit audit: scan the skills already installed for Claude Code and Hermes, or any folder with --path
  • askit update: refresh skills that were installed from a registry when the registry has newer files
  • askit eval --judge: grade open ended eval cases with the model, next to the plain string checks
  • Look-alike check in askit lint: warns about very similar descriptions (W020) and duplicate names (W021)
  • Scanner rules for chat template markers (SC008), links that can carry data out (SC009) and long encoded blobs (SC112); SC101 now also catches downloads piped into python, node, perl, ruby and php
  • Export targets for Aider, Cline and Roo Code
  • 11 new starter skills, 27 in total: docker-compose-reviewer, requirements-auditor, markdown-link-checker, openapi-summarizer, cron-explainer, jwt-inspector, git-conflict-resolver, unit-test-writer, release-checklist, incident-postmortem-writer and regex-explainer
  • Animated terminal demo in the README

Changed

  • SC003 no longer fires on prohibitions such as "do not force push without telling the user"
  • Lint ignores links inside code spans and fenced code blocks
  • File lists in the registry index are sorted by path on every operating system

Install

pip install -U agentskill-toolkit

Lint, scan, test and export agent skills for any coding agent

Choose a tag to compare

@Lethe044 Lethe044 released this 26 Oct 11:50

First release of agentskill-kit. The project began as a Hermes Agent hackathon prototype and is now a model independent toolkit for agent skills (SKILL.md).

Added

  • askit lint: structure checks for SKILL.md, offline and without a model
  • askit scan: detects prompt injection, hidden unicode, secrets, dangerous commands and exfiltration patterns
  • askit test and askit score: run bundled tests in an isolated folder and combine lint, scan and test results into one quality score
  • askit forge: generate a skill from a task or from notes, verify it and refine it until it passes
  • askit eval: compare how a model follows a skill with and without it
  • askit mine: find repeated workflows in shell history and draft skills from them
  • askit export: Claude Code, Hermes, Cursor, Windsurf, GitHub Copilot, Codex (AGENTS.md) and Gemini CLI
  • Static skill registry with checksum verified installs, askit search, install, publish and a searchable website
  • askit serve: local web interface for your skills
  • Providers: Ollama, LM Studio, Groq, Gemini, OpenRouter, any OpenAI compatible endpoint, optional OpenAI and Anthropic keys
  • Turkish and English CLI (--lang tr)
  • 16 starter skills, a GitHub Action, a pre-commit hook and the Hermes integration

Changed

  • Removed the simulated publish step. Quality scores now come from real checks
  • Moved the Hermes training environment to integrations/hermes/

Install

pip install agentskill-toolkit

The PyPI package is named agentskill-toolkit. It installs the askit command.