How Skills Work
The Agent Skills Standard
This project uses the Agent Skills open standard (SKILL.md format) — a portable way to provide contextual knowledge to AI coding assistants.
Each skill is a directory containing:
skill-name/
SKILL.md # YAML frontmatter + Markdown instructions
references/ # Documentation fetched from upstream
REFERENCE.md # Index of available documents
setup.adoc # Actual docs from the source repo
...
scripts/ # Optional automation scripts
How the AI Uses Skills
When you ask your AI assistant about a topic covered by a skill, the assistant:
- Detects relevance from the
SKILL.mdfrontmatter (name, description, “When to Use” section) - Reads instructions from the Markdown body
- Consults references in the
references/directory for detailed docs - Follows conventions from the skill’s best practices
Dual-Mode: Skills + Rules
For Cursor IDE, skills can also install .mdc rule files to .cursor/rules/. These rules are:
- Glob-scoped: Only activate when editing files matching specific patterns
- Convention-enforcing: Remind the AI about project-specific patterns
- Optional: Skills work without rules; rules add Cursor-specific enhancements
Skill Lifecycle
graph LR
A[install.sh install] --> B[Clone upstream repo]
B --> C[Copy SKILL.md]
B --> D[Copy docs to references/]
C --> E[Install to IDE skills dir]
D --> E
E --> F[Record in registry.json]
F --> G[check-updates compares commits]
G --> H[update re-fetches if changed]
The Registry
Installed skills are tracked in ~/.rhel-devops-skills/registry.json:
{
"version": "1.0",
"auto_check_updates": true,
"installed_skills": [
{
"name": "agnosticd",
"source_repo": "https://github.com/agnosticd/agnosticd-v2",
"docs_commit_hash": "abc123...",
"installed_to": [
{"ide": "claude", "path": "~/.claude/skills/agnosticd"},
{"ide": "cursor", "path": "~/.cursor/skills-cursor/agnosticd"}
]
}
]
}