v0.1.0 // agentskills.io open standard

The developer toolkit
for agent skills. Build, validate, route, and generate SKILL.md files — the portable format for AI agent capabilities.

SKILL.md is a markdown file that teaches an AI agent one capability. This toolkit lets you parse, validate, lint, score, route, and generate them — no LLM required.

npm install skill-tools ↓ Read the Docs →
WHAT

SKILL.md

A markdown file with YAML frontmatter that describes one capability an agent can perform.

WHY

Portable skills

One format, any agent. Share skills across teams, vendors, and frameworks.

HOW

This toolkit

Parse, validate, lint, score, route by intent, and auto-generate from OpenAPI specs.

@skill-tools/core skill-tools CLI @skill-tools/router @skill-tools/gen 114 tests passing zero LLM dependencies 100% TypeScript Apache 2.0 @skill-tools/core skill-tools CLI @skill-tools/router @skill-tools/gen 114 tests passing zero LLM dependencies 100% TypeScript Apache 2.0

Get started
in seconds.

Four packages. One ecosystem. Parse and validate skills with the core library, lint them with the CLI, route by intent with embeddings, or generate from your existing API specs.

~/project 
$ npm i @skill-tools/core       # parser & types
$ npm i -g skill-tools           # validate, lint, score
$ npm i @skill-tools/router   # semantic routing
$ npm i -g @skill-tools/gen   # generate from OpenAPI

The ecosystem.

Four tools that cover the full skill lifecycle.

01 — FOUNDATION

@skill-tools/core

Parse SKILL.md files, extract YAML frontmatter, count tokens, resolve file references. Every other package depends on this.

parsertypestokenizerresolver
@skill-tools/ scope — the shared foundation under the skill-tools npm org
02 — QUALITY

skill-tools

The ESLint for agent skills. Validate against the spec, lint for quality issues, score 0‑100 across five dimensions.

CLIvalidatelintscore
03 — ROUTING

@skill-tools/router

Semantic skill selection via TF-IDF embeddings + cosine similarity. Find the right skill from thousands — zero external dependencies.

embeddingsroutingsearchvector
04 — GENERATION

@skill-tools/gen

OpenAPI 3.x in, SKILL.md out. Unified or per-endpoint modes. Auth detection, curl examples, error handling baked in.

OpenAPIgeneratorCLI

The format.

YAML frontmatter + Markdown body. That's it.

deploy-vercel/SKILL.md 
---
name: deploy-vercel
description: Deploy web apps to Vercel.
  Use when pushing to production.
version: 1.0.0
license: Apache-2.0
---

# Deploy to Vercel

## Prerequisites
- Vercel CLI (`npm i -g vercel`)
- Project linked (`vercel link`)

## Steps
1. Run `vercel --prod`
2. Verify deployment URL in output
3. Report the live URL to the user

## Error Handling
If deployment fails, run `vercel logs`.
Frontmatter

YAML metadata for discovery. Name is required, must match the parent directory, max 64 chars.

Description

Required. Tells agents when to use this skill. Powers semantic routing and catalog discovery.

Instructions

Markdown body the agent follows step-by-step. Keep under 5,000 tokens for optimal context window usage.

Resources

Progressive disclosure. Reference scripts/, references/, and assets/ directories to keep the core skill lean.

Open standard.
Portable skills.

Built on the Agent Skills specification at agentskills.io — a vendor-neutral format for packaging AI agent capabilities.

Read the Spec → Reference Library