bkarlovitz/agent-skills

This repo is a personal collection of agent skills that follow an open standard called agent-skills. A skill in this world is just a folder with a SKILL.md file (a short markdown spec with some metadata at the top, like the skill's name and description) plus any extra reference documents that the skill wants to load when it runs. Because the format is open and plain text, the same skill folder works in several different AI coding assistants, including Claude Code, Cursor, opencode, and Codex, without being tied to any one of them. The README's main message is how to install these skills into your agent. The recommended path is a small command line tool called skills (from vercel-labs/skills) that you run through npx, so you do not have to install anything globally. The tool detects which supported agents you have on your machine and copies the skill into the right place for each of them. The README gives example commands for installing one named skill, installing every skill in the repo at once, installing to your user directory instead of the current project, and limiting the install to specific agents. There is also a manual install path: each skill folder under skills/ is self-contained, so you can copy it by hand into your agent's skills directory, for example into a project's .claude/skills folder. At the moment the repo ships one skill, called spec-an-agent. It is described as a grilling session: an interactive prompt that interrogates the design of a planned agentic workflow before any code is written, covering its trigger, scope, authority, escalation, failure modes, and evaluations, and writing the result into spec files like WORKFLOW.md, CAPABILITIES.md, EVALS.md, and architecture decision records. The repo is MIT licensed and has zero stars.