Agent Skills repo

A new open-source repository called agent-skills packages software engineering checklists for coding agents, aiming to make them follow stricter workflows instead of jumping straight to code. (github.com) The repository was published by Addy Osmani on GitHub and had about 9,600 stars and nearly 1,000 forks when checked on April 13, 2026. Its README says it is built for “AI coding agents” and organizes work across six stages: define, plan, build, verify, review, and ship. (github.com) In this system, a “skill” is a reusable instruction file, like a recipe card for a narrow task, while the agent is the cook using it. The Agent Skills specification says a skill is usually a folder with a `SKILL.md` file that tells the model what to do and when to use it. (agentskills.io, agentskills.io) The technical problem is context overload: if every rule lives in one giant prompt, the model pays to carry all of it on every turn. The Agent Skills documentation says compatible agents load only a skill’s name and description at startup, then pull in the full instructions only when the task matches. (agentskills.io, agentskills.io) Osmani’s repo bundles 19 skills, including spec-driven development, test-driven development, security-and-hardening, code-review-and-quality, performance-optimization, and shipping-and-launch. The skills directory also lists newer entries such as source-driven-development, which was updated hours before this check. (github.com) The repo also defines seven slash commands that map those skills onto a development lifecycle: `/spec`, `/plan`, `/build`, `/test`, `/review`, `/code-simplify`, and `/ship`. The README says each command activates the relevant skills automatically instead of requiring the user to paste long instructions by hand. (github.com) The setup guides show the project is not limited to one coding assistant. The README lists Claude Code, Cursor, Windsurf, GitHub Copilot, Codex, and “other agents,” and a separate guide shows native installation steps for Gemini Command Line Interface. (github.com, github.com) That portability lines up with the broader standard behind it. The Agent Skills site says the format was originally developed by Anthropic, released as an open standard, and is now supported by multiple agent products, including Claude Code and OpenAI Codex. (agentskills.io, agentskills.io) Osmani’s own course material draws the distinction plainly: tools let an agent act, while skills tell it how and when to act. His lesson on agent skills says the point is to package domain expertise into modular files that can be shared, versioned, and loaded only when relevant. (addyosmani.com) The repo’s pitch is simple: make coding agents behave more like cautious senior engineers and less like autocomplete with shell access. Whether teams adopt this exact package or write their own, the files are plain Markdown, which makes the rules easy to inspect, edit, and reuse. (github.com, github.com)