coleam00/skills

What it solves

It addresses the problem of "prompt bloat" in AI coding agents, where large system prompts (like a 2,000-line CLAUDE.md) become inefficient. Instead of loading all instructions at once, this project provides a library of modular "skills" that an agent loads only when the specific task requires them, allowing for better scalability and precision in agentic workflows.

How it works

Each skill is a standalone folder containing a SKILL.md file. This file defines the skill's name, when it should be used, and the specific procedure the agent must follow. Agents load the descriptions at startup to know which skills exist, but only pull in the full procedural content when the work matches the skill's description. These skills are designed to be plain Markdown, making them readable and editable by humans.

Who it’s for

Software developers using AI coding agents (such as Claude Code, Cursor, Copilot, Cline, or Windsurf) who want to standardize and automate their development loop—from PRD creation and architecture planning to implementation, validation, and PR submission.

Highlights

• PIV Loop: A core workflow consisting of Plan, Implement, and Validate.
• Modular Library: 33 pre-defined skills covering codebase priming, epic slicing, issue investigation, and parallel work via git worktrees.
• Meta-Skills: Tools to help users build their own "AI Layer," including rules drift checking and opportunity scanning.
• Broad Compatibility: Supports Claude Code as a plugin or can be installed into 75+ other agents via a CLI tool.
• Dark Factory: A high-autonomy skill that can build a repository from a PRD with minimal human intervention.