Skip to content
FrankX.AI
Agent skills hub

Claude Code skills and portable SKILL.md workflows for your agent stack.

For engineers and technical creators shipping with coding agents: package your domain rules, verification scripts, and architecture standards into modular SKILL.md folders that load on demand across Claude Code, Codex, Gemini CLI, and Cursor.

Quick install (Claude Code plugin)

Install from the open-source MIT repository

/plugin marketplace add frankxai/claude-skills-library
/plugin install claude-skills-library@claude-skills-library
Example trigger phrases inside a session:
Activate Technical Translator
Run Release Notes Writer
Engage Creation Engine
Consult System Architect
File contract

What goes inside a SKILL.md folder?

Every skill uses progressive disclosure so your context window stays lean. At session start, the harness reads only the YAML name and description. When your prompt matches that description, the agent loads the markdown body and any referenced helper scripts.

  • SKILL.md — YAML routing description plus step-by-step operational rules.
  • scripts/ — Deterministic Python, Bash, or Node validators that run without burning model tokens.
  • references/ — Deep schema tables or brand rules read on demand.
.claude/skills/release-contract-check/SKILL.mdAgent Skills format
---
name: release-contract-check
description: Verify Next.js build, internal links, JSON-LD schema, and WCAG contrast before opening a pull request. Use when shipping routes or MDX guides.
---

# Release contract check

## Steps
1. Run `pnpm run type-check` and `pnpm run content:check`.
2. Inspect canonical URLs and JSON-LD blocks on every changed route.
3. Verify keyboard focus states and heading hierarchy (`h1` -> `h2` -> `h3`).
4. Summarize test output in the pull request body.
Architecture comparison

Skills vs plugins vs slash commands vs subagents vs MCP servers

Choose the lightest primitive that solves the failure mode. Keep static standards in skills, external API connections in MCP servers, and isolated parallel execution in subagents.

PrimitiveHow it worksBest jobContext window impact
Skill (`SKILL.md`)Markdown folder with YAML frontmatter; the description stays in context for routing, the body loads when the task matches or you type /name.Domain standards, repeatable checklists, brand voice, framework rules.Model-invocable skills add a name and description to the skill listing (description plus when_to_use is cut off at 1,536 characters); the body loads only on use.
Plugin (`/plugin`)Installable bundle packaging skills, slash commands, subagents, hooks, and MCP configs from a marketplace repo.Versioned distribution across teams, IDEs (VS Code, JetBrains), and repositories.Depends on bundled components; skills inside plugins stay lazy-loaded.
Slash command (`.claude/commands/`)Single-file prompt invoked with `/command-name`. Commands are now merged into skills: `.claude/commands/ship.md` and `.claude/skills/ship/SKILL.md` both create `/ship`.Deterministic operator workflows such as `/ship`, `/handover`, or `/review`.Same as a skill. With `disable-model-invocation: true` its description leaves the listing and it runs only when you type /name.
Subagent (`.claude/agents/`)Isolated context window with its own system prompt, tool allowlist, and model assignment.Parallel research, independent code review, and bounded multi-file tasks.Separate token budget; returns a compact summary to the parent session.
MCP server (`.mcp.json`)JSON-RPC process or HTTP endpoint exposing live tools, schemas, and external API actions.Database queries, browser inspection, issue trackers, design systems, and cloud APIs.Tool schemas load into context unless lazy-loaded by the harness.

Skill workflow

How to author and verify a skill

Go to vault

Design your skill

Use the Skill Builder to define YAML frontmatter (`name`, `description`), trigger phrases, and output contracts.

Install in your harness

Place the folder at `.claude/skills/<name>/SKILL.md` for one repo or `~/.claude/skills/<name>/SKILL.md` across projects.

Verify with evals

Test routing accuracy and output constraints with promptfoo test cases before sharing the skill with your team.

Implementation guides

Deep-dive playbooks and starter kits

Frequently asked questions

What is a Claude Code skill (`SKILL.md`)?

A Claude Code skill is a directory containing a SKILL.md file with YAML frontmatter plus optional scripts, templates, and reference files. Every frontmatter field is optional; `description` is the one to write well, and `name` defaults to the folder name. Claude keeps the name and description in a short listing and reads the full instructions when the task matches.

Where do Claude Code skills live on disk?

Project-scoped skills live in `.claude/skills/<skill-name>/SKILL.md` inside your repository so every contributor shares them through Git. Personal skills shared across repositories live in `~/.claude/skills/<skill-name>/SKILL.md`.

How do Claude Code skills differ from plugins and MCP servers?

A skill teaches the model how to perform a workflow using markdown instructions and local scripts. An MCP server connects the model to external tools or live data sources over JSON-RPC or HTTP. A plugin packages skills, commands, hooks, subagents, and MCP configurations into a single installable unit.

Do SKILL.md files work in Codex, Gemini CLI, and Cursor?

Yes. The open Agent Skills directory layout (`SKILL.md` with YAML frontmatter and optional `scripts/` and `references/` folders) works across Claude Code, OpenAI Codex CLI, Gemini CLI, OpenCode, Antigravity, and Cursor adapters.

Generate a ready-to-install SKILL.md

Use the interactive Skill Builder to draft frontmatter, trigger phrases, and verification steps for your repository.