skls.to

Skills Loader

Universal skill discovery and composition for AI agents. Browse over 2,000 community skills, measure what works, and give your agents the knowledge they need — without reinventing the wheel.

What is Skills Loader

Skills Loader is a universal skill discovery and recommendation engine for AI agents. It helps agents find the right capabilities for the task at hand — across any tool that supports the Agent Skills standard.

The problem is straightforward: tens of thousands of agent skills exist across GitHub, but there's no way to know which ones are safe, which ones actually work, or which ones fit what you're doing right now. GitHub stars are the only quality signal. That's not enough.

Skills Loader indexes community skills from curated open-source repositories, reviews every one for security, and recommends the right skills for each task. It works with Claude Code, Cursor, Codex, Gemini CLI, and 25+ other tools. It's open and free.

2,052
community skills indexed
1,334
reviewed and approved
5
curated source repos

How It Works

The Composition Model

Skills Loader treats agent configuration as composition. An agent's capabilities are built from three layers: a role (what kind of agent it is), skills (composable modules that shape what it can do), and knowledge (project-specific context like codebases and docs).

Profiles bundle these layers into coherent agent personas. A "developer" profile combines different skills than a "QA" or "orchestrator" profile. There are 11 profiles available, each assembling the right skills for a type of work.

Recommendation Engine

Tell Skills Loader what you're working on — the task, the languages, the frameworks, the tool you're using — and it returns the 3-5 most relevant skills, already assembled into the format your tool needs.

There are no LLM calls in the recommendation path. It's deterministic and fast — typically under 50ms. Scoring is based on task relevance, project context match, and community signal.

Your task context → Recommendation engine → Matched skills → Formatted for your tool

Multi-Format Output

The same skills, formatted for whichever tool you use. One API call, one format flag:

  • Claude Code (claude-code)
  • Cursor (cursor-rules)
  • Codex (codex)
  • Raw Markdown (markdown)
  • Structured data (json)

The Skill Catalog

2,052 community skills from 5 curated open-source repositories on GitHub — all with permissive licenses (CC0, MIT).

Skills arrive in many formats: .cursorrules, CLAUDE.md, AGENTS.md, SKILL.md. Ingestion normalizes them all into a common format, tracks full provenance, and deduplicates across sources so the same skill from multiple repositories appears only once.

New skills are ingested automatically as source repositories update.

Categories Agent Skills docs ↗

  1. Library & API Reference
  2. Product Verification
  3. Data Fetching & Analysis
  4. Business Process & Team Automation
  5. Code Scaffolding & Templates
  6. Code Quality & Review
  7. CI/CD & Deployment
  8. Runbooks
  9. Infrastructure Operations

Skills also carry two meta-types: Capability (what an agent can do) and Preference (how an agent should behave). Both are composable and can be mixed within a single profile.

Security & Trust

Every community skill goes through automated security scanning before it enters the catalog. Skills flagged for risk are escalated to a multi-model review panel where independent AI reviewers assess the content from different perspectives.

The review system uses consensus-based verdicts — no single model decides whether a skill is safe. Skills that fail review are blocked and never appear in the catalog.

Before review, all content is normalized to detect obfuscation and evasion techniques. The full provenance chain is tracked for every skill: its origin, every review decision, every status change. If a skill's source content changes upstream, it gets re-reviewed.

1,334
skills reviewed and approved
Full
provenance chain tracked

What's Next — Effectiveness Tracking

Most skill catalogs have no quality signal beyond GitHub stars. We're building something better.

The infrastructure for tracking skill effectiveness is in place. Agents can report outcomes after using skills, and those signals feed back into the recommendation engine. Skills that consistently help get recommended more often.

This is early work. Measuring whether a skill genuinely helped is hard — it requires deeper integration with development workflows like task completion, test results, and code review outcomes to produce trustworthy signal. We're investing in this because it matters.

As adoption grows and the feedback loop matures, recommendations will get meaningfully better. The vision: skills earn their ranking through real-world results, not popularity contests.

Get Started

Three ways to start using Skills Loader:

MCP Install (Claude Code)

One command. Skills Loader appears as an MCP tool in your agent's context.

Install via Claude Code 
claude mcp add skills-loader --transport http --url https://skills.agentsandswarms.ai/mcp

REST API

Anonymous recommendations — no API key needed for public endpoints.

Get recommendations 
curl -X POST https://skills.agentsandswarms.ai/api/public/recommend \
  -H "Content-Type: application/json" \
  -d '{"workType": "dev", "taskText": "build a REST API with Express and TypeScript"}'

TypeScript Client

Install the client 
npm install @agentsandswarms/skills-loader
Recommend skills for a task 
import { SkillsLoaderClient } from '@agentsandswarms/skills-loader';

const client = new SkillsLoaderClient();
const { skills } = await client.recommend({
  workType: 'dev',
  taskText: 'build a REST API with Express and TypeScript',
  projectContext: { languages: ['typescript'], frameworks: ['express'] }
});

Browse the Catalog

Explore the full skill catalog at agentsandswarms.ai/skills.

Roadmap & Current State

Where we are now

  • 2,052 community skills indexed from 5 curated sources
  • 1,334 reviewed and approved (65% of catalog)
  • Multi-model security review pipeline active
  • 11 agent profiles available
  • MCP integration live for Claude Code
  • REST API and TypeScript client published

What's next

  • skls.to as the primary home for skills discovery
  • Author ownership and verification — claim your skill, request removal
  • User accounts with personal dashboards
  • Voting system for community quality signals
  • Category browsing and search improvements
  • Agent memory integration via npx aismemory — cross-session skill history
  • Expanded source repository coverage

Further out

  • Skill authoring tools
  • Team-level analytics
  • Custom skill repositories — bring your own source