Skip to content
contextlint

Agent Skills

contextlint publishes three Skills following the agentskills.io spec. A Skill is a unit that tells AI what you want done; internally it captures a workflow combining command execution, file traversal, and reasoning. Once invoked from an AI host, the Skill produces the expected outcome without you walking the AI through each step.

Available Skills

Section titled “Available Skills”
Skill Purpose Details
contextlint-init Initial setup for a repository — scans the doc layout, infers rules, and installs the CLI contextlint-init
contextlint-fix Resolves detected violations through a mix of mechanical fixes and human confirmation contextlint-fix
contextlint-impact Analyzes the impact of a file change using the Context Graph contextlint-impact

Installation

Section titled “Installation”

Requires GitHub CLI v2.90 or newer. If you don’t have it yet, see the GitHub CLI website.

Terminal window
gh skill install nozomi-koborinai/contextlint contextlint-init
gh skill install nozomi-koborinai/contextlint contextlint-fix
gh skill install nozomi-koborinai/contextlint contextlint-impact

contextlint-init alone is enough to start. After init finishes the repository setup, add fix or impact as needed.

Syntax

Section titled “Syntax”
gh skill install OWNER/REPO SKILL
gh skill install OWNER/REPO SKILL@VERSION
gh skill install OWNER/REPO SKILL --agent claude-code --scope user
Option Description
OWNER/REPO The GitHub repository where the Skill is published
SKILL The <name> part of skills/<name>/SKILL.md in the repo
@VERSION Version tag (optional)
--agent Target host (claude-code etc., varies by host)
--scope user (user-wide) or repo (repo-only)

gh skill update, gh skill search, and gh skill publish are also available.

Supported hosts

Section titled “Supported hosts”

Any host that follows the agentskills.io spec runs the same Skills.

  • Claude Code
  • Cursor Agent
  • Cline
  • Codex
  • Gemini CLI
  • GitHub Copilot
  • Other hosts compatible with the agentskills.io spec

For per-host registration steps and supported versions, refer to agentskills.io and the host’s own documentation.

How to invoke a Skill

Section titled “How to invoke a Skill”

After installation, just tell your AI host what you want done — the Skill is invoked automatically. You don’t need to mention the Skill name.

For example, any of the following requests will trigger contextlint-init:

Set up contextlint

I want to add a Markdown linter

Add document integrity checking

contextlint-fix is triggered by requests like “fix lint”, “fix the broken links”, or “contextlint is showing errors”, and contextlint-impact by requests like “what breaks if I change design.md?” or “is it safe to delete this doc?”. See each Skill’s own page for the detailed trigger conditions and behavior.

Differences from the MCP server

Section titled “Differences from the MCP server”
Aspect Skills MCP server
Abstraction High (workflow level) Low (tool level)
How AI is told Natural-language request Tool call
Distribution skills/<name>/SKILL.md in a GitHub repo npm package
Project-specific config Not required (the Skill infers it) contextlint.config.json is required

You can use them together. The contextlint-fix Skill, for example, prefers the MCP lint tool when available — Skills and MCP complement each other.

Why we recommend going through Skills

Section titled “Why we recommend going through Skills”

contextlint has 21 rules across seven categories. Reading all of them by hand to pick the right configuration is heavy lifting, so the contextlint-init Skill inspects the repository and proposes a configuration instead. You only need to confirm at the end, which dramatically reduces the effort needed for the initial setup.