Agent Skills That Treat Prompts Like Assets

What it does

MengTo/Skills is a library of 75 folder-based “agent skills” — essentially SKILL.md playbooks that tell AI coding agents like Codex, Claude, and Cursor how to execute specific design and frontend tasks. Each skill packages prompts, constraints, guardrails, and references into versioned files the agent can load and follow, covering everything from turning screen recordings into detailed build prompts to implementing GSAP scroll animations and WebGL effects.

The interesting bit

The project treats prompts as infrastructure rather than chat history: skills are procedural, portable, and scoped with explicit triggers like “Use when…” so the agent knows exactly which playbook to run and what defaults to apply. It is essentially an attempt to systematize taste by replacing vague adjectives with screenshot-backed references and step-by-step constraints.

Key highlights

• 75 skills across Codex workflows, media sourcing, UI prompting, and web design implementation
• Portable folder structure: each skill contains a SKILL.md playbook plus optional references, assets, and scripts
• Philosophy centers on constraints over vibes: explicit defaults, acceptance checks, and negative prompts
• Covers niche frontend territory like stitched full-page capture, cinematic GSAP storytelling, and WebGL steering
• Designed to work across agents (Codex, Claude Code, Cursor) via plain Markdown contracts

Caveats

• The README notes some maintenance ideas are still pending, including install notes for specific agents and lightweight validation for required frontmatter
• Most skills appear to be single-author curation (the repo literally tells contributors to “Write like Meng To”), so community contribution patterns are unclear

Verdict

Grab this if you are a designer or frontend developer trying to get consistent, repeatable output from AI agents without retyping the same constraints every session. Skip it if you are looking for a framework or tool with its own runtime; this is strictly a curated knowledge base.