Addy Osmani's Agent Skills for Coding Agents

Addy Osmani has published the Agent Skills repository to give AI agents the operational skills of senior developers. The project forces models to follow structured workflows instead of taking the fastest shortcut to generate code.

What was announced

Addy Osmani released the open-source Agent Skills repository on GitHub, which has surpassed 115,000 stars. The project includes 24 structured skills and eight specific commands to guide coding agents. The commands map the entire software development lifecycle, from idea definition to production release. The system is compatible with over 70 different agents, including:

• Claude Code
• Cursor
• Copilot
• Windsurf
• Cline

The repository is distributed under the MIT license. The official announcement on the author's blog is dated May 3, 2026.

How it works

AI agents tend to write the requested code while ignoring specs, tests, and reviews to get the result as fast as possible. Agent Skills steps in by injecting text files into the agent's context that describe precise workflows, not just tips or documentation. Each development phase has a dedicated command:

• `/spec` to define requirements
• `/plan` to plan the work
• `/build` to implement the code
• `/test` to verify functionality

The system uses anti-excuse tables to stop the agent from skipping checks, providing pre-written responses to every attempt to avoid work. Every skill ends with mandatory exit criteria based on concrete evidence. It also includes a skill for doubt-driven development, forcing the agent to reconsider decisions when the code is unfamiliar. The release command triggers several virtual roles in parallel, like a code reviewer and a security expert, to approve the final work.

What changes for businesses

Adopting these structured workflows reduces bug-fixing time in AI-assisted software development. For example, a company building an internal app can use the review command to have a virtual panel of experts automatically check the code before release. A software house can integrate UI engineering skills to ensure the agent meets accessibility standards when building user screens. A development team can use the test-driven development skill to force the agent to write tests before code, preventing production failures.

Limitations and what to watch out for

The repository focuses exclusively on development methodology and does not include proprietary AI models to run locally. The skills are markdown text files and do not execute direct actions on company servers without the host agent. It has not been communicated whether or how companies need to adjust their internal cybersecurity protocols for these automated workflows. The system's effectiveness depends on the underlying coding agent's quality and its ability to correctly interpret the commands. There is no information on official or commercial support for companies integrating these tools into critical production environments.

How to try it

The repository is publicly available on GitHub, and the official website provides full documentation. Installation happens via the `npx skills add addyosmani/agent-skills` command in the terminal. You must use one of the over 70 compatible agents listed in the documentation to make the skills work.