coleam00/skills

What it solves

This project provides a library of 35 modular "skills"—structured markdown files that act as procedural instructions for AI coding agents. It solves the problem of bloated system prompts (like a 2,000-line CLAUDE.md file) by allowing agents to load only the specific instructions needed for the current task, reducing context window usage and improving reliability.

How it works

Each skill consists of a folder containing a SKILL.md file that defines a name, a description of when to use it, and a step-by-step procedure. The agent loads the descriptions at startup and only pulls in the full procedure when the work matches the description. These skills are organized around a core development loop: prime $\rightarrow$ plan $\rightarrow$ implement $\rightarrow$ validate $\rightarrow$ review $\rightarrow$ commit $\rightarrow$ PR.

Who it’s for

Developers using AI coding agents (such as Claude Code, Cursor, Windsurf, or Copilot) who want to standardize and scale their agentic workflows without relying on massive, monolithic prompt files.

Highlights

• Modular AI Layer: A collection of 35 skills covering everything from PRD creation and architecture planning to atomic commits and PR reviews.
• PIV Loop: A specialized set of skills for the "Plan, Implement, Validate" cycle to ensure high-quality, validated code.
• Agent Agnostic: While optimized for Claude Code, the skills are plain markdown and can be installed into 75+ different agents via a CLI tool.
• Deterministic Hooks: Includes Python-based hooks that guarantee specific behaviors (e.g., blocking secret leaks or preventing PRs when tests are red) regardless of the model's memory.
• Meta-Skills: Tools to help users create new skills, audit their AI layer, and scan for opportunities to encode new processes.