Dev

Introducing agent-skills: Teaching Development Processes to AI Agents

Agent-skills, released by Addy Osmani, injects senior engineering workflows and quality gates into AI coding agents. Available for over 70 agents.

3 min read Reviewed & edited by the SINGULISM Editorial Team

Introducing agent-skills: Teaching Development Processes to AI Agents
Photo by Mohamed Nohassi on Unsplash

Addy Osmani, a renowned engineer from the Google Chrome team, has released “agent-skills,” a collection of skills for AI coding agents, on GitHub. This project encodes the workflows, quality gates, and best practices used by senior engineers in software development, enabling AI agents to consistently adhere to them across all phases of development.

Injecting Engineering Expertise into Agents

As AI agents for code generation become more prevalent, concerns about the quality and consistency of generated code have emerged. Agent-skills addresses these issues by packaging the tacit knowledge of experienced engineers into explicit rule sets that the agents can automatically reference. Through this project, Osmani aims to empower AI agents not just to generate code, but to follow processes that create verifiable and maintainable software.

Six Phases and Eight Commands for Development

Lifecycle

At its core, agent-skills revolves around a six-phase development lifecycle: Define, Plan, Build, Verify, Review, and Ship. Corresponding to these phases, eight slash commands are provided, allowing developers to interactively control the agent’s actions.

Each command is grounded in specific principles. For instance, /spec creates specifications before coding, /plan breaks work into small atomic tasks, /test emphasizes test-driven development, and /code-simplify applies the principle that “clarity beats cleverness.”

The /build command automates task execution based on a pre-approved plan. Each task is committed individually with test-driven validation, and the process halts if any step encounters issues or risks. This design minimizes human intervention while maintaining rigorous validation.

Integration with a Wide Range of Agents

The installation process uses the Open Skills CLI. With a single command, npx skills add addyosmani/agent-skills, users can install 24 skills at once. Individual skills can also be selected, such as adding only the five axes of code review skills with npx skills add addyosmani/agent-skills --skill code-review-and-quality.

Agent-skills is compatible with over 70 agents, including Claude Code, Cursor, Codex, Copilot, and Cline. For native integration, Claude Code users are advised to install it via the marketplace, while Cursor users can copy the SKILL.md file into the .cursor/rules/ directory for use.

Automatic Skill Activation

One notable feature of agent-skills is its ability to recognize the current task and automatically activate the appropriate skill. For example, api-and-interface-design is triggered during API design tasks, while frontend-ui-engineering is activated for UI development. This ensures that best practices are applied contextually, without developers needing to explicitly specify commands.

Editorial Opinion

In the short term, agent-skills is expected to bring noticeable improvements to the output quality of AI coding agents. By enforcing quality gates and test-driven development, the project could help streamline the transition from prototyping to production-grade software. Additionally, by allowing development teams to explicitly define and share expected agent behaviors, it contributes to the standardization of team knowledge. However, the initial adoption phase may require time for customizing skills and aligning them with existing workflows.

From a long-term perspective, this “codification of engineering skills” has the potential to revolutionize software engineering education and talent development. If the tacit knowledge of senior engineers can be accumulated in a reusable format, even less-experienced developers may leverage advanced practices. That said, challenges such as maintaining skill sets and addressing operational differences between agents remain. The community’s efforts to expand skills and establish compatibility will likely be key to broader adoption. The editorial team views agent-skills as a promising step toward enhancing the reliability of AI agents.

References

Frequently Asked Questions

What agents can agent-skills be used with?
It can be used with over 70 AI coding agents, including Claude Code, Cursor, Codex, Copilot, and Cline. Installation is simple with the Open Skills CLI.
How are the skills applied to agents?
Skills can be executed via slash commands corresponding to development phases or automatically triggered based on the task at hand. For example, the `api-and-interface-design` skill is activated during API design tasks.
Is it possible to install individual skills only?
Yes, you can install specific skills using `npx skills add addyosmani/agent-skills --skill skill-name`. Key skills like code-review-and-quality and test-driven-development are available for individual installation. ## References - [agent-skills GitHub Repository](https://github.com/addyosmani/agent-skills) — Published on July 9, 2026 - Related: [AI Agent Collective "Agency Agents" Launched: One-Click Integration](https://singulism.com/ja/agency-agents-ai-workflow-desktop-app)
Source: GitHub Trending

Comments

← Back to Home