Inspired by mattpocock/skills.
Source: mattpocock/skillsReverse-engineered from real GitHub workflow.
Skills for Real Engineers. Straight from my .claude directory.
Build composable, production-ready AI agent skills with installer and configuration system for multi-agent environments.
I want you to create a collection of reusable AI agent skills - think of them as prompt templates and workflows that engineers can install and customize. These are practical, production-ready skills designed to improve how AI coding agents like Claude, Codex, and similar tools work on real projects.
The project should have a shell-based installer that lets users pick and choose which skills to install for their specific agent. Key architectural goals: skills should be small, composable, and easy to hack on rather than monolithic black boxes. They solve real pain points like miscommunication between developers and agents, lack of test coverage, inconsistent code quality, and integration with existing tools.
Each skill should live in its own directory under a skills folder, organized by category like "engineering" and "productivity". Include documentation for each skill explaining what problem it solves. The main skills I'm envisioning are things like /grill-me (asks detailed questions before coding), /grill-with-docs (grill-me plus documentation generation), and /triage (integration with issue trackers).
Create a setup script that walks users through initial configuration - asking them which issue tracker they use (GitHub, Linear, or local files), what labels they use for triage, and where to save generated docs. Build this as a shell-based system that can work across different coding agents and models.
Include a newsletter signup mechanism and clear documentation showing why each skill exists and what common failure modes it prevents. The whole thing should feel like opinionated but hackable tooling that gives developers back control instead of taking it away. Make the installer npm-based so it's easy to discover and install, with clear READMEs and CONTEXT files explaining the philosophy behind the approach.