Addy Osmani, formerly of Google Chrome DevRel and currently at Anthropic, has released agent-skills, an open-source collection of production-grade engineering workflows designed to help AI coding agents follow senior-level development practices. The project features 20 structured workflows, 7 slash commands, and comprehensive checklists covering the full development lifecycle from specification to deployment.
Comprehensive Development Lifecycle Coverage
The agent-skills framework organizes workflows into six distinct phases: Define, Plan, Build, Verify, Review, and Ship. The system includes 7 slash commands that map directly to this lifecycle: /spec, /plan, /build, /test, /review, /code-simplify, and /ship. Each command triggers structured workflows that guide AI agents through specific engineering tasks with clear quality gates and evidence requirements.
The project also includes 3 specialist agent personas and 4 reference checklists designed to maintain consistency across different stages of development. The framework is compatible with major AI coding tools including Claude Code, Cursor, Gemini CLI, Windsurf, GitHub Copilot, and Kiro.
Built-In Quality Gates and Common Failure Prevention
A distinctive feature of agent-skills is its documentation of common excuses AI agents use to skip critical steps. Each skill includes a table listing typical shortcuts agents might take—such as "I'll add tests later" or "Error handling can come in a future iteration"—along with documented counter-arguments. Every skill concludes with specific evidence requirements, such as passing tests, build output, or runtime data, ensuring agents complete work to production standards.
Part of a Larger AI Development Ecosystem
Osmani has positioned agent-skills as a companion to his Dictionary of AI Coding project, together forming a comprehensive knowledge base for developers working with AI coding agents. He has written extensively on related topics including self-improving coding agents, agentic engine optimization (AEO), and multi-agent coding orchestration through his work on Beyond Vibe Coding, a comprehensive guide to AI-assisted development. The project was discussed on Hacker News, where it received feedback from the developer community.
Key Takeaways
- Agent-skills provides 20 core engineering workflows and 7 slash commands covering the complete software development lifecycle from specification to deployment
- Each skill includes tables documenting common AI agent shortcuts (like skipping tests) with counter-arguments to enforce production-grade standards
- The framework requires evidence of completion for every workflow—passing tests, build output, or runtime data—preventing incomplete work
- Compatible with major AI coding tools including Claude Code, Cursor, Gemini CLI, Windsurf, GitHub Copilot, and Kiro
- Released by Addy Osmani as part of a broader ecosystem of AI development resources including the Dictionary of AI Coding and research on self-improving agents